개회사 안녕하십니까? 국립국어원장소강춘입니다. 귀한시간을내어참석해주신모든분들께감사의말씀을올림니다. 풍요로운계절에이자리에참석하신분들을뵙게되어반갑습니다. 국립국어원은 2010년부터 2년에한번씩세계의언어정책을국내에소개하고관련기관들과교류와협력관계를다지는국제학술대회를마련해왔습니

Transcription

1 발표자료집 2018 국립국어원국제학술대회 보이는 언어의기록, 수어사전 Records of Visible Language: Sign Language Dictionary ( 목 ) 이화여자대학교이화삼성교육문화관 1 층강당 주최. 주관

2 개회사 안녕하십니까? 국립국어원장소강춘입니다. 귀한시간을내어참석해주신모든분들께감사의말씀을올림니다. 풍요로운계절에이자리에참석하신분들을뵙게되어반갑습니다. 국립국어원은 2010년부터 2년에한번씩세계의언어정책을국내에소개하고관련기관들과교류와협력관계를다지는국제학술대회를마련해왔습니다. 지난 2016년의 수어언어와사회적의사소통 주제에이어, 올해에는 보이는언어의기록, 수어사전 이라는주제로국제학술대회를개최하게되었습니다. 국립국어원장소강춘 이러한뜻깊은자리에소중한경험을나누어주시기위해멀리서오신예테크리스토페르센선생님, 요한나메시선생님, 트레버존스턴선생님, 토마스항케선생님께감사와환영의말씀을올립니다. 애초기조연설을해주시기로하셨던글래디스탕교수님은안타깝게도갑작스러운개인사정으로참석하지못하셨습니다. 2016년 한국수어언어법 의제정과시행으로농인의고유한언어인한국수어는우리사회에서언어로서의자격을법적으로보장받게되었으며, 27만농인과한국수어사용자의언어권과삶의질을실질적으로보장하기위한발판이마련되었습니다. 국립국어원은기존에구축된수어사전을정비하여더욱체계적인수어사전구축을계획하고있는데이번학술대회를계기로한걸음더나아간한국어수어사전의모습을전망하고농사회의발전과공감을이끌어내려고합니다. 이번학술대회에서많은논의가이루어짐으로써우리사회가소통의장벽없는사회로한발더다가설수있는계기가마련될것이라고생각합니다. 오늘이자리에서말뭉치중심의수어사전, 수어사전의구조와표제어표기, 정보기술을활용한수어사전의미래등에대하여논의하면서앞으로의한국수어사전의발전방향을모색하는귀중한자리가되기를바랍니다. 오늘이자리를빛내주신귀빈여러분, 바쁜일정에도우리학술대회를찾아주신여러청중들, 그리고발표를맡아주신세계각국의수어언어전문가여러분께다시한번깊은감사의말씀을올립니다. 모쪼록이번국제학술대회를통해수어사전의발전을이끌어내는풍성한수확을거두기를기원합니다 국립국어원국제학술대회 보이는언어의기록, 수어사전

3 Opening Remarks Greetings. I am So Kang-Chun, Director General of the National Institute of Korean Language (NIKL). I would like to start by expressing my gratitude to all participants for taking the time to attend this conference. It is a pleasure to see everyone here in this season of harvest. The National Institute of Korean Language has hosted the international academic conference biannually since 2010 to introduce language policies from around the world and to strengthen ties for cooperation and exchange with other organizations. Following the theme of Sign Language and Social Communication in 2016, this year s conference is being held under the theme of Records of Visible Language: Sign Language Dictionary. Director General of NIKL So Kang-Chun I want to extend my heartfelt gratitude and welcome to our presenters Dr. Jette Kristoffersen, Dr. Johanna Mesch, Dr. Trevor Johnston, and Dr. Thomas Hanke, who travelled far to share their valuable insights with us. Dr. Gladys Tang was also sched uled to join us to deliver the keynote address, but unfortunately, she could not be here with us due to personal circumstances. With the enactment of the Korean Sign Language Act in 2016, Korean sign language gained legal recognition as a language and laid the foundation for 270,000 deafs and sign language users to ensure their right to language and quality of life. The National Institute of Korean Language is currently working on a more systematic revision of the Korean Sign Language Dictionary. We believe this conference will bring us one step closer to the new and improved dictionary and help us develop and gain a better understanding of the deaf society. Through active discussions, we can lay the groundwork toward a society free of communication barriers. As we touch on topics such as corpus-based dictionaries, dictionary structure and entries, and the future of sign language dictionaries with IT, I hope there will be meaningful discussions about the future direction of the Korean Sign Language Dictionary. I would like to once again express my deepest appreciation to our distinguished guests and members of the audience for attending the conference, as well as sign language experts from around the world who will be presenting today. I hope we can reap a fruitful and rich harvest from this conference for the future growth of sign language dictionaries. November 8, NIKL Conference Records of Visible Language: Sign Language Dictionary 5

4 축사 안녕하십니까? 문화체육관광부문화예술정책실장이우성입니다. 보이는언어의기록, 수어사전 국립국어원 2018 국제학술대회의개최를축하드리며, 이행사에참석하여자리를빛내주신모든분들께감사와환영의마음을전합니다. 특히, 이번학술대회를위해홍콩, 덴마크, 스웨덴, 호주, 독일등에서찾아와주신발표자여러분께특별한감사의인사를전합니다. 문화체육관광부문화예술정책실장이우성 올해는한글을창제하신세종대왕이즉위하신지 600돌이되는해입니다. 누구나자유롭게소통할수있는세상. 이것이세종대왕이꿈꾸던세상이고, 한글은바로그꿈의산물이아닌가합니다. 600년이흐른지금한글은한국어사용자들을잇는다리로서그소임을다하고있습니다. 2016년에제정된 한국수어언어법 에는한글창제에담긴정신을한국수어사용자와도함께나누기위한뜻이담겨있습니다. 한국수어언어법 은한국수어가한국어와동등한자격을가진농인의고유한언어임을천명하고, 한국수어사용자가자유롭게소통할수있는환경을만들어나가는것을목적으로하고있기때문입니다. 이번학술대회는이러한목표를이루기위한의미있는발걸음이라고생각합니다. 지난해문체부는 제1차한국수어발전기본계획 (2018~2022) 을발표한바있습니다. 이계획에따라한국수어교육자료개발, 한국수어교육기관지원, 한국수어교원자격제도운영, 한국수어문화정보구축, 한국수어사전편찬등한국수어발전을위해다양한정책을추진하고있습니다. 특히, 새롭게편찬하고있는 한국수어사전 은표제어, 뜻풀이, 용례모두한국수어가중심이되는사전을목표로하고있습니다. 농인과농인, 농인과청인을이어주는 한국수어사전편찬 은매우중요한작업입니다. 이번학술대회가우리수어사전의편찬수준을한층높일수있는계기가되길바랍니다. 오늘학술대회를위해멀리서오신발표자선생님을비롯한청중여러분들께다시한번감사드리며, 행사준비를위해애써주신소강춘국립국어원장님을비롯한관계자여러분께도감사의마음을전합니다. 유난했던여름에여봐란듯이쌀쌀한날씨가이어지고있습니다. 모두건강에유의하시면서즐겁게한해를마무리하시기바랍니다. 고맙습니다 국립국어원국제학술대회 보이는언어의기록, 수어사전

5 Congratulatory Remarks Greetings. I am Lee Woo-sung, Head of the Culture and Arts Policy Office at Korea s Ministry of Culture, Sports and Tourism. First of all, I would like to congratulate the opening of the National Institute of Korean Language s 2018 International Academic Conference, held under the theme of Records of Visible Language: Sign Language Dictionary. I would like to extend my warmest appreciation and welcome to everyone who came out to the conference today. In particular, I especially want to thank the presenters who travelled all the way from Hong Kong, Denmark, Sweden, Australia, and Germany. Head of Culture and Arts Policy Office Ministry of Culture, Sports and Tourism Lee Woo-sung This year marks the 600th anniversary of the enthronement of King Sejong, the inventor of the Korean alphabet Hangul. King Sejong dreamt of a world where all people can communicate freely, and Hangul is the product of that dream. Now, 600 years later, Hangul is fulfilling its role as a bridge between Korean language users. The Korean Sign Language Act, enacted in 2016, captures the determination to share the spirit of Hangul with sign language users. This Act declares Korean sign language as the native language of the deaf people with an equal status as Hangul, and aims to create an environment where Korean sign language users can communicate freely. I believe this conference is a meaningful step toward achieving this goal. Last year, Korea s Ministry of Culture, Sports and Tourism announced its first Basic Plan on Korean Sign Language Development ( ). Following this plan, many policies have been carried out for the advancement of Korean sign language, including development of educational resources, support for educational institutions, operation of an instructor qualification system, building culture information, and compilation of a dictionary. In particular, the Korean Sign Language Dictionary will consist of entries, definitions, and examples that are specific for Korean sign language. Compilation of a Korean Sign Language Dictionary is an important task, bringing together deaf people with other deaf people, as well as non-deaf people. I hope this conference will be a valuable opportunity to enhance the quality of the new dictionary. Once again, I would like to thank the presenters and the members of the audience for attending today s conference. I also want to extend my gratitude to Director General So Kang-Chun of the National Institute of Korean Language and his staff for all of their hard work in preparation of this event. With the scorching summer behind us, the chilly weather is finally making an ostentatious appearance. I hope you take care of your health, and I wish you all a good rest of the year. Thank you. November 8, NIKL Conference Records of Visible Language: Sign Language Dictionary 7

6 일정표 Program 시간 내용 09:30~10:00 ( 30) 등록 / Registration 10:00~10:30 ( 30) 개회식(개회사, 축사) / Opening Ceremony 1부 / Session 1 10:30~11:10 ( 40) 주제 1: 수어 말뭉치 중심의 수어사전 발표자: 트레버 존스턴(Trevor Johnston, 호주, Macquarie University) 11:10~11:50 ( 40) 주제 2: 수어사전의 구조 발표자: 예테 크리스토페르센(Jette H. Kristoffersen, 덴마크, University College Capital) 11:50~13:50 ( 120) 점심 식사 / Luncheon 2부 / Session 2 13:50~14:30 ( 40) 주제 3: 수어사전 편찬의 과제 발표자: 요한나 메시(Johanna Mesch, 스웨덴, Stockholm University) 14:30~15:10 ( 40) 주제 4: 대한민국 수어사전의 현황과 발전 방향 발표자: 최혜원, 이현화(한국, 국립국어원) 15:10~15:30 ( 20) 휴식 / Coffee Break 3부 / Session 3 15:30~16:10 ( 40) 주제 5: 미래의 수어사전 발표자: 토마스 항케(Thomas Hanke, 독일, Hamburg University) 16:10~16:30 ( 20) 폐회 / Closing 국립국어원 국제 학술 대회 보이는 언어의 기록, 수어사전

7 2018 국립국어원 국제 학술 대회 보이는 언어의 기록, 수어사전 Records of Visible Language: Sign Language Dictionary 목차 / Contents 1부 발표 Session 1 주제 1 수어 말뭉치 중심의 수어사전 11 트레버 존스턴_ 호주, 맥쿼리 대학교 A Corpus-oriented Sign Language Dictionary 28 Trevor Johnston_ Macquarie University, Australia 주제 2 수어사전의 구조 43 예테 크리스토페르센_ 덴마크, 유시시(UCC) 대학교 Structure of a Sign Language Dictionary 52 Jette H. Kristoffersen_ University College Capital, Denmark 2부 발표 Session 2 주제 3 수어사전 편찬의 과제 61 요한나 메시_ 스웨덴, 스톡홀름 대학교 Challenges of creating a Sign Dictionary 70 Johanna Mesch_ Stockholm University, Sweden 주제 4 대한민국 수어사전의 현황과 발전 방향 79 최혜원, 이현화_ 한국, 국립국어원 Current Status and Development Direction of Korean Sign Language Dictionary 95 Hyewon Choi, Hyunhwa Lee_National Institute of Korean Language, Korea 3부 발표 Session 3 주제 5 미래의 수어사전 113 토마스 항케_ 독일, 함부르크 대학교 Sign Language Dictionaries of Tomorrow 125 Thomas Hanke_ Hamburg University, Germany 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 9

8

9 주제 1 수어말뭉치중심의수어사전 트레버존스턴호주, 맥쿼리대학교 Trevor Johnston Macquarie University, Australia

10 1 부발표 수어말뭉치중심의수어사전 트레버존스턴호주, 맥쿼리대학교 개요 [ 슬라이드 2] 본강의에서는말뭉치중심수어사전, 특히 오슬란사인뱅크 (Auslan Signbank) 로알려진호주수어 ( 이하 오슬란 ) 온라인사전을다룬다. 본강의는두부분으로구성된다. 먼저파트 1에서는 1985년이후로출판되어온다양한오슬란사전의발전과역사를살펴본다. 특히 1984년에만들어졌으며 1985년이후책, 시디롬, 온라인형식으로출판되어온모든오슬란사전의기초가되는수어어휘데이터베이스를논의한다. 이들모두오슬란말뭉치 (Auslan Corpus) 가만들어지기전에출판되었다. 해당데이터베이스는 오슬란어휘데이터베이스 라불린다. 또다양한사전형식의항목구조에대해서도설명한다. 파트 2에서는오슬란사인뱅크를사례로들어말뭉치중심수어사전을설명한다. 오슬란사인뱅크는 2004~2007년사이에오슬란기록보관소가만들어지고 (Johnston, 2008) 해당기록보관소가오슬란말뭉치라는이름의언어말뭉치에주석이달리기시작하면서말뭉치중심수어사전이되었다. 본사전에기록된수어는현재크게두가지측면에서말뭉치에서확인된수어와연관이있다. 첫째는식별용주석을사용한것이고, 둘째는수어사용자와그들이사용하는수어관련말뭉치로부터정보를얻은사전의항목으로수어를통합한것이다. 파트 1: 말뭉치이전수어사전 [ 슬라이드 3-39] 오슬란어휘데이터베이스 [ 슬라이드 4-5] 오슬란어휘데이터베이스는엠에스워드 (MS word) 문서형태로된일련의표항목으로 1984 년에시작 되었다 년부터 2018 년까지 34 년동안해당데이터베이스는지속적인조사에따른수정및추가를통 국립국어원국제학술대회 보이는언어의기록, 수어사전

11 수어말뭉치중심의수어사전 트레버존스턴 해끊임없이개선되어왔다. 1984년이후로반복을거듭해온오슬란어휘데이터베이스는표제어로입력된기존오슬란어휘의저장소로간주되어왔다. 수어사용자들이서로의사소통할때만들어내곤하는다른유형의수어항목은체계적으로본데이터베이스에저장되지않는다. 비어휘적인이들수어는묘사 ( 또는 식별자 기호 ), 지시기호, 행위및몸짓으로불려왔다. 굴절형태소에해당하는일부표제어가변형되거나바뀐경우가있다. 동작주 피동작주역할이나상정보의동사를표시하는변화또는명사의복수형을표시하는변화가그예이다. 그처럼굴절된형태는사전에별도로들어가지않는다. 대부분의언어사전들과마찬가지로변형된형태는어휘보다는해당언어의문법에서논의되도록한다. 하지만경우에따라서는기호의이전굴절변경은사실상, 그리고불규칙적으로새롭고예측불가능한의미, 다시말해새로운어휘와관련이있다. 그와같은경우는사전에추가된다. 이조사는다음의몇가지기법을이용하여수행되었다. (1) 원어민수어사용자의직관과자기성찰, (2) 농인공동체에서의참여자관찰세션동안녹화한청각장애원주민수어사용자간의대화영상시청및분석, (3) 원주민수어사용자를대상으로실시한그룹 개인면담상에서개별기호의용법과의미를논의, 검사또는유도 2004년오슬란어휘데이터베이스는오슬란사인뱅크사전의온라인출판을목적으로대대적인정비를거쳤다. 이사전에는피드백메커니즘이적용되어있어서등록사용자들이누락된수어나부정확한정의를지적하거나보고할수있었다. 머지않아오슬란말뭉치가탄생했다. 그이후로오슬란사인뱅크상에이미기록되어있는수어와관련하여식별용주석의형태로말뭉치의주석을표시했다. 시간의경과에따른오슬란어휘데이터베이스 [ 슬라이드 6] 오슬란어휘데이터베이스상의데이터는 1984년이후로다양한유형의플랫폼사이를이동했다. 다양한버전의오슬란사전이기초를두고있는데이터셋인것이다. 오슬란어휘데이터베이스는 년부터엠에스워드 (MS Word) 의초창기버전에저장된기록약 2,000건의목록으로존재했다. 1990년대초엠에스워드 (MS Word) 표는애플의하이퍼카드 (HyperCard) 스택으로이동했다 ( ). 하이퍼카드스택덕분에컴퓨터상의각기호기록내에시각정보를쉽게포함시킬수있게되었다. 각기호항목 / 기록은라인드로잉 ( 라인드로잉이오슬란사전에등장한것은 1989 년이다 ) 을이용하여표시하였다. 1990년대초데이터는또다시상업적으로이용가능한데이터베이스로이동했는데처음은폭스프로 (FoxPro) 였고그다음이파일메이커프로 (FileMaker Pro) 였다. 이들새로운데이터베이스상의각기록은기호별로음운론, 의미론, 방언, 도상성등다양한정보가기호화된 100여개의필드를지니고있었다 ( 본강의파트 2에서이를다시다룰예정이다 ). 이들프로그램은일부필드에영상및그래픽정보를입력할수있었으며이를통해스프레드시트내에서각각의기호표제어를시각적으로보여주는것이가능했다. 이와비슷한신규오슬란어휘데이터베이스가 1990년대말에생성되었다. 파일메이커프로데이터베이 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 13

12 1 부발표 스의데이터를내보내시디롬을지원하는전용컴퓨터프로그램으로통합시키는방식이었다. 해당프로그램은멀티미디어시디롬오슬란사전을만들기위해사용되었다. 시디롬을지원하는프로그래밍언어는각기호와관련있는복수의필드, 그리고데이터베이스상의기호사이에, 다시말해동의성관계, 방언변이, 제한적용법등다양한기록및필드사이에수립된수많은교차참조를이용할수있었다. 2000년대초비디오필드를포함한파일메이커프로의전체데이터셋은시디롬과기본적으로기능이동일한온라인웹사이트로이동했다. 이웹사이트에서는파이선 (Python) 프로그래밍언어로구현된장고 (Django) 웹애플리케이션프레임워크를사용했다. 2004년에출범한이웹사이트는오슬란사인뱅크 (Auslan Signbank) 라는이름으로불렸다 (2004). 오슬란사인뱅크를지원하는스크립팅및프로그래밍은 2014년부터오픈소스가되었고깃허브 (Github) 웹사이트 ( 상에서제공된다. 2018년현재오슬란의항목수는 7,298개에이른다 ( 이가운데대략절반은 stand-alone 핵심표제어이고나머지절반은핵심표제어가변형된형태이다 ). 사전형식 : 인쇄, 시디롬, 온라인 [ 슬라이드 7-14] 오슬란어휘데이터베이스를이용하여출판된사전은세가지형태로나뉜다. 첫째, 인쇄형식 : 1987 호주수어 ( 오슬란 ) 예비수어사전, 1989 오슬란사전 : 호주농인공동체수어사전, 1997 호주수어 : 새로운오슬란사전, 2003 오슬란생존가이드 : 초보자를위한호주수어포켓사전. 둘째, 시디롬멀티미디어형식 : 1997 호주수어 : 새로운오슬란사전, 2002 오슬란의발견 : 초보자를위한호주수어사전. 셋째, 웹기반온라인형식 : 2004, 2008, 2018 오슬란사인뱅크 ( 사전의다양한버전과서로다른형식에도불구하고위에서언급한모든사전의구성원칙이바뀌지않았다는사실은주목할필요가있다. 지난수년동안있었던변화는 (1) 형식상의특징에따른기호의순차적처리와관련한사소한변경, (2) 용어검색및분류항목의데이터베이스기능개선, (3) 기존매체에서새로운매체로의전환, 즉인쇄형식에서시디롬형식, 나아가온라인형식으로의전환에따른적응및발전정도이다. 어떤형식으로든오슬란어휘데이터베이스는호주농인공동체의수어어휘를정확하게문서화하고자한다. 사전구조 [ 슬라이드 15-17] 오슬란사전은모두제한적인양방향, 2 개국어사전이었다. 단, 한가지예외가있었는데그것은바로 오슬란생존가이드 : 초보자를위한호주수어포켓사전 (Johnston & Schembri, 2003) ( 아래내용참 국립국어원국제학술대회 보이는언어의기록, 수어사전

13 말뭉치중심수어사전 트레버존스턴 조 ) 이다. 진정한사전이라면한언어의단어나기호는단순히나열되는것이아니라정의가내려진다. 2개국어사전에서는 A 언어의단어나기호가 B 언어로정의내려진다. 양방향사전의경우에는 B 언어로정의내려진 A 언어단어또는기호의목록과 A 언어로정의내려진 B 언어단어또는기호의목록으로구성되어있다. 그러한이유에서수어의양방향, 2개국어사전은어휘정보에접근하는두가지방법이존재한다. 하나는수어를통한방법, 다른하나는음성언어를통한방법이다. 한언어의각단어또는기호는상대언어의단어또는기호로정의내려진다. 오슬란사전의본문은영어 (B 언어 ) 로정의내려진오슬란기호 (A 언어 ) 를열거한다. 오슬란의표제어는손모양과위치등언어내적인구성원칙에따라순서대로나열된다. 인쇄사전의표제어는라인일러스트로나타냈다. 1990년대말까지는데이터베이스의표제어또한함노시스 (HamNoSys) 를이용한전용수어전사로나타냈으나일반인이볼수있는것은아니었고인쇄사전에실리지도않았다. 함노시스를이용한전사는기호의형태를알고자할때더이상필수적인것으로간주되지않는다. 영상클립 ( 또는이전버전의경우에는라인드로잉 ) 이포함된필드를확인하는것만으로도기호의형태를쉽게알수있기때문이다. 또음운론적특성과관련하여기호항목을분류또는설명하려고할때전사가필요하지않다 ( 과거에는손모양, 위치등에사용된유니코드심볼에순서를지정하는방식으로함노시스전사를분류하는것이가능했다.). 게다가새로운데이터베이스는특별한전사스크립트에의존하지않고도기호의음운론적특징또는그밖의특징을태그로달수있는상당수의필드를수월하게수용할수있다. 함노시스전사는 1990년대말까지생성되었고이후에는디지털영상기술과데스크톱성능이전용영상클립으로각각의표제어를나타내는것이가능할만큼충분히강력해졌고비용도저렴해졌다. 결국함노시스전사는중단되었다. 그에따라오슬란어휘데이터베이스의모든항목은영상클립으로만족해야하며음운론적형태를기호화하는관련필드를지니고있다. 인쇄, 시디롬, 웹사이트형식의사전에는오슬란-영어란또는데이터를볼수있는뷰가있다. 하지만세가지형식모두영어-오슬란뷰 (B 언어-A 언어 ) 는제한적이다. 인쇄형식의경우그이유는단순하다. 라인드로잉또는함노시스를이용하여오슬란으로영어표제어를정의하기란사실상불가능했기때문이다. 따라서비교방향은단순히각단어에해당하는오슬란번역어의항목번호를사용자에게알려주는영어단어색인을통해드러냈다. 시디롬과사인뱅크웹사이트형식의경우에는음운론적특성 ( 손모양, 위치, 움직임등 ) 및관련영어키워드에대한각기호항목 ( 기호형태 ) 을태그할수있다. 사용자는시디롬또는사인뱅크웹사이트의검색창에영어단어를입력하여해당키워드와관련있는기호를찾는다. 따라서여전히양방향사전의특징을지니고있지만제한적인의미에서만그러하다. 오슬란사전은형식상의특징 ( 아래참조 ) 에기반한원칙에따라표제어를분류하고배열했다. 본사전은수어초반우세손의모양을바탕으로하여수어를손모양섹션으로분류한다. 각섹션내에는그외에도손의위치와방향, 움직임및손의개수, 그리고경우에따라서는얼굴표정등형식상의특징이담겨있다. 비 ( 非 ) 우세손의모양과기호의대칭은우세손모양이동일한기호의분류및배열을결정한다 NIKL Conference Records of Visible Language: Sign Language Dictionary 15

14 1 부발표 인쇄사전의정렬원칙 ( 알파벳순서 ) [ 슬라이드 18-25] 여기서는기호항목의 알파벳순서 근간을이루는주요원칙만을다룬다. 더자세한내용은존스턴 (Johnston, 2003 년 ) 을참조하기바란다. 손모양기준 [ 슬라이드 18-20] 손모양은하나의기호를다른기호와구분하고여러기호를섹션으로분류하기위한원칙으로삼기에이상적이다. 오슬란의손모양은어느정도자의적인순서에따라정렬된다. 초창기오슬란사전 (1987년, 1989년 ) 의경우에는어휘에서차지하는손모양의빈도를바탕으로손모양의순서를정하되가장빈도가높은손모양 ( 평평한손모양 ) 이가장먼저나온다. 호주의수어 (1997) 에서는수를셀때사용하는오슬란손모양을바탕으로더자연스럽고직관적인순서를사용했다. 순서는 0부터시작해서수를셀때사용하는손모양을따른다. 이러한손모양의미세한변형은수를세는주요손모양들가운데에서순서가정해진다. 예를들어둥글게오므린손 (Round Hand) 은오슬란손모양으로숫자 0을의미하기때문에사전상에서가장먼저등장한다. 오케이모양의손 (Okay Hand) 은숫자 0 을의미하는대체가능한손모양이기때문에두번째로온다. 그다음은숫자 1을의미하는검지를세운손 (Point Hand) 이다. 검지를세운손과비슷한갈고리모양의손 (Hook Hand) 이그뒤에나오며이와같은식으로숫자 2부터 12까지의손모양을통해순서가정해진다 ( 오슬란에서는 12까지의숫자가운데그어느것도낮은숫자에서이미나온손모양을사용하지않는다 ). 오슬란에서사용하는손모양중몇가지는숫자와아무런관련이없으며임의적으로목록과사전의마지막에온다. 예를들어 SUN 은사전속둥글게오므린손 (Round Hand) 섹션, IGNORE 는검지를세운손 (Point Hand) 섹션, FRONT 는평평한손 (Flat Hand) 섹션에나온다 [ 슬라이드 20]. 기본위치기준 [ 슬라이드 21-22] 그다음으로오슬란사전은위치에따라기호의순서를나열한다. 수어공간또는신체상의높은위치에서표시한기호 ( 예 : 머리또는얼굴위에서의기호 ) 는낮은위치에서표시한기호 ( 예 : 허리또는팔위에서의기호 ) 보다먼저온다. 그러한이유에서 THINK 는 STOMACH 보다앞서는 WHO 의앞에온다. 손의개수기준 [ 슬라이드 23] 각각의손모양섹션내에서한손만을사용하는기호가두손모두를사용하는기호보다먼저나온다. 예를들어한손만사용하는기호 CONSIDER 는두손을모두사용하는변형기호보다앞선다. 동일한또는다른손모양기준 [ 슬라이드 24-25] 각각의손모양섹션내에서두손을사용해야하는기호는한손만사용해야하는기호보다뒤에나온다. 두손을사용하는기호내에서는양손의모양이같은기호가비 ( 非 ) 우세손의모양이다른기호보다먼저나온다. 그러한이유에서 CONSIDER 는 CROSS-OUT 보다앞선다. 두손을사용하지만비 ( 非 ) 우세손의모양이다른기호의순서는단순히사전전반의손모양순서를따른다. 결국우세손의모양이해당기호의손모양섹션을결정하고, 비 ( 非 ) 우세손의모양이우세손모양이동일한두손기호들사이의순서를결정한다. 그러한이유에서 QUESTION, POTATO, DEBT 는상대적으로이같은순서를따른다. 오케이모양의손은검지를세운손보다앞서고, 국립국어원국제학술대회 보이는언어의기록, 수어사전

15 말뭉치중심수어사전 트레버존스턴 검지를세운손은평평한모양의손보다앞서기때문이다. 생존가이드구조 [ 슬라이드 26] 오슬란생존가이드 : 초보자를위한호주수어포켓사전 (Johnston & Schembri, 2003) 은정확한의미에서사전이라기보다는단어목록이라할수있다. 그러한이유에서과거에비언어학자들이만든대부분의이른바수어사전들과다를바없었다. 이들은본질적으로단일어단어목록이었으며표제어는사실해당언어의알파벳문법에따라배열된공동체음성언어의단어였다. 표제어와동일한기호의일러스트는바로옆에위치해있었다 ( 생존가이드 는한정된특정독자를염두에두고출판되었다. 농인과기본적인의사소통을할수있을정도의매우기본적인어휘력을쌓고싶지만시간적여유가많지않은성인이그대상이다 ). 이같은유형의단순한어휘목록에대한특정한틈새분야가있지만오늘날언어학자들이만들고있고본강의에서다루고있는진정한수어사전과혼동해서는안된다. 시디롬구조 [ 슬라이드 27-33] 시디롬버전의 < 호주수어 >(1997) 는오슬란최초의멀티미디어사전이다. 오슬란어휘데이터베이스는시디롬제작에필요한데이터를제공했다. 시디롬의가장중요한혁신을꼽자면시디롬형식상의특징을기반으로기호를검색할수있다는점이다. 친숙한 영어단어로검색 하는방법 ( 인쇄형식사전의영어색인을참고하는방법 ) 대신 오슬란으로검색하기 옵션버튼을누르는방법으로기호검색을요청할수있다. 사용자는시디롬상에서 오슬란으로검색하기 옵션을선택한후다음의구체적인내용을묻는일련의추가질문에답해야한다. (1) 한손을사용하는기호인지, 두손을사용하는기호인지의여부, (2) 수어시작시우세손의모양, (3) 수어의위치, (4) 두손을사용하는기호의경우비 ( 非 ) 우세손의모양, (5) 2차적인위치 ( 즉, 두손이만나거나서로상호작용하는경우의비 ( 非 ) 우세손의위치 ) 그다음으로중요한시디롬의혁신은시멘틱필드 ( 음식, 신체부위, 감정, 건강등 ) 또는방언 ( 호주의주요방언지역또는주기준 ) 과같이다양한기준에따라즉시기호를선택하고분류할수있다는점이다. 세번째로중요한시디롬의혁신은미리설정된여러링크 ( 예 : 동의어, 반의어, 동형이의어 ) 를기반으로사전내의기호사이를즉시이동할수있다는점이다. 온라인사인뱅크구조 [ 슬라이드 34-39] 오슬란어휘데이터베이스는또한일반적으로는사인뱅크 (Signbank), 구체적으로는오슬란사인뱅크 (Auslan Signbank) 라고하는온라인버전오슬란사전의근간을이룬다. 사실데이터베이스자체는현재 웹사이트내에있다. 다시말해오슬란사인뱅크는오슬란어휘데이터베이스의웹사이트명에불과하다 NIKL Conference Records of Visible Language: Sign Language Dictionary 17

16 1 부발표 오슬란사인뱅크는단순히데이터베이스상의웹포털인것이다. 하지만연구원들은.csv 파일형태로사인뱅크웹사이트상의내용을다운로드받아추가데이터처리나조사를목적으로다양한데이터베이스로통합할수있다 ( 따라서데이터베이스버전은여전히파일메이커프로및기타스프레드시트형식으로남아있다.). 모든이전버전의오슬란사전과마찬가지로오슬란사인뱅크는영어정의와함께표제어로표시된언어의전통적인어휘적기호를포함하고있다. 오슬란사전에는묘사하는기호또는식별자기반의기호가다수포함되어있는데이들기호의어휘상태에는의심의여지가있다. 오슬란사인뱅크상에서사전을보여주는방식에몇가지혁신이적용되어있다. 첫째, 인터넷접속이가능한사람이라면누구든퍼블릭뷰를통해이용가능하다. 둘째, 등록사용자만이용가능한제한적뷰또는프라이빗뷰가있다. 등록뷰의경우데이터베이스에대한접근이여러단계로나뉘어져있다. 일반사용자는웹사이트에대한피드백 ( 사전상의기호오류또는누락보고등 ) 을작성하기위해등록을해야한다. 그외에도애플리케이션, 통역사, 연구자, 편집자, 출판사를위한네단계의조사용접근이가능하다. 데이터베이스내에는이와관련하여다양한접근및편집권한이존재한다. 데이터베이스열람및변경접근은다양한등급의사용자들에게웹사이트에대한서로다른단계의권한을부여하는방식으로통제한다. 추가적인페이지뷰에서는이미시디롬에서이용중인유형의정보를보여준다. 하지만시디롬과는달리오슬란사인뱅크는어느때라도업데이트가가능하며 ( 단, 접근권한을갖고있다는전제하에 ) 이때의변경사항은즉시효력을발휘하여뷰에반영된다. 또한새로운정보가이용가능해지면기록은삭제되거나추가될수있다. 이는오슬란사전이공동체의피드백또는오슬란말뭉치의점검을통해확보한새로운정보에신속하게대응할수있음을의미한다 ( 다음섹션참조 ). 오슬란사인뱅크의 2004년초기웹버전은당시의스크립트언어에따른제약때문에시디롬의기능몇가지만을지니고있었다. 2004년과 2018년사이에오슬란사인뱅크에사용된스크립트언어는전면적으로개선되었고이전까지시디롬에서만제공하던기능이웹사이트에서도이용할수있게되었다. 하지만현재이기능의대부분은특별접근권한을부여받은등록사용자만이용가능하다. 수어는인쇄버전의사전에서먼저부여받은내재적순서를지니고있지만웹사이트상에서의기호순서는인쇄사전에서만큼명확하지않다. 각각의페이지는단하나의기호항목을보여주므로사용자가사전한페이지에서다음페이지로이동하는경우에만순서가분명히드러난다. 가장최근인 2018년에이행된오슬란사인뱅크의주요혁신은 오슬란상의정의 필드가제공되는것이다. 이필드에는영상클립이삽입될수있으며이때해당표제어에대한오슬란상의정의가제공된다. 이러한유형의항목이생겨난것이불과최근의일이지만향후단일어사전에서처럼오슬란사인뱅크항목을보여주는것이가능할수있음을시사한다. 이는획기적인일이다. 어떤수어라도단일어사전은아직까지출판된적이없다. 수어용문자체계가존재하거나개발되지않고서는인쇄형태의수어단일어사전이란사실상불가능했다. 비록영상의도래에힘입어단일어수어사전이이론상가능해졌지만실행은여전히불가능했다. 그래도디지털영상기술이나오면서단일어사전제작이기술적으로가능해졌지만아직까지제작되거 국립국어원국제학술대회 보이는언어의기록, 수어사전

17 말뭉치중심수어사전 트레버존스턴 나출판된사전은하나도없다. 최신 2018 년버전오슬란사인뱅크는현재단일어항목생성을가능하게해 준다. 오슬란을이용하여오슬란표제어가정의내려지는항목말이다. 이기술을이용한최초의항목이생 성되기시작했다. 파트 2: 말뭉치중심수어사전 [ 슬라이드 40-70] 말뭉치중심수어사전에대한근거 [ 슬라이드 40-44] 앞에서설명했다시피오슬란사전은참여자관찰, 면담, 도출등을수반한 전통적 언어기록과서술연구프로그램의산물이다. 이와같은모델은아직문서화되지않고서술되지않은언어에대한연구를목적으로언어학과인류학에서폭넓게사용되고있다. 일반적으로이러한유형의연구를진행하는언어공동체는대체로서술의역사가없고지리적으로동떨어져있으며소규모인전통사회에서찾아볼수있다. 하지만수어공동체는이와는다른경향을보인다. 거의모든곳에서공동체수어는음성언어 ( 및때로는문자언어 ) 를사용하고구성원대부분이청인인공동체내의소수농인들이사용하고있다. 일부국가또는지역의몇몇농인들은수세대에걸쳐청각장애를앓아온가정에서태어나지만이는비교적드문경우이다. 농인대부분의부모와형제자매는청각장애를갖고있지않다. 그러므로대부분의수어를습득하여사용하는것은음성언어를습득하여사용하는것에비해매우이례적인것으로보인다. 농인공동체의잘알려진사회언어학적특성에따르면상당수연구자들은수어공동체가다른예상보다어휘와문법양측면에서훨씬높은수준의변화를보이는듯하다고말한다. 1990년대말과 2000년대초오슬란연구자들을포함한상당수수어연구자들에게분명해진사실이있다. 그것은바로수어의어휘와문법에대하여경험적으로건전한서술을하려면원어민또는그에가까운수어사용자의자연스러운수어를바탕으로언어학적말뭉치를만들어야한다는점이다 (Johnston, 2010, 2014). 언어학적말뭉치를만드는일은기계가읽을수있도록전사, 주석, 번역을통해언어사용기록을변환하는작업이수반된다. 수어의경우, 물론이는디지털영상기록물을제작하고멀티미디어주석소프트웨어를사용하는것을의미한다. 일단수어말뭉치를만드는준비과정에서기록보관소가마련되면해당수어의기존수어사전은여기에적응해야할필요가있다. 예를들어식별되어말뭉치에주석이달린기존의모든어휘기호들또한사전에실려야한다. 그렇지않을경우해당사전은불완전한것으로여겨질것이다. 사실상사전과말뭉치간의관계는두가지방향으로존재한다. 한편으로는사전이기호의식별과주석을알려주고, 다른한편으로는말뭉치가지금까지문서화되지않은기존기호와관련하여잠재적으로새로운데이터를제공하거나이미문서화된형태의가능한변이를식별해준다 NIKL Conference Records of Visible Language: Sign Language Dictionary 19

18 1 부발표 말뭉치주석과사전내기호항목간의결부 [ 슬라이드 45-49] 식별용주석 사전과말뭉치가조화를이루기위해서는사전에입력된기호유형의형태가말뭉치내의토큰 (token) 과부합할수있도록해야한다. 이말뭉치중심데이터베이스의기호항목은영어로된키워드와고유의식별용주석 (Identifying-gloss, ID-gloss) 과관련이있다. 고유의식별용주석은사전내각각의표제어마다배정되어있다. 사전내기호유형과관련한식별용주석은말뭉치내기호유형의모든토큰에주석을다는목적으로사용된다. 하지만식별용주석을효과적, 효율적으로사용하려면어휘유형이문서화되어있는참조데이터베이스가필요하다. 이것이바로말뭉치중심수어사전의정수이다. 해당언어의모든기존어휘기호는사전상에등재되며말뭉치주석으로분할되어표시를할때고유의식별자를갖게된다. 따라서말뭉치에주석을다는동안수어사전에끊임없이참조표시를해야한다. 이를위해오슬란주석자는오슬란사인뱅크를참조한다. 주석자는특정기호의식별용주석에확인이필요할때면언제든지해당기호와관련있는영어키워드를오슬란사인뱅크검색창에입력하고사인뱅크에서제시하는일치여부를확인한다. 그후주석자는말뭉치에서주시하고있는기호의형태와가장유사한기호의식별용주석을선택한다. 현재오슬란사인뱅크는말뭉치구축에사용되는 ELAN 주석프로그램의외부에있다. 주석자는 ELAN 상에서주석을만들때데스크톱상에온라인사전이열려있도록할뿐이다. 하지만사인뱅크 NGT( 오슬란사인뱅크기반의네덜란드수어용사인뱅크. 뒤에서이를다룰계획이다 ) 에서둘은서로연결되어있다 (Crasborn & Meijer, 2012; Crasborn 등, 2012). 예를들어사인뱅크 NGT는 NGT 말뭉치의주석을다는데사용되는 ELAN 소프트웨어템플릿과연결되어있다. 주석자는 ELAN상에기호의식별자주석을입력할수있으며사인뱅크 NGT상의모든관련식별자주석을포함하는통제어휘가생겨나게되면최적의조합을선택할수있을것이다. 만약의구심이들경우주석자는통제어휘내의어떠한주석이라도클릭할수있고사인뱅크로부터의식별용주석을위한섬네일영상클립이팝업창으로나타나게된다. 음운변이 [ 슬라이드 50-51] 가장흔한형태의기호 ( 말뭉치의토큰카운트에서가장빈번하게발생하는변이의횟수로평가 ) 는어휘기호의어간또는기본형태로여겨진다. 바로중심표제어인것이다. 표제어의변이는해당항목에서드러난다. 근본적으로변이는기본기호와는다른형식상의특징이하나또는많아야둘인기본기호의한형태이다 ( 변이는보통다른손모양, 위치또는수어에서사용되는손의개수를수반한다.). 또한변이는형태기반사전의적절한위치에고유의항목기록을갖고있다. 만약말뭉치에기록된것처럼사용데이터에따라기본및변이형태의지위가뒤바뀌어야된다고한다면그대로될것이다. 예를들어사용데이터는두손을모두사용한형태의기호가한손만을사용한형태보다훨씬더빈번하다는점을보여줄수있다. 만약 국립국어원국제학술대회 보이는언어의기록, 수어사전

19 말뭉치중심수어사전 트레버존스턴 한손만을사용한형태가사전상에중심표제어로올라와있을경우관계는뒤바뀌어야한다. 입증된사용 [ 슬라이드 52] 시간이지나면서, 그리고말뭉치가모든사용자와모든텍스트유형을대표하고또충분히크다면수어사전항목의대부분이말뭉치를통해입증된다는점을확인할수있다. 일반적인말뭉치에서찾아보기힘든기호는너무오래전에사용되어쓸모없어졌거나중요도가떨어지는기호정도일것이다 ( 예 : 특정분야의소규모사람들만이사용하는어휘 ). 말뭉치에서입증된사용은사전이경험적으로볼때견고하다는점을보여준다. 2012년, 4년간의주석사용이후 (Johnston 2012 참조 ) 오슬란사인뱅크의 3,500개핵심항목가운데약 75% 가오슬란말뭉치에서증명되었다. 당시오슬란말뭉치의규모 ( 약 70,000 토큰 ) 가음성언어의말뭉치에비해작다는점을고려하면이는매우높은비율이다 ( 오슬란사인뱅크의항목은 7,000여개로두배가넘지만이는모두 3,500개핵심어휘기호가변이된형태이다 ). 누락데이터 [ 슬라이드 53] 물론어떠한사전도완전하지는않기때문에지금까지기록되지않은기호가말뭉치에서식별될가능성도높다. 만약해당언어의진정한기존어휘기호로확인된다면일정기간동안공동체구성원들과의협의및말뭉치에서발견한추가토큰의확인을통해사전에추가될수있다. 이런식으로말뭉치는지금까지기록되지않은기호로어휘데이터베이스를확대할수있다. 그것이기존어휘항목의변이형태이든, 언어에새롭게추가된기호이거나 ( 신조어또는차용어 ) 단순히이전까지간과되었던기호이든말이다. 약 50개의새롭거나이전에기록되지않았던기존어휘기호가오슬란말뭉치에서식별되었고 4년간의주석사용이후오슬란사인뱅크에추가되었다. 사용데이터에근거한정의개선 [ 슬라이드 54] 애초에말뭉치에서의입증된사용을고려했을때지나치게좁은의미가기호의형태에부여되었을수있 다. 그러한경우해당기호의실제자연스러운사용을반영하여사인뱅크내의정의를개선하게된다. 이는 정의에대한변경및개선, 또는관련키워드의추가, 제거또는변경이될수있다. 일부기호의어휘지위조정 [ 슬라이드 55] 초기현장조사에근거하여일부매우상징적이거나묘사적인기호가기존어휘기호로사전에포함되었 을수있다 ( 직관, 자기성찰, 도출또는공동체피드백결과관습적이고어휘화된것으로보고된기호이다 ). 하지만말뭉치에서입증된것처럼기호를사용한다는것은기호가애초에생각했던것만큼구체적이거나 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 21

20 1 부발표 관습적 ( 또는 어휘적 ) 인의미를지니고있지않음을시사할수있다. 그와같은기호는어휘화측면에서경계선상의기호이다. 즉의미측면에서해당언어의특징을지니지못하고그러한이유로오슬란사전에적절하게포함되지못했을수있다 ( 물론해당언어의문법측면에서논의가필요할것이다 ). 따라서그러한항목은그지위를제대로반영할수있도록업데이트되거나아예제거되어야할수있다. 말뭉치에서사전까지? [ 슬라이드 56-57] 이상적인말뭉치구축상황에서는그누구도먼저언어에대하여기본적인어휘론적연구를수행한다음그결과로나온초기어휘를데이터베이스나사전에기록및반영하는일없이수어의기록을분할하고주석을달려고하지않을것이다. 하지만최근몇년동안대규모수어말뭉치프로젝트의일환으로몇가지수어사전프로젝트가시작되었다. 그목적은말뭉치를사용하여사전을만드는것에있다. 거의아무것도없는상황에서주석자체를바탕으로참조어휘를구축하는것은가능한일이다. 즉, 어휘의기초로말뭉치자체의입증된사례만을사용하는방법이다. 이와비슷한접근방식이 BSL 말뭉치와관련하여설명되었다 (Cormier 등, 2012). 비록이런식으로진행할수도있겠지만이접근방식은어렵고많은시간이소요되며일관성이떨어지기쉽다. 특히주석작업초기에주석자가이용할공통의참조점이없기때문이다. 연구자들은초기의데이터베이스와사전을끊임없이보완하고수정해야할상당수의다양한사용및변이형태와맞닥뜨릴가능성이높다. 이는커져가는어휘데이터베이스에도연쇄효과를일으킬수있다 (Crasborn, Meijer, 2012). 실제로일부연구자들은실증적인수어사전이참조말뭉치상에서입증된기호에만의존해야한다고생각한다. 어쩌면혹자는이를두고본강의에서언급한말뭉치중심수어사전이라기보다는말뭉치기반또는말뭉치주도수어사전이라할것이다. 하지만수어사전에기호를포함시키기위해그처럼엄격한기준을적용하기에는위험이따른다. 한언어의어떠한말뭉치라도 ( 무수히많은단어나기호와접하지않는한 ) 해당언어의알려진모든단어나기호의최소하나의토큰을포함할가능성은매우낮다. 만약매우기본적인일상어휘를말뭉치에서찾아볼수없게된다면특히문제가되며, 규모가그리크지않다는점에서이미구축되었거나구축중인수어말뭉치의경우에는그렇게될가능성이매우높다. 수어사전의사용자라면누구라도기본개념을위한기호가기록되어있지않다는사실을깨닫게될것이다. 말뭉치상에서과거에단한번도없었던일이기때문이다. 이는만족스럽지못한일이며사전의사용자는사전이불완전하다는타당한결론을내릴것이다. 만약입증된데이터를제한하고싶다면이문제를피할수있는방법한가지는관습적인또는서술적인텍스트뿐만아니라도출된텍스트와어휘까지아우르는말뭉치를자체적으로구축하는것이다. 그렇게되면기본적인어휘를도출하는기록과정에서작업을통합할수있다. 말뭉치결과를이용한사전항목강화 [ 슬라이드 58-68] 오슬란어휘데이터베이스와더불어오슬란사인뱅크는각항목의각기호에대한매우다양한정보를수 국립국어원국제학술대회 보이는언어의기록, 수어사전

21 말뭉치중심수어사전 트레버존스턴 용할수있도록만들어졌다. 해당정보는수많은전용필드나셀에서찾을수있다. 여기에는다음에관한정보가담겨있다. (1) 표제어의형태, (2) 표제어의의미, (3) 표제어의사용자, (4) 표제어의검색이나분류또는온라인상표시를위한기준. 이들정보상당부분은말뭉치자체를조사해보면찾을수있다 ( 예 : 방언선호도, 연령집단, 학교기반기호, 소속종교, 빈도등 ). 이정보는끊임없이검토되며사인뱅크연구자용인터페이스상에서쉽게편집가능하다. 일부정보는일반인이즉시열람할수있고또다른일부정보는데이터베이스의프라이빗뷰에서만연구자가참고목적으로열람가능하다. 동형이의어처리 [ 슬라이드 69-70] 오슬란사인뱅크는형태기반의기호데이터베이스이기때문에만약두기호가동형이의어인경우, 다시말해오슬란상에서는서로다른의미를갖고있지만기호형태가동일한경우 ( 예 : 금속용기 와 할수있는 을의미하는영어단어 can ), 두기호모두동일한형태를참조하게된다. 오슬란사인뱅크에서의해결책은영어사전에서와마찬가지이다. 즉표제어가되풀이되고생성된두개 ( 또는그이상 ) 의항목에서동일한음운론적형태에대하여두가지 ( 또는그이상 ) 의의미로각각규정한다. 오슬란사인뱅크에서는의미 1, 의미 2 등과같은식으로번호를부여한다. 다수의수어 [ 슬라이드 71-78] 오슬란사인뱅크 & 기타사인뱅크 [ 슬라이드 72] 오슬란어휘데이터베이스구조의기본원칙은처음만들어져데이터베이스에적용된이후로어떠한수어에도적용가능했다 (Johnston, 2001). 오슬란사인뱅크 ( 및인쇄, 시디롬초기형식 ) 의표제어는형태에기반한것이었다. 이들표제어는그형식상의특징이전용전사시스템 ( 함노시스등 ) 또는각기호가지닌관련형식상의특징 ( 손모양, 위치등 ) 의존재여부를명시한일련의필드에기호화되어있다. 그결과어떠한수어의기호라도데이터베이스내에수용될수있다. 단순히형태별로서술된항목이생성되는것이다 ( 만약데이터베이스가형태별로정렬된다면새로운기호는알파벳순서에따라적절한자리에들어가게된다.). 심지어기호들사이의상관관계유형 ( 가령동의어, 반의어, 동형이의어, 의미필드항목, 방언분포등 ) 이프로그램속으로구축되어있다면이또한어떠한수어의경우에도적절하게이용가능하다. 이러한잠재력이실현되자다른수어를연구하는타언어학자들도오슬란사인뱅크의기본아키텍처를본인의수어연구에도입하게되었다. 그결과네덜란드수어 (NGT) 사인뱅크, 영국수어 (BSL) 사인뱅크, 핀란드수어 (FinSL) 사인뱅크의온라인사전이제작되었고미국수어 (ASL) 사인뱅크도작업에착수했다 NIKL Conference Records of Visible Language: Sign Language Dictionary 23

22 1 부발표 동음어에서다수의수어까지 [ 슬라이드 73-77] 오슬란사인뱅크구조의한가지중요한점에힘입어단일사인뱅크, 다시말해범용어휘데이터베이스 (Universal Lexical Database) 에다수의수어를수용하는어휘데이터베이스구축의가능성이열렸다. 그러한데이터베이스는기존오슬란사인뱅크를타사인뱅크의템플릿으로사용하는데그치지않고범용사인뱅크 (Universal Signbank) 를구축해준다. 앞에서언급한것처럼오슬란사인뱅크는형성과정의특징에따라수어형태를기록해낸다. 그결과인쇄형식으로제공된경우에는라인드로잉또는사진이시각적인순서로정렬되고, 스프레드시트의기록과셀의형태로제공된경우에는유사한영상클립들이늘어서게된다. 비록사인뱅크같은온라인사전에서는페이지당하나의항목만눈에보이지만동일한원칙에따라각항목에기호화된관련형식상의특징별로데이터베이스내에서기록을즉시분류할수있다. 그러한이유로기호는손모양 ( 인쇄형식에서처럼 ) 이나위치, 또는요청받은다른특징별로정렬할수있다. 더욱중요한점은이러한형태기반항목들이쉽게기호화될수있고음운론적변이, 방언변이, 의미관계 ( 반의어, 동의어, 동형이의어 ) 및타수어 ( 영국수어, 미국수어등 ) 와의관계를대변한다는사실이다. 오슬란사인뱅크의기호형태는이제방언뿐만아니라수어에도태그로달릴수있다. 태그는기호형태가영국수어나미국수어에도존재한다는것을의미한다. 언어에태그를달수있고더불어다른기호와동형이의어관계에있는기호에태그를달수있게되면서확장가능한데이터베이스의가능성이열렸다. 어떠한수어로부터의기호라도동일한데이터베이스에들어갈수있다. 오슬란어휘데이터베이스가처음구상되었을당시에만해도컴퓨터의데이터처리속도 ( 및비용 ), 디지털영상기록물의품질, 서버의저장공간 ( 및비용 ) 에실질적인제약이있었다. 오늘날에는사인뱅크의항목수, 각항목과관련한코드의수와유형, 표제어나정의에사용되는영상클립의길이등은더이상실행불가능하지도, 많은비용이들지도않는다. 그결과이러한유형의범용사인뱅크의가능성과사용을시험해보려는노력이시작되었다. 이는글로벌사인뱅크 (Global Signbank) 라한다. 글로벌사인뱅크? [ 슬라이드 78] 글로벌사인뱅크는실현가능한가? 지금까지심층적으로연구되어온수어어휘의두가지관찰에따르면그가능성은충분하다. 첫째, 규모이다. 지난 50여년동안출판되어온수어사전대부분은수어표제어의수가수천개를넘지못한다 (3,000~8,000개사이 ). 이는대다수음성언어의사전, 심지어는문자가존재하지않는음성언어와비교해보더라도낮은수치이다. 둘째, 상당수연구자들은수어의도상성이높다보니아주똑같거나매우유사한형태의기호가서로다른수어에서거듭나타나는상황이발생하곤한다고밝혔다. 비록도상적으로자극을받은이들기호형태가 국립국어원국제학술대회 보이는언어의기록, 수어사전

23 말뭉치중심수어사전 트레버존스턴 종종서로관계없는수어사이에서똑같거나유사한의미를갖기도하지만그의미는전혀다를수도있다. 그럼에도불구하고이들수어는동일한기호형태를공유하고있다. 한가지수어를사용하는사람들이외국수어를본후에형태측면에서본인이사용하는수어의기호화매우유사해보인다고말하는경우가종종있다. 때로그들은그의미를정확하게추측할수있을것이고그러지못할때도있을것이다. 반면서로관계없는음성언어사이에서동일한단어형태가거듭나타나는경우는매우드물다. 그이유는다음과같다. 대부분의음성언어에서음운론적으로형태가올바른단어의잠재적수는음소와가능한음절수의조합을고려했을경우사실상무제한인것으로보이기때문이다. 하지만수어에서의상황은일부분도상성의영향때문에다를수있다. 가령글로벌사인뱅크가실현가능하다면사인뱅크의다국어버전이라할수있는글로벌사인뱅크는현재재구성되고있는구성상의순서로부터가장큰수혜를입을가능성이높다. 글로벌사인뱅크는심지어손모양이앞으로도가장 중요한 수어의특징으로남을가능성이높다하더라도대다수수어연구자들이받아들이는 중립적인 ( 다시말해, 특정언어의특징을지니지않는 ) 구성상의순서를갖게될것이다. 예를들어특정언어와관계없는순서는수를세는별도의체계보다는각손모양에서펼치고있는특정손가락의수에근거할수있다. 두가지이상의수어가동일하거나거의동일한기호형태를지니고있다면이들은마치오슬란사인뱅크상의동형이의어와마찬가지로글로벌사인뱅크상의동일한영상에태그가달릴것이다. 글로벌사인뱅크는수어음운목록, 도상성의정도와유형, 어휘화 문법화패턴등에대한교차언어 비교연구목적으로이용가능하다. 참고문헌 Cassidy, S., Crasborn, O., Nieminen, H., Stoop, W., Hulsbosh, M., Even, S., Komen, E.and Johnston, T Signbank: Softrware to Support Web Based Dictionaries of Sign Language. In N. Calzolari, K. Choukri, C. Cieri, T. Declerck, S. Goggi, K. Hasida, H. Isahara, B. Maegaard, J. Mariani, H. Mazo, A. Moreno, J. Odijk, S. Piperidis and T. Tokunaga (eds.). Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), May 7-12, 2018, Miyazaki, Japan. Cormier, K., Fenlon, J., Johnston, T., Rentelis, R., Schembri, A., Rowley, K., Adam, R. and Woll, B From Corpus to Lexical Database to Online Dictionary: Issues in Annotation of the BSL Corpus and the Devel-opment of BSL SignBank. In O. Crasborn, E. Efthimiou, E. Fotinea, T. Hanke, J. Kristoffersen and J. Mesch (eds.). Proceedings of 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 25

24 1 부발표 the 5th Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon [Workshop as part of 8th International Conference on Language Resources and Evaluation, LREC] (pp. 7-12). Istanbul, Turkey., Istanbul, Turkey May Crasborn, O & Meijer, A de From corpus to lexicon: the creation of ID-glosses for the Corpus NGT. In Proceedings of the 5th Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon (pp ). Crasborn, O., Hulsbosch, M. and Sloetjes, H Linking Corpus NGT annotatons to a lexical database using open source tools ELAN and LEXUS. In Proceedings of the 5th Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon (pp ). Istanbul, Turkey., Istanbul, Turkey May Johnston, T A preliminary signing dictionary of Australian Sign Language (Auslan). Adelaide: TAFE Na-tional Centre for Research and Development. Johnston, T Auslan Dictionary: A dictionary of the sign language of the Australian deaf community. Sydney: Deafness Resources Australia. Johnston, T (ed.) Signs of Australia: A new dictionary of Auslan. Sydney: North Rocks Press. Johnston, T The lexical database of Auslan (Australian Sign Language). Sign Language & Linguistics 4(1/2) DOI: /sll joh. Johnston, T., Thornton, D. and Napier, J Discovering Auslan: A beginner's dictionary of Australian Sign Language. Sydney: North Rocks Press. Johnston, T & Schembri, A (eds.) The Survival Guide to Auslan: a beginner's pocket dictionary of Australi-an Sign Language. Sydney: North Rocks Press. Johnston, T Language standardization and signed language dictionaries. Sign Language Studies 3(4) /sls Johnston, T Auslan Signbank ( Sydney: Royal Institute for Deaf and Blind Children & Catalyst Training Systems. Johnston, T The Auslan Archive and Corpus. In D. Nathan (ed.). The Endangered Languages Archive London: Hans Rausing Endangered Languages Documentation Project, School of Oriental and African Studies, University of London 국립국어원국제학술대회 보이는언어의기록, 수어사전

25 말뭉치중심수어사전 트레버존스턴 Johnston, T From archive to corpus: transcription and annotation in the creation of signed language corpora. International Journal of Corpus Linguistics 15(1) DOI: /ijcl joh. Johnston, T Lexical Frequency in Sign Languages. Journal of Deaf Studies and Deaf Education 17(2) Johnston, T The reluctant oracle: using strategic annotations to add value to, and extract value from, a signed language corpus. Corpora 9(2) NIKL Conference Records of Visible Language: Sign Language Dictionary 27

26 Session 1 A Corpus -oriented Sign Language Dictionary Trevor Johnston, Macquarie University, Australia Outline [Slide 2] In this lecture I describe a corpus-oriented sign language dictionary, specifically the on-line internet dic-tionary of Australian Sign Language (henceforth Auslan) known as Auslan Signbank. The lecture consists of two parts. In Part 1, I give the history and evolution of the various dictionaries of Auslan that have been published since In particular I discuss the lexical database of sign forms that was first created in 1984 and which underlies all the dictionaries of Auslan published since 1985 in book, CD-ROM, and internet formats. These were all published before the Auslan Corpus was created. The database is referred to as the Auslan Lexical Database. I also explain the structure of entries in the various dictionary formats. In Part 2, I describe a corpus-oriented sign language dictionary, using Auslan Signbank as an example. Auslan Signbank became a corpus-oriented dictionary after the creation of the Auslan Archive between (Johnston, 2008), and the beginning of annotation of that archive into a linguistic corpus, called the Auslan Corpus. The signs recorded in the dictionary are now related to the signs identified in the corpus in two main ways: by the use of ID-glosses and by incorporating into the entries for these signs in the dictionary of information gleaned from the corpus about signers and the signs they use. Part 1: The pre-corpus sign language dictionary [Slides 3-39] The Auslan Lexical Database [Slide 4-5] The lexical database of Auslan (or Auslan Lexical Database) began in 1984 as a series of tabular entries in a Microsoft Word document. Over the 34 years from 1984 to 2018 the database has been constantly improved with 국립국어원국제학술대회 보이는언어의기록, 수어사전

27 A Corpus-oriented Sign Language Dictionary Trevor Johnston corrections and additions as a result of on-going research. In all its iterations since 1984, the Auslan Lexical Database has been conceived as a repository of con-ventional lexical signs of Auslan, which are entered as headsigns. The database does not systematically include entries for the other possible types of signs that signers often produce when communicating with each other. These nonlexical signs have been called depicting (or classifier signs), pointing signs, en-actments and gestures. There are variations or modifications of some headsigns that represent inflectional morphology, e.g., changes that mark verbs for agent and patient roles or for aspectual information, or that mark nouns for plurality. Such inflected forms are not entered separately in the dictionary. As with most language dictionaries, these modified forms are left to be discussed in the grammar of the language, rather than in a lexicon. However, in some cases an erstwhile inflectional modification of a sign is actually and irregularly associated with a new and unpredictable meaning, i.e., a new lexical item. Such cases are added to the dictionary. The research was conducted using several techniques: (i) native signer intuitions and introspections; (ii) videotape recordings of natural signed exchanges between deaf native signers recorded during participant observation sessions in the deaf community were repeatedly viewed and analysed; and (iii) group and individual interviews of native signer informants in which the usage and meaning of individual signs were discussed, tested or elicited. The Auslan Lexical Database underwent a major overhaul for the on-line publication of Auslan Sign-bank dictionary in This dictionary included a feedback mechanism that allowed registered users to and comment and report on missing signs or inaccurate definitions. Soon after, the Auslan Corpus was created. From that point on, the annotation of the corpus was made with reference to the signs already recorded in Auslan Signbank, by way of ID-glosses. The Auslan Lexical Database over time [Slide 6] The data in the Auslan Lexical Database has migrated across several different types of platforms since It has been the dataset upon which the various editions of the Auslan dictionaries were based. From the Auslan Lexical Database existed as a list of approximately 2,000 records stored in a Microsoft Word processing program table (the earliest versions of Microsoft Word). In the early 1990s the Microsoft Word tables was migrated to an Apple HyperCard stack ( ). The HyperCard stack made it possible to include visual information easily within each sign record on a computer. Each sign entry/record was illustrated with a line-drawing (the line drawings appeared in the 1989 Auslan Dictionary). In the early 1990s the data was again migrated to commercially available databases: first to FoxPro, and then to FileMaker Pro. Each record in these new databases had more than 100 fields that encoded a variety of information about each sign, e.g., phonology, semantics, dialect, iconicity, etc. (I will mention some of these again in the second part of this lecture.) These programs allowed for video and graphic information to be entered in some fields, making it possible to have a visual representation of each sign headword within a spreadsheet NIKL Conference Records of Visible Language: Sign Language Dictionary 29

28 Session 1 A new parallel Auslan Lexical Database was created in the late 1990s by exporting the data from the FileMaker Pro database and incorporating it within into a dedicated computer program that supported the CD-ROM. The program was used to create a multi-media CD-ROM dictionary of Auslan. The programming language that supported the CD-ROM was able to exploit the multiple fields associated with each sign and the numerous cross references that had been established between signs in the database, i.e., between numerous records and fields, such as relationships of synonymy, dialectal variations, restricted usage, and so forth. In the early 2000s the entire FileMaker Pro dataset, including video fields, was migrated into an internet website which had the same basic functionality as the CD-ROM. The site used Django web applica-tion framework written in Python programming language. The site was called Auslan Signbank (2004). From 2014 the scripting and programming that supports Auslan Signbank and has become open source and house on a Github site ( Today (2018), Auslan had at last count 7,298 entries (approximately half of these are core stand-alone headsigns, and half are common variant forms of those core headsigns). Dictionary formats: print, CD-ROM and Internet [Slides 7-14] The dictionaries published using the Auslan Lexical Database have appeared in three formats: One, print format: 1987 A preliminary signing dictionary of Australian Sign Language (Auslan); 1989 Auslan Dictionary: A dictionary of the sign language of the Australian deaf community; 1997 Signs of Australia: A new dictionary of Auslan; and 2003 The Survival Guide to Auslan: a beginner's pocket dictionary of Australian Sign Language. Two, CD-ROM multi-media format: 1997 Signs of Australia: A new dictionary of Auslan; and 2002 Discovering Auslan: A beginner's dictionary of Australian Sign Language. Three, in a web-based internet format; 2004, 2008, 2018 Auslan Signbank ( It is important to note that despite the various editions and different formats, the organizing principles of all these dictionaries have remained unchanged. The changes that have happened over the years relate to (i) minor alterations in sign sequencing according to formational features, (ii) improvements to the functionality of the database in terms searching and sorting entries, and (iii) adaptations and developments occasioned by the transfer from old to new media, i.e., from printed book format to CD-ROMs through to the web-based application. In all its manifestations, the Auslan Lexical Database attempts accurately document the lexicon of the sign language of the deaf community in Australia. Dictionary Structure [Slides 15-17] The Auslan dictionaries have all been limited bidirectional, bilingual dictionaries. There has been one exception The Survival Guide to Auslan: a beginner's pocket dictionary of Australian Sign Language (Johnston 국립국어원국제학술대회 보이는언어의기록, 수어사전

29 A Corpus-oriented Sign Language Dictionary Trevor Johnston & Schembri 2003) (see below). In a true dictionary the words or signs of a language are defined, not just listed. In a bilingual diction-ary, the words or signs of Language A are defined in Language B. In a bidirectional dictionary one has a list of Language A words or signs defined in Language B, and a list of Language B words or signs defined in Language A. Thus in a bidirectional, bilingual dictionary of a sign language there should be two ways to access lexical information: one through the sign language and one through the spoken language. Each word or sign of each language is defined in the words or signs of the other language. The main part of the Auslan dictionaries lists Auslan signs (Language A) which are defined in English (Language B). The headwords of Auslan, i.e., the headsigns, are ordered according to language-internal formational principles such as handshape and location. In the print dictionaries, the headsign was repre-sented by a line illustration. In the database until the late 1990s, the headsigns were also are represented by a dedicated sign language transcription in HamNoSys, but this was not viewable by the public, nor included in the printed dictionary. Transcriptions in HamNoSys are no longer regarded as essential in order to know what the form of a sign is because one can easily see the form of a sign by looking at the field that contains the video clip (or, in earlier editions, line drawing). Nor are transcriptions required in order to sort or describe sign entries in terms of phonological features. (Previously, it was possible to sort the HamNoSys transcriptions them-selves by assigning an order to the Unicode symbols used for handshape, location, and so on.) Moreover, the new databases can easily accommodate large numbers of fields in which phonological and other features of signs can be tagged, without recourse to a special transcription script. The HamNoSys transcriptions were created up to the late 1990s, before digital video technology and desktop computing power became powerful and cheap enough to make it feasible to represent each headsign with its own dedicated videoclip. They have been discontinued. All entries in the Auslan Lexical Database, therefore, make do with a video clip, and have associated fields that code for phonological form. In the print, CD-ROM and web-site formats of the dictionary there is an Auslan to English section or view of the data. However, in all three the English to Auslan view (Language B to Language A) is limited. In the print format, the reason is simple: a definition of an English headword in Auslan using line draw-ings or HamNoSys was virtually impossible. Thus, this direction of comparison was represent simply by an English word index which pointed users to the entry numbers of relevant Auslan translation equiva-lents of each word. In the CD-ROM and Signbank web-site formats, one can tag of each sign entry (sign form) for its phonological characteristics (handshape, location, movement, etc.) and for the English key-words associated with each sign. The user typed an English word in the search field of the CD-ROM or the Signbank web-site to find signs associated with those keywords. Thus it is still bidirectional, but only in this limited sense. The dictionaries sorted and arranged the headsigns according to principles based on formational fea-tures (see below). The dictionary groups the signs into handshape sections on the basis of the handshape of the strong hand at the beginning of the sign. Within each section, other formational features, such as location, orientation, 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 31

30 Session 1 movement, the number of hands, and, in some cases, facial expression. The hand-shape on the weak hand, and the symmetry of the sign determine the grouping and sequencing of signs with the same dominant handshape. Ordering principles for print dictionary ( alphabetical order) [Slides 18-25] Only the main principles that underlie the alphabetic order of sign entries are discussed here. For full details, see Johnston (2003). By handshape [Slides 18-20] Handshape is ideally suited to act as the principle for distinguishing one sign from another and for grouping signs into sections. The handshapes of Auslan are ordered according to a semi-arbitrary sequence. The earlier editions of the Auslan dictionaries (1987, 1989) based the se-quence of handshapes on the frequency of occurrence of each in the lexicon, with the most frequent handshape the flat handshape appearing first. Signs of Australia (1997) used a more natural and intuitive sequence based on Auslan handshapes used in counting. The order follows the handshape used in counting starting from zero to then. Slight variations on these handshapes are ordered between the major counting handshapes. For example, the Round hand is first in the dictionary because it is the Auslan handshape for the number 0. The Okay hand is second because it is an alternative handshape for 0. The Point hand follows because it is the handshape for the number 1. It is followed by the Hook hand because the latter is similar to the Point hand (it resembles a hooked 1) and so on through handshapes for the numbers 2 through 12 (In Auslan, none of the numbers above twelve involves the use of a handshape not already found in lower numbers.) A few handshapes found in Auslan cannot be associated with numbers in any way and are arbitrarily placed at the end of the list and at the end of the dictionary. For example, SUN is found in the Round hand section of the dictionary, IGNORE is found in the Point hand section, and FRONT is found in the Flat hand section [Slide 20]. By primary location [Slides 21-22] The dictionary then sequences the signs according to location (tabulation). It lists signs made high in the signing space or high on the body (e.g., on the head or face) before those that are made low in the signing space or low on the body (e.g., on the waist or arm). Thus THINK comes before WHO which comes before STOMACH. By number of hands [Slide 23] Within each handshape section, the dictionary lists as a group the signs that use only one hand before those that use two hands. For example, the one-handed sign CONSIDER precedes its double-handed variant. By same or different handshapes [Slides 24-25] Within each handshape section, the dictionary arranges signs that require the use of two hands after those that require only one hand. Within these two-handed signs those that have the same handshape on both hands are listed before those that have a different handshape on the subordinate or weak hand. Thus CONSIDER precedes CROSS-OUT. The order of two-handed signs with a different handshapes on the weak hand simply repeats the order of handshapes in the dictionary overall. Thus the dominant or strong hand determines the handshape section of the dictionary the sign appears in, and the subordinate or weak handshape determines its ordering with respect to other two handed signs with the same 국립국어원국제학술대회 보이는언어의기록, 수어사전

31 A Corpus-oriented Sign Language Dictionary Trevor Johnston dominant handshape. Thus QUESTION, POTATO and DEBT appear in this order relative to each other because the Okay hand comes before the Point hand and the Point hand comes before the Flat hand. The structure of the survival guide [Slide 26] The Survival Guide to Auslan: a beginner's pocket dictionary of Australian Sign Language (Johnston & Schembri 2003) is a wordlist rather than a true dictionary. As such, it was not unlike most so-called sign language dictionaries that have been created in the past by non-linguists.. These were essentially mono-lingual wordlists, in which the headwords were actually the words of the community spoken language arranged according to the alphabetic writing conventions of that language. An illustration of a sign equated with the headword was place next to it. (The Survival Guide was published with a specific narrow audience in mind: adults with limited time, who only wanted to know very simple vocabulary items to facilitate basic communication with deaf people.) There is a specific niche for these types of simple wordlists but they should not be confused with true dictionaries of sign languages which present day linguists are creating and which are the topic of discussion here. Structure of CD-ROM [Slides 27-33] The CD-ROM edition of Signs of Australia (1997) was the first multi-media dictionary of Auslan. The Auslan Lexical database was again the source of data used in creating the CD-ROM. The most important innovation in the CD-ROM was the ability to search for signs based on their for-mational features. A sign search request could be launched selecting the search by Auslan button option rather than the familiar search by English word (by consulting the English index in the book format). After selecting the search by Auslan option on the CD-ROM the user was guided through a series of filtering questions which asked them to specify: (i) if the sign was one-, double- or two-handed; (ii) the handshape on the strong hand at the beginning of the sign; (iii) the location of the sign; and, if the sign was used two hands, (iv) the handshape on the weak hand, and (v) the secondary location (i.e., the place on the weak hand where the two hands contacted or interacted with each other). The second most important innovation was the ability to instantly select and sort signs according to various criteria, such as semantic field (food, body parts, emotions, health, etc.) or dialect (according to major dialect zones or states of Australia). The third most important innovation was the ability to jump instantly between signs within the diction-ary based on a number of pre-established links between them, such as synonyms, antonyms and homo-nyms. Structure of Internet Signbank [Slides 34-39] The Auslan Lexical Database also lies at the basis of the internet version of the Auslan Dictionary, called 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 33

32 Session 1 Signbank generally, or Auslan Signbank specifically. Indeed, the database itself now resides in the web-site. In other words, Auslan Signbank is just the website name of the Auslan Lexical Database in its cur-rent form: Auslan Signbank is just web portal on the database. It is however possible for researchers to download the contents of the Signbank site as a.csv file and then import the records into a variety of databases for further processing or investigation. (Versions of the database thus still exist in FileMaker Pro and other spreadsheet formats.) As with all previous versions of the dictionary Auslan Signbank contains conventional lexical signs of the language displayed as the headsign with a definition in English. The dictionary also contains a num-ber of depicting or classifier-based signs the lexical status of which is open to question. There are some innovations in the presentation of the dictionary in Auslan Signbank. First, there is now a public view available to anyone with an internet connection. Second, there is a restricted or private view available only to registered users. The registered view has several levels of access to the database. Ordi-nary users register in order to be able to post feedback on the site (including reporting errors or missing signs from the dictionary). There are four other levels of research access available on application, inter-preter, researcher, editor, publisher. These have various access and editing rights within the database. Access to viewing and/or changing the database is controlled allocating different classes of users with different levels of rights and privileges to the site. These additional page views display the type of infor-mation already exploited in the CD-ROM. However, unlike the CD-ROM, Auslan Signbank can be up-dated at any time (provided one has the access privileges) and these changes can be effected and dis-played immediately. Records may also be deleted or added as new information becomes available. This means that the dictionary is able to respond quickly to new information that comes available either through community feedback or through examination of the Auslan Corpus (see next section). The initial web version of Auslan Signbank (2004) had only some of the functionality found in the CD-ROM due to constraints imposed by the scripting language used at that time. Between 2004 and 2018, the scripting language behind Auslan Signbank was completely overhauled and functionality previously found only on the CD-ROM became available on the site. However, to date, most of this functionality is only available to registered users who have special access rights. The signs still have the inherent order first assigned in the print versions of the dictionary, but the or-dering of signs is not transparent on a web-site as it would be in a printed book. The order is only evident if one scrolls through the dictionary from one page to the next, as each page only presents one sign entry. The latest major innovation in Auslan Signbank, implemented in 2018, is the provision of an Defini-tion in Auslan field. In this field, a video clip can be inserted in which a definition in Auslan is given for the headsign. The creation of entries of this type has only recently begun, but it does mean that it will be possible, in the future, to present the Auslan Signbank entries as in a monolingual dictionary. This is ground-breaking. No monolingual dictionary of any signed language has yet been published. Without the existence or development of writing systems for signed languages, monolingual dictionaries of signed languages in printed form were virtually 국립국어원국제학술대회 보이는언어의기록, 수어사전

33 A Corpus-oriented Sign Language Dictionary Trevor Johnston impossible. Although monolingual sign language dictionaries became possible in principle with the advent of film and video, they were still totally impracticable. How-ever, with the availability of digitized video, it has become technically possible to make a monolingual dictionary, but none have yet to be made or published. This latest version of Auslan Signbank (2018) now makes it possible to create monolingual entries entries in which an Auslan headsign is defined using Auslan. The first entries using this technique have begun to be created. Part 2: corpus-oriented sign language dictionary [Slides 40-70] Rational for a corpus-based sign language dictionary [Slides 40-44] Recall that the Auslan dictionaries described above were the product of a traditional language documentation and description research program which involved participant observation, interviews, and elicitation. Such a model is widely used in linguistics and anthropology for work on undocumented and undescribed languages. Generally speaking, the language communities in which this type of research is undertake tend to be found in geographically discrete, small-scale traditional societies without a history of writing. Sign language communities, however, tend not to be like this. Almost everywhere, community sign languages are used by small numbers of deaf individuals who are embedded in much larger hearing communities which use a spoken (and sometimes written) language. In some countries or regions, some deaf people are born into families that have several generations of deafness, but this is relatively uncom-mon. Most deaf people have hearing parents and siblings. Thus the acquisition and use of most sign lan-guages appears quite a-typical when compared to that of most spoken languages. These and other well-known sociolinguistic characteristics of deaf communities mean that many researchers have remarked that sign language communities seem to display a much higher degree of variation in both lexis and grammar than might otherwise be expected. In the late 1990s and early 2000s, it became obvious to many sign language researchers, including Auslan researchers, that an empirically sound description of the lexicon and grammar of a sign language requires the creation of linguistic corpus based on naturalistic signing from native or near-native signers (Johnston 2010, 2014). The creation of a linguistic corpus entails the transformation of recordings of language use through transcription, annotation, and translation so that it becomes a machine readable resource. In the case of sign languages this, of course, means creating digital video recordings, and using multimedia annotation software. Once an archive of recordings has been created in preparation for creating a corpus of a sign language, any preexisting sign language dictionary of this sign language needs to orient itself towards it. For ex-ample, one needs to ensure that every conventional lexical sign that is identified and thus glossed in the corpus is also listed in the dictionary, otherwise the dictionary would clearly be judged as incomplete. Actually, the relationship between the dictionary and the corpus is two way: on the one hand, the diction-ary informs sign identification and annotation; 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 35

34 Session 1 on the other hand, the corpus provides potentially new data on hitherto undocumented conventional signs, or identifies possible variants of already documented forms. Relating corpus annotations to sign entries in the dictionary [Slides 45-49] ID-glosses In order to harmonize the dictionary and the corpus, it is necessary to be able to relate the form of sign types entered in the dictionary with tokens found in the corpus. In this corpus-oriented database, sign entries are associated with keywords and a unique identifying gloss in English. The unique identifying gloss (which I have called the ID-gloss) is assigned to each headsign in the dictionary. The ID-gloss associated with a sign type in the dictionary is used to annotate any token of this sign type in the Corpus. However, in order to use ID-glosses effectively and efficiently, one needs a reference database in which lexical types are documented. This is the essence of a corpus-oriented sign language dictionary: all conventional lexical signs of the language are listed in the sign dictionary and have a unique identifier when segmented and labelled in the corpus annotations. Thus during the annotation and glossing of the corpus, constant reference needs to be made to the sign language dictionary. To do this the Auslan annotator refers to Auslan Signbank. Whenever the ID-gloss for a particular sign needs to be checked, the annotator types in a English keyword associated with the sign into Auslan Signbank search window and views the matches suggested in Signbank. The annotator then select the ID-gloss for the sign that most closely resembles the form of the sign they observe in the corpus. Auslan Signbank is currently external to the ELAN annotation software used to create the corpus. The annotator simply has the on-line dictionary open on their desktop as they create annotations in ELAN. However, in Signbank NGT (a Signbank for Sign Language of the Netherlands based on Auslan Signbak we will discuss this later) the two are linked (Crasborn & Meijer 2012, Crasborn et al 2012). For exam-ple, Signbank NGT is linked to the ELAN software templates that are used to annotate the NGT corpus. Annotators can begin to type the ID-gloss of a sign annotation in ELAN and a controlled vocabulary listing all related ID-glosses in Signbank NGT will appear so they may choose the best fit. If in doubt, the annotator may click on any gloss within the controlled vocabulary and a thumbnail video clip for that ID-gloss from Signbank will appear in a pop-up window. Phonological variation [Slides 50-51] The most common form of a sign (judged by the most frequent occurrence of a number of variant in a token count from the corpus) is treated as the stem or base form of a lexical sign. It is the main headsign. Variants of the headsign are shown at that entry. A variant is basically a form of base sign that has one, or at most two, formational features which are different to the base sign (the variant usually involves a different handshape, 국립국어원국제학술대회 보이는언어의기록, 수어사전

35 A Corpus-oriented Sign Language Dictionary Trevor Johnston location, or the number of hands used in the sign). Variants also have their own entry record at the appropriate place in the form-based dictionary If usage data, as recorded in the corpus, suggests that the status of base and variant forms need to be reversed, they will be. For example, usage data may show that a double-handed form of a sign is far more frequent than a one-handed form. If the one-handed form has been listed as the main headsign in the dictionary, the relations should be reversed. Attested usage [Slide 52] Overtime, and providing that the corpus is representative of all users and all text-types, and also large enough, one should expect to find that the majority of one s sign language dictionary entries are attested in the corpus. The only signs expected not to appear in a general corpus are those already known to be archaic, obsolete or marginal (e.g., vocabulary only used by small groups of people in specialist fields). Attested usage in the corpus indicates that the dictionary is empirically sound. In 2012 after 4 years of annotation (see Johnston 2012) approximately 75% of the 3,500 core entries in Auslan Signbank had been attested in the Auslan Corpus. This is a very high percentage of the known lexicon given the relatively modest size of the Auslan Corpus at that time (approximately 70,000 tokens), when compared to spoken language corpora. (There is double this number of entries in Auslan Signbank, i.e., over 7,000 entries, but these are all variant forms of the core 3,500 lexical signs.) Missing data [Slide 53] Of course, no dictionary is ever complete so it is more than likely that hitherto unrecorded signs will be identified in the corpus. If these are confirmed as genuine conventional lexical signs of the language, then, after a period of consultation with community members and/or confirmation by additional tokens being found in the corpus, they should be added to the dictionary. In this way the corpus is able to augment the lexical database with hitherto unrecorded signs, either variants of existing lexical entries, or lexical signs which may be recent additions to the language (neologisms or borrowings) or which were simply previ-ously overlooked signs. Approximately 50 such new or previously unrecorded conventional lexical signs had been identified in the Auslan Corpus and so they were, in turn, added to Auslan Signbank, after 4 years of annotation. Improved definitions based on usage data [Slide 54] A sign form may be initially assigned a meaning which is too narrow given its attested usage in the corpus. In such cases, the definition in Signbank is improved to reflect the way the sign is actually used in naturalistic signing. This could mean changing and improving the definitions, or adding, removing or changing associated keywords NIKL Conference Records of Visible Language: Sign Language Dictionary 37

36 Session 1 Adjust the lexical status of some signs [Slide 55] Some highly iconic or depicting signs may have been entered in the dictionary as conventional lexical signs on the basis of initial fieldwork (intuitions, introspection, elicitation or community feedback had reported the sign as conventional and lexcialized). However, the use of the sign as attested in the corpus may indicate that the sign does not have a meaning which is as specific or conventional (or lexical ) as first thought. Such signs are borderline signs in terms of lexicalizaiton: they may not be language-specific in their meanings and thus not really appropriately listd in a dictionary of Auslan. (Of course, they would still need to be discussed in a grammar of the language.) Such entries may therefore need to be updated to better reflect their status, or even removed. From corpus to dictionary? [Slides 56-57] In the ideal corpus-building situation, it is not expected that one would begin to segment and gloss a re-cording in a sign language without first having conducted basic lexicological research into the language, and then having recorded and described the resulting initial lexicon in a database or dictionary. However, in recent years some sign language dictionary projects have begun as part of large sign language corpus projects. The aim is to use the corpus to build the dictionary. It is possible to construct a reference lexicon more or less from scratch based the annotations them-selves as they are being made, i.e., using only attested instances in the corpus itself as the basis for one s lexicon. Something akin to this approach has been described for the BSL corpus (Cormier et al., 2012). Though this may provide a way to proceed, this approach is difficult, time consuming and prone to inconsistency because annotators have no common reference point, especially in the early stages of annotation and glossing. Researchers are likely to encounter many different uses and variant forms that require constant revision and correction of the incipient database and dictionary. This can also have knock on effects in the growing lexical database (Crasborn and Meijer, 2012). Indeed, some researchers believe that an empirical dictionary of a sign language should only draw on signs attested in the reference corpus. One could perhaps call these acorpus-based or corpus-driven sign language dictionaries rather corpus-oriented sign language dictionaries, like the one described here. However, there are dangers in adopting such a strict criteria for inclusion of signs in a sign language dic-tionary it is highly unlikely that any corpus of a language (unless it runs into hundreds of millions of words or signs) will include at least one token of every known word or sign in the language. This is par-ticularly problematic if very basic everyday vocabulary items just happen not to be found in a corpus, something that is highly likely to be the case in the sign language corpora that have been or are in the process of being created because they are really not that large. Any user of such a dictionary may find that signs for basic concepts is not recorded because it simply never occurred in the corpus. This is not satisfactory and users of such a dictionary would rightly conclude that it is incomplete. One way to avoid this problem, if one wants to restrict oneself to attested data, is to construct one s 국립국어원국제학술대회 보이는언어의기록, 수어사전

37 A Corpus-oriented Sign Language Dictionary Trevor Johnston corpus so that it includes elicited text and vocabulary items, not just conversational or narrative text. One could then incorporate tasks during recording sessions that elicit basic vocabulary items. Enriching dictionary entries from corpus findings [Slides 58-68] The Auslan Lexical Database, and hence Auslan Signbank, is designed to accommodate a large variety of information about each sign at each entry. This information is found in scores of dedicated fields or cells. These contain information on: (i) the form of the headsign; (ii) its meanings; (iii) who uses it; and (iv) criteria by which the entry can be searched, sorted or displayed on-line. Much of this information emerges from the study of the corpus itself (e.g., dialect preferences, age cohorts, school-based signs, religious affiliation, frequency, etc.) This information is constantly under review and is easily editable in the Signbank researcher interface. Some information is visible to the public immediately, other information is only visible on the private view of the database, where it is available for researchers to consult. Dealing with homophones [Slide 69-70] Because Auslan Signbank is a form-based database of signs, it follows that if two signs are homonyms, i.e., two distinct meanings in Auslan have the same sign form (cf. English can meaning metal container and able to ), then they would both reference the same form. The solution in Auslan Signbank is the same as in an English dictionary the headword or headsign is repeated, and at the two (or more) entries created the same phonological form is defined with each of its two (or more) distinct meanings. In Auslan Signbank these are numbered: Sense 1, Sense 2, and so on. Multiple sign languages [Slides 71-78] Auslan Signbank & other Signbanks [Slide 72] From their first conception and application in the Auslan Lexical database, the principles underlying its structure were claimed to be applicable to any sign language (Johnston 2001). Auslan Signbank (and its earlier instantiations in book and CD-ROM formats) has form-based entries. These entries have their formational features encoded either in a dedicated transcription system (such as HamNoSys) or in a series of fields that specify the presence or absence of relevant formational features of each sign (such as handshape, location, etc.). The result is that signs from any sign language can be accommodated within the database entries are simply created which are described by form. (If the database is ordered by form, then the new sign is simply entered at its appropriate slot according to alphabetical order.) Even the types of interrelationships between signs (such as synonymy, antonymy, homonymy, semantic field membership, dialect distribution, etc.) that are built into the 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 39

38 Session 1 program can also be exploited, as appropriate, for any language. Realization of this potential has lead other linguists working on other sign languages to adapt the basic architecture of Auslan Signbank for their own sign languages. As a result, on-line dictionaries have been created for Sign Language of the Netherlands (NGT Signbank), British Sign Language (BSL Signbank), and Finnish Sign Language (FinSL Signbank), and work on an American Sign Language (ASL Signbank) has commenced. From homophones to multiple sign languages [Slide 73-77] One important aspect of the structure of Auslan Signbank opens up the further possibility of creating a lexical database that accommodates multiple sign languages in the one Signbank, i.e., in a Universal Lex-ical Database as it was described in Johnston (2001). Such a database does not simply use the existing Auslan Signbank as a template for other Signbanks, it creates a Universal Signbank. Recall, that Auslan Signbank creates records of sign forms according to formational features. This re-sults in a visual ordered sequence of line drawings or photographs when presented in print format, or in columns of similar video clips when presented as records and cells in a spreadsheet. Though only one entry per page is visible in an on-line dictionary such as Signbank, the same principle allows instant sort-ing of records within the database by any one of the relevant formational features encoded with each entry. Thus signs may be ordered by handshape (as in the print format), or by location, or by any other feature requested. More importantly, as we have seen, these form-based entries also easily encoded and represent phono-logical variation, dialect variation, semantic relations (antonyms, synonyms, homonyms), and relations to other sign languages, such as BSL and ASL. In Auslan Signbank sign forms can now be tagged not only dialect, but also sign language. The tag means the sign form exists in BSL or ASL. The ability to tag for language together with the ability to tag a sign as being in a homonym relation with another creates the potential for an open-ended database. Signs from any sign language can be en-tered into the same database. When the Auslan Lexical Database was first conceive there were practical limitations on personal computer processing speeds (and cost), the quality of digital video recordings, and the storage capacity (and cost) of servers. Today, the number of entries in Signbank, the number and type of codes associated with each entry, or the length of the video clips used as the headsign or as definitions are no longer impracticable nor too costly. As a result, work to test the possibilities and uses of this type of universal Signbank, has begun. It is called Global Signbank. Global Signbank? [Slides 78] Is a Global Signbank practical? Two observations about the lexicons of sign languages that have been studied in depth to date suggests it may well be. First, size. Most dictionaries of sign languages that have been published over the past 50 years do not seem to contain more than several thousand main headsigns (somewhere between 3,000 to perhaps 8,000). This appears relatively modest compared to most spoken language dictionaries, even spoken languages without a written form 국립국어원국제학술대회 보이는언어의기록, 수어사전

39 A Corpus-oriented Sign Language Dictionary Trevor Johnston Second, many researchers have remarked that the high degree of iconicity in sign languages has tended to create a situation in which the same or very similar sign forms reappear repeatedly in different sign languages. Though these iconically motivated sign forms often have the same or similar conventional meaning in unrelated sign languages, their meanings may also be completely different. Nonetheless they share the same sign form. It is not uncommon for signers of one sign language reporting that when they view a foreign sign language they often see signs that appear to be very similar in form to ones in their own sign language. Sometimes they may be able to correctly guess the meaning, often they will not be able to. By contrast, it is quite rare to see the same word form appear repeatedly in unrelated spoken languages. The reason for this is that the potential number of phonologically well-formed words in most spoken languages appears to all practical purposes to be without limit when the combination of phonemes and the possible number of syllables are taken into account. However, the situation in signed languages may be different partly due to the impact of iconicity. Assuming it is practicable, a multiple language version of Signbank Global Signbank would most likely benefit from the current formational order being re-designed. Global Signbank would preferably have neutral (i.e., non-language specific) formational order accepted by a majority sign language re-searchers, even if the primary feature is most likely to remain handshape. A non-language specific order could, for example, be one based on the number of selected fingers extended in each handshape, rather than any given counting system as such. Whenever two or more sign languages have an identical or near identical sign form they would be tagged to the same video in Global Signbank, just as are homophones in Auslan Signbank. Global Signbank could be used for the cross-linguistic and comparative study of the sign language phonological inventories, degrees and types of iconicity, and lexicalization and grammati-cization patterns. References Cassidy, S., Crasborn, O., Nieminen, H., Stoop, W., Hulsbosh, M., Even, S., Komen, E.and Johnston, T Signbank: Softrware to Support Web Based Dictionaries of Sign Language. In N. Calzolari, K. Choukri, C. Cieri, T. Declerck, S. Goggi, K. Hasida, H. Isahara, B. Maegaard, J. Mariani, H. Mazo, A. Moreno, J. Odijk, S. Piperidis and T. Tokunaga (eds.). Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), May 7-12, 2018, Miyazaki, Japan. Cormier, K., Fenlon, J., Johnston, T., Rentelis, R., Schembri, A., Rowley, K., Adam, R. and Woll, B From Corpus to Lexical Database to Online Dictionary: Issues in Annotation of the BSL Corpus and the Development of BSL SignBank. In O. Crasborn, E. Efthimiou, E. Fotinea, T. Hanke, J. Kristoffersen and J. Mesch (eds.). Proceedings of the 5th Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon [Workshop as part of 8th International Conference on Language Resources and Evaluation, LREC] (pp. 7-12). Istanbul, Turkey., Istanbul, Turkey May NIKL Conference Records of Visible Language: Sign Language Dictionary 41

40 Session 1 Crasborn, O & Meijer, A de From corpus to lexicon: the creation of ID-glosses for the Corpus NGT. In Proceedings of the 5th Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon (pp ). Crasborn, O., Hulsbosch, M. and Sloetjes, H Linking Corpus NGT annotatons to a lexical database using open source tools ELAN and LEXUS. In Proceedings of the 5th Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon (pp ). Istanbul, Turkey., Istanbul, Turkey May Johnston, T A preliminary signing dictionary of Australian Sign Language (Auslan). Adelaide: TAFE National Centre for Research and Development. Johnston, T Auslan Dictionary: A dictionary of the sign language of the Australian deaf community. Sydney: Deafness Resources Australia. Johnston, T (ed.) Signs of Australia: A new dictionary of Auslan. Sydney: North Rocks Press. Johnston, T The lexical database of Auslan (Australian Sign Language). Sign Language & Linguistics 4(1/2) DOI: /sll joh. Johnston, T., Thornton, D. and Napier, J Discovering Auslan: A beginner's dictionary of Australian Sign Language. Sydney: North Rocks Press. Johnston, T & Schembri, A (eds.) The Survival Guide to Auslan: a beginner's pocket dictionary of Australian Sign Language. Sydney: North Rocks Press. Johnston, T Language standardization and signed language dictionaries. Sign Language Studies 3(4) /sls Johnston, T Auslan Signbank ( Sydney: Royal Institute for Deaf and Blind Children & Catalyst Training Systems. Johnston, T The Auslan Archive and Corpus. In D. Nathan (ed.). The Endangered Languages Archive London: Hans Rausing Endangered Languages Documentation Project, School of Oriental and African Studies, University of London. Johnston, T From archive to corpus: transcription and annotation in the creation of signed language corpora. International Journal of Corpus Linguistics 15(1) DOI: /ijcl joh. Johnston, T Lexical Frequency in Sign Languages. Journal of Deaf Studies and Deaf Education 17(2) Johnston, T The reluctant oracle: using strategic annotations to add value to, and extract value from, a signed language corpus. Corpora 9(2) 국립국어원 국제 학술 대회 보이는 언어의 기록, 수어사전

41 Ҷ Ҷ ਙ Ҷ ೡ ധ ࠃ י ӟ 3FDPSET PG 7JTJCMF -BOHVBHF 4JHO -BOHVBHF %JDUJPOBSZ 주제 2 수어사전의 구조 예테 크리스토페르센 덴마크, 유시시(UCC) 대학교 Jette H. Kristoffersen University College Capital, Denmark

42 1 부발표 수어사전의구조 - 예테크리스토페르센 덴마크, 유시시 (UCC) 대학교 초록 수어사전학은수어연구계에서신생학문에해당된다. 수어사전학자가직면하는가장큰어려움은수어가시각양식 (visual modality) 을사용하는언어로수어사용자간널리사용되는문어체계가없다는점이다. 지난수십년간수어사전학은전자매체의혜택을크게누렸다. 수어사전은더이상과거와같이어휘만을정리한목록이아닌구어사전과유사한정보를담은과학에기반한사전이다. 하지만수어의시각적요소는어휘검색과수어 (sign) 에관한정보수록측면에서여전히과제로남아있다. 본논문에서는덴마크수어사전 (Danish Sign Language Dictionary, Kristoffersen & Troelsgård, 2016) 의구조를소개하고수어사전의구조를제안하고자한다. 1. 서론 가장최근까지도대부분의수어사전은구어에서해당되는어휘와수어를나타내는그림, 사진또는설명 을수록한어휘목록에불과했다. 이때원천언어는주로구어에해당되며수어가목표언어이다. 지난수십년간수어사전학은빠르게발전하였으며일련의과학적수어사전이편찬되었다. 그첫시도중하나로언어학적원칙에기반한미국수어사전 (A Dictionary of American Sign Language on Linguistic Principles, Stokoe et. al. 1965) 이있다. 오늘날수어사전은더이상어휘의나열이아닌구어사전과유사한정보를담은진정한의미의사전이다. 또한수어가원천언어에해당된다. 수어사전학자는기술발전의혜택을크게누렸다. 과거에는수어를사전에표기하는방식을결정하는데어 려움을겪었다면, 오늘날에는수어를디지털동영상형태로기록하여쉽게전자사전에통합시킬수있다 국립국어원국제학술대회 보이는언어의기록, 수어사전

43 수어사전의구조 예테크리스토페르센 (Zwitserlood et al, 2013 참조 ). 하지만여전히수어사전학자가겪는몇가지어려움이있다. 예를들어, 수어를지칭하거나기호검색을수행하고검색한결과, 기호의순서정렬등을위해서는수어에대한어떠한공식표기법이필요하다. 또한수지적요소 (manual features) 검색이가능하도록하기위해서수어를원천언어로하는수어사전은구어사전에비해고도화된검색기능이필요하다. 덴마크수어사전 (Kristoffersen & Troelsgård, 2016) 은사실상덴마크수어 (DTS)-덴마크어이중언어양방향사전이지만, 그구성은이중언어사전과단일어사전의중간정도에해당된다 (Kristoffersen & Troelsgård, 2012). 표제어는 DTS 기호만을기준으로하며수어의의미에따라정렬되어있고, 해당되는덴마크어의동형이의어또는다의어를구별하기위한설명을제외한, 용례를포함한거의모든정보는 DTS를기준으로작성되어있다. 덴마크어휘에대한정보는기존의 덴마크단일어범용사전 으로쉽게확인할수있기에본사전에서는 DTS를중심으로기술하도록이접근법을채택했다. 따라서덴마크수어사전은표제어의정의대신해당사전의또다른메타언어인, 검색가능한대응덴마크어를수록한단일어사전이라할수있다. 2. 덴마크수어사전표제어배열 덴마크수어사전의표제어는 4 개부분으로구성되어있다 ( 그림 1 참조 ). 섹션 1: 표제어항이섹션은수어표제어의사진과용어해설 (gloss) 을담고있다. 첫번째수위 (location) 와제일두드러지는수형 (handshape) 을아이콘으로설명한다. 마지막으로버튼을클릭하여해당표제어해설을출력할수있도록별도창을열거나전체사전에수록된기호를포함한모든용례를담은용어색인을확인할수있다. 섹션 2: 동영상창수어표제어의기본형을볼수있다. 단어의음운적변이형 (variant) 이있는경우, 동영상상단에각변이형별재생버튼으로확인할수있다. 컨트롤버튼으로재생속도를낮추거나창을확대할수있다. 용례등의동영상은다른섹션에있는재생버튼을클릭하여확인할수있다 NIKL Conference Records of Visible Language: Sign Language Dictionary 45

44 1 부발표 섹션 3: 의미창섹션3에서는수어단어의뜻을설명한다. 대응덴마크어혹은단어의용법에관한설명 ( 기능어의경우 ), 입모양에대한정보, 동의어에관한상호참조, 사용제한설명및예문등이기술되어있다. 의미상이해하기어려운합성어도일반적인뜻아래기술하고있다. 대응덴마크어는덴마크어사전 (Danish Dictionary) 1) 에서검색할수있으며연결링크가제공된다. 따라서사용자들은해당덴마크어의뜻과용법에대해더자세히알아볼수있다. 섹션 4: 추가정보 동일한것으로보이는수어 ( 동형이의어 ) 와분류사의경우에는자주사용되는고정형태 (frozen form) 에 관한상호참조를제시한다. 또한표제어와흔히함께사용되는수어도기록하고있다. [ 그림 1. 덴마크수어사전표제어의 4 개섹션구성예시 ] 1) 덴마크어사전 (Danish Dictionary) 은현대덴마크어를기록한단일어, 범용사전이다. 사전은 에서무료로 사용할수있다 국립국어원국제학술대회 보이는언어의기록, 수어사전

45 수어사전의구조 예테크리스토페르센 3. 덴마크수어사전검색기능 3.1. 검색기준선택 덴마크수어사전에등재된수어는여러카테고리에따라검색할수있다. 수형, 수위, 주제및텍스트로 검색이가능하다. 수형검색검색이가능한수형은총 65가지이다. 검색기준으로사용될수형은연관수형그룹별로선택되지만원치않는수형은개별적으로제외할수있다. 이방식으로사용자는기본적으로묶이는수형그룹과별개로단일수형또는원하는유형의수형을선택할수있다. 그림 2는집게손가락을편유형을검색한수형선택창의예시를보여주고있다. [ 그림 2. 수형선택창부분 ] 움직이는손 (active hand) 과수동적인손 (passive hand) 별특정수형검색이가능하다. 수위검색수위는 15개의신체부위별혹은신체와가까운위치의수위를나타내는수위아이콘페이지에서선택할수있다. 또한 4개의그룹 ( 선택창상단 ) 중에서수위를선택하는방법도있다. 그룹을지정하면해당그룹에속한모든수위가검색기준에포함된다. 추가로 ( 우측열에서 ) 두개의수위를선택할수있다. 하나는전체몸의앞부분이며다른하나는얼굴앞수위이다. 그림 3은수위선택창을나타낸다 NIKL Conference Records of Visible Language: Sign Language Dictionary 47

46 1 부발표 [ 그림 3. 수위선택창 ] 텍스트검색대응덴마크어, 수어용어해설 ( 글로스 ) 및예문 ( 표기와번역문 ) 에대한텍스트검색이가능하다. 이방법을통해기본형은아니지만예문에나타나는수어단어를찾을수있다. 구단위검색및불특정검색도가능하다. 주제검색 70가지주제중에서검색기준을선택할수있다. 70가지주제는 9개그룹과 2개의하위그룹으로구분된다. 상위그룹에해당되는주제를지정하면자동적으로모든하위주제가검색기준으로선택된다. 그림 4 는주제구성체계를묘사하고있다. 복수의검색기준을결합하여검색도가능하다. 카테고리별로는 또는 (or) 조건이적용되며검색카테고 리의조합으로검색하는경우 그리고 (and) 조건을사용한다. 현재선택한기준은화면의좌측에표시된다. 그림 4 는 집게손가락을편 수형과 종교 주제로검색한수 어목록을보여주고있다 국립국어원국제학술대회 보이는언어의기록, 수어사전

47 수어사전의구조 예테크리스토페르센 [ 그림 4. 주제검색에사용되는 3 단계체계 ] 3.2. 검색기능조절 검색을조율할수있도록검색기준과검색결과는좌우나열되어표시된다 ( 그림 5. 우측에검색결과확 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 49

48 1 부발표 인가능 ). 원치않는검색기준을개별적으로클릭하여제외하거나리셋 (reset) 버튼을클릭하여모두삭제 할수있다. [ 그림 5. 수형및주제결합검색에따른기준목록과검색결과일부 ] 모든연관결과가검색결과에포함되도록하기위해덴마크수어사전의검색기능은상당히포괄적이다. 수지적요소의경우, 수어의수위및수형은그중요도와무관하게모두검색기능에고려된다는것을의미한다. 같은이유로텍스트기준으로검색할때글로스, 대응어및용례번역모두가검색결과에반영되어표시된다. 검색결과는연관성순서대로나타난다. 수형및수위로검색하는경우, 수어에나타나는순서에따라점수가부여돼먼저나타나는수형또는수위가뒤따르는수형또는수위보다점수가높다. 텍스트검색결과는글로스, 대응덴마크어, 용례글로스, 용례의번역문에있는단어순으로가중치가부여된다. 즉, 글로스또는대응어가용례에사용된단어보다점수가높다. 이렇게계산된 연관성점수 에따라각검색결과 국립국어원국제학술대회 보이는언어의기록, 수어사전

49 수어사전의구조 예테크리스토페르센 는 0-3 연관성별 (relevance star) 로평가되어높은순서대로표시된다 ( 그림 5 우측검색결과참조 ). 연관성점수가동일한결과는수위-수형순으로표시된다. 또한사용자는연관성점수와무관하게수위및수형기준으로검색결과를정렬하여확인할수있다. 이방법은검색결과목록상단에있는탭을클릭하여정렬할수있다. 세번째탭은다시연관성기준의정렬방법으로검색한결과를표시한다. 4. 향후과제 덴마크수어사전은몇가지분야에서개선의여지가있다. 단일어구어사전은흔히원천언어를메타언어로사용한다. 이를수어사전에적용하기란거의불가능에가깝지만 DTS에관한추가정보를더함으로써사전이더욱풍부해질수있다 ( 예. 의미의정의 ). 또한굴절형에관한설명또는어원및품사와같이아직연구가많이진행되지않은분야등새로운유형의정보도추가할수있을것이다. 참고문헌 Det Danske Sprog- og Litteraturselskab ( ). Den Danske Ordbog [The Danish Dictionary]. Available at http//:ordnet.dk/ddo (last accessed 18th September 2018) Kristoffersen, J. H. & Troelsgård, T. (2012). The electronic lexicographical treatment of sign languages: The Danish Sign Language dictionary. In S. Granger & M. Paquot (eds.), Electronic Lexicography. Oxford: Oxford University Press, pp Kristoffersen, J. H. & Troelsgård, T. (eds) (2016). Ordbog over Dansk Tegnsprog 3. udgave [The Danish Sign Language Dictionary 3rd edition]. Center for tegnsprog, Copenhagen. Available at dk (last accessed 18th September 2018)Stokoe, W.C.; Casterline, D.C.; Croneberg, C.G. (1965). A Dictionary of American Sign Language on Linguistic Principles, Washington D.C.: Gallaudet College Press. Zwitserlood, I, Kristoffersen, J.H. & Thomas Troelsgård (2013). Issues in Sign Language Lexicography. In: Jackson, H. (ed.): The Bloomsbury Companion to Lexicography. London: Bloomsbury Publishing 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 51

50 Session 1 Structure of a Sign Language Dictionary Jette H. Kristoffersen, University College Capital, Denmark 0. Abstract Sign language lexicography is a young discipline in the world of sign language research. The main problem faced by the sign language lexicographer is that sign languages use the visual modality and have no written representation commonly used among the native signers. Sign language lexicography has over the last decades benefited tremendously from the electronic medium, and sign language dictionaries are no longer mere wordlists as in the past, they are real scientifically based dictionaries, with information types similar to those found in dictionaries of spoken languages. The visual modality is still a challenge on both how to perform searches and how to present the information on the signs. This paper describes the structure of the Danish Sign Language Dictionary (Kristoffersen & Troelsgård, 2016) and demonstrate a possible structure of a Sign Language Dictionary. 1. Introduction Until recent times most sign language, dictionaries were mere wordlists, with words from a spoken language, each accompanied by a picture or a verbal description of a sign. The source language was typically the surrounding spoken language and the sign language was the target language. Over the last decades, sign language lexicography has developed rapidly and a series of scientifically based sign language dictionaries has emerged, the first being A Dictionary of American Sign Language on Linguistic Principles, Stokoe et. al. (1965). Today Sign language dictionaries are no longer wordlist they are real dictionaries with the types of information similar to those found in dictionaries of spoken languages. And they typically have sign language as the source language. The sign language lexicography has benefited tremendously from technological advances. Thus, the decision of 국립국어원국제학술대회 보이는언어의기록, 수어사전

51 Structure of a Sign Language Dictionary Jette H. Kristoffersen how to represent signs in a dictionary used to be a major problem, but now digital video recordings of signs can easily be integrated into electronic dictionaries (cf. Zwitserlood et al, 2013). Still, certain obstacles remain for the sign language lexicographer. For example, some form of formal notation of the signs is needed in order to refer uniquely to the signs, to perform sign searches and to order the signs in search results etc. Also, a sign language dictionary with sign language as the source language requires more advanced search functionalities compared to a dictionary of a written language, for example in order to enable searches on manual features. The Danish Sign Language Dictionary (Kristoffersen & Troelsgård, 2016) is practically a bidirectional bilingual Danish Sign Language (DTS) - Danish dictionary, but its structure places it somewhere between bilingual and monolingual (Kristoffersen & Troelsgård, 2012). There is only one set of entries, with DTS signs as headwords. The entries are structured according to the semantics of the signs, and almost all information, including usage examples, concerns DTS, except for disambiguating explanations of homographic or strongly polysemous Danish equivalents. This approach was chosen in order to keep focus on DTS, as the need for information about Danish words can easily be met by existing Danish all-round monolingual dictionaries. As a result of this decision, the Danish Sign Language Dictionary could be described as monolingual dictionary, which instead of definitions has (searchable) equivalents in another language, Danish, which is also the general metalanguage of the dictionary. 2. Entry layout in the Danish Sign Language Dictionary The entries of the Danish Sign Language Dictionary have four sections (cf. figure 1): Section 1: Entry header In this section the sign headword is shown as a photo and a gloss. The first occurring location and the most prominent handshape of the sign are shown as icons. Lastly, there are buttons for opening separate windows with a printable sign overview and with a concordance view of all usage examples in the entire dictionary including the sign. Section 2: Video window By default the base form of the sign headword is shown. If there are phonological variants of the sign, a play button for each variant is shown in top of the video window. There are also control buttons for slow-motion and enlarged view. Other videos, e.g. of usage examples, are activated by clicking play buttons in different sections of the entry. Section 3: Meanings window In this section the meanings of the sign are described. The meaning description includes: Danish equivalents and/or a description of the sign's usage (for function signs), information about possible mouth movements, cross-references to synonyms etc., information about restricted use, and example 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 53

52 Session 1 sentences. Semantically opaque compounds with the sign are shown below the regular meanings. Danish equivalents that can be looked up in the Danish Dictionary 1), are supplied with a clickable link to this dictionary. Thus, users can learn more about the meaning and use of the Danish words. Section 4: Additional information This section holds cross-references to homonyms and to common frozen forms of the sign (for classifier entries). In addition to this, frequent co-occurrences with the sign are shown in this section. [Figure 1. The four sections of an entry in the Danish Sign Language Dictionary.] 1) The Danish Dictionary is an all-round, monolingual dictionary of contemporary Danish. The dictionary is freely accessible at: 국립국어원국제학술대회 보이는언어의기록, 수어사전

53 Structure of a Sign Language Dictionary Jette H. Kristoffersen 3. Search facilities in the Danish Sign Language Dictionary 3.1. Choosing search criteria The signs in the Danish Sign Language Dictionary can be looked up through search criteria from different categories: handshape, location, topic and text. Handshape There are 65 searchable handshapes. The handshapes to be used in the search criteria are selected in groups of related handshapes, but unwanted handshapes can be deselected individually. This enables the user to select single handshapes or custom groups of handshapes, independently of the default handshape grouping. Figure 2 shows a part of the handshape selection window, with the extended index finger group currently activated. [Figure 2. Partial view of the handshape selection window.] Particular handshapes for the active and the passive hand can be specified. Location Location is chosen from a page with 15 location icons, representing locations on or near the body. In addition to these it is possible to choose location through four group headers (= top row of the selection window). If a group header is chosen, all locations in the group are included in the search criteria. Additionally (in the rightmost column) two options are available for choosing locations in the space in front of the body: one for the neutral space in general, and one for locations in front of the face. The location selection window is shown in figure NIKL Conference Records of Visible Language: Sign Language Dictionary 55

54 Session 1 [Figure 3. The location selection window.] Text Text searches are performed on Danish equivalents as well as on sign glosses and example sentences (both transcriptions and translations). This enables users to find signs that are not themselves lemmas in the dictionary, but appear in example sentences. Phrase searches and wildcard searches are possible. Topic Topics can be chosen as search criteria from a list of 70 topics. The topics are ordered into nine groups, with subgroups in two levels. Topics which are group headers automatically include all subordinate topics in the search criteria. Figure 4 shows an overview of the topic hierarchy. Several criteria categories can be combined. Inside each category an or condition is used, while and conditions are used for category combinations. The currently chosen criteria are shown in the left side of the screen. Figure 4 shows the criteria list for a search for signs that are produced with the handshape extended index finger and concern the topic religion 국립국어원국제학술대회 보이는언어의기록, 수어사전

55 Structure of a Sign Language Dictionary Jette H. Kristoffersen [Figure 4. The three-level hierarchy used for topic search.] 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 57

56 Session Adjusting the search The search criteria and the result list are shown side by side in order to facilitate adjustments of the search (cf. figure 5, where the result list can be seen to the right). For the same purpose the search criteria can be deselected individually, by clicking them, or cleared altogether, by clicking a reset button. [Figure 5. Criteria list (and part of the result list) for a combined search on handshape and topic.] In order to secure that all possibly relevant matches are included in a search result, the search function in the Danish Sign Language Dictionary is quite broad. For manual features, that means that all occurring locations and handshapes in a sign, both prominent and non-prominent, are taken into account by the search function. Similarly, all text string matches, both among glosses, equivalents, and translations of usage examples are included in the search result. The matches of a search are ordered by relevance. For handshape and location matches this means that they are weighted according to their appearance in the sign, so that the first occurring handshape or location scores higher than the following. Text search matches are weighted in the following order: glosses, Danish equivalents, glosses 국립국어원국제학술대회 보이는언어의기록, 수어사전

57 Structure of a Sign Language Dictionary Jette H. Kristoffersen in usage examples, words in the translations of usage examples. Thus, a matching gloss or equivalent scores higher, than a match in a usage example. Based on the calculated relevance scores each match receives from none to three relevance stars, and the matches are ordered accordingly (see the result list shown to the right in figure 5). Matches with equal relevance score are ordered firstly by location, secondly by handshape. In addition to this, the user can choose between two sort orders: firstly by location, secondly by handshape, with no regard to the relevance score. These alternative sort orders are activated by clicking their according tabs in top of the result list. A third tab re-establishes the default, relevance-based sort order. 4. Future challenges The Danish Sign Language Dictionary could be developed in several areas. Spoken monolingual dictionaries often use the source language also as meta-language, for sign language dictionaries that would be an almost impossible challenge, but the dictionary would benefit from having more information in DTS, e.g. sense definitions. Also some additional information types could be added, e.g. descriptions of inflected forms, or lesser studied fields like etymology and part-of-speech. References Det Danske Sprog- og Litteraturselskab ( ). Den Danske Ordbog [The Danish Dictionary]. Available at http//:ordnet.dk/ddo (last accessed 18th September 2018) Kristoffersen, J. H. & Troelsgård, T. (2012). The electronic lexicographical treatment of sign languages: The Danish Sign Language dictionary. In S. Granger & M. Paquot (eds.), Electronic Lexicography. Oxford: Oxford University Press, pp Kristoffersen, J. H. & Troelsgård, T. (eds) (2016). Ordbog over Dansk Tegnsprog 3. udgave [The Danish Sign Language Dictionary 3rd edition]. Center for tegnsprog, Copenhagen. Available at dk (last accessed 18th September 2018)Stokoe, W.C.; Casterline, D.C.; Croneberg, C.G. (1965). A Dictionary of American Sign Language on Linguistic Principles, Washington D.C.: Gallaudet College Press. Zwitserlood, I, Kristoffersen, J.H. & Thomas Troelsgård (2013). Issues in Sign Language Lexicography. In: Jackson, H. (ed.): The Bloomsbury Companion to Lexicography. London: Bloomsbury Publishing 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 59

58

59 주제 3 수어사전편찬의과제 요한나메시스웨덴, 스톡홀름대학교 Johanna Mesch Stockholm University, Sweden

60 2 부발표 수어사전편찬의과제 요한나메시스웨덴, 스톡홀름대학교 본글은스웨덴수어사전 (Swedish Sign Language Dictionary) 이어떻게개발되었는지설명하고수어사전을편찬하는데오랜시간이걸리는이유를살펴볼것이다. 스웨덴의수어사전화작업은 1998년스톡홀름대학에서시작되었으며, 그결과 2001년첫온라인사전이나오게되었다. 스웨덴수어사전 (SSLD) 은 2008년에처음만들어진후개발이계속되고있다. 2003년말뭉치 (corpus) 구축이시작되고 스웨덴수어말뭉치 2009~2011 3개년프로젝트덕분에어휘주석으로말뭉치데이터가확대되자, 이를온라인스웨덴수어사전의지속적인개발과스웨덴수어사전의새로운수어및어휘변형을위한자원으로활용할수있는방안에대한논의가시작되었다. 키워드 : 수어사전, 사전학, 스웨덴수어 서론 수어사전은여러국가내수어의지위를높인다. 또한학습자와교수자, 가족및수어연구자들의요구에부응하기위해필요한도구이다 (McKee & Vale, 2017). 책형태의스웨덴수어사전은구화주의시대의영향아래 1916년처음출판되었다 (Österberg, 1916). 가장최근책으로출판된사전은스웨덴전국농인협회가 1997년 10년간의작업끝에편찬한것이다 (Svenskt teckenspråkslexikon, 1997). 스웨덴수어사전화작업은 1988년스톡홀름대학교언어학과수어부가처음시작했다. 이프로젝트의결과로 3,132개의수어표제어가담긴첫사전이 2001년스웨덴수어사전디지털버전이라는제목으로온라인에공개되었다. 이사전에는스웨덴전국농인협회가출판한책자사전에서따온수어동작영상파일이포함되었다. 하지만책자사전의기준을따르는것이수어사전을제한적인자원으로만든다는사실이곧명확해졌다. 목적이다른유연한사전을만들어낼필요가있었다. 그이후주요어휘데이터베이스인스웨덴수어사전 (Swedish Sign Language Dictionary) 이 2008 년에 국립국어원국제학술대회 보이는언어의기록, 수어사전

61 수어사전편찬의과제 요한나메시 만들어졌으며지금까지개발이계속되고있다 (Mesch, Wallin, & Björkstrand, 2012). 현재스웨덴수어사전에는 17,800개의수어표제어와 4,070개의예시문장이포함되어있다 (2018년 9월기준 ). 이프로젝트와관련된사전화실무단은스웨덴수어의단어를기록하고온라인사전을지속적으로업그레이드한다. 사전편찬자들은자신들이속한언어관련기관뿐아니라다른자원을기반으로작업하는경우가많다. 어떤언어라도언어자원은해당언어나그언어에포함된사용역 (register) 을대표적으로보여주는것이중요하다. 그렇기때문에언어자원또한언어의기록문서로사용될수있는것이다. 더중요한것이있다. 수어사전은다양한목적에적합한자료여야하고수어학습자, 통역사, 교수자, 농인커뮤니티및다운증후군이나기타언어장애를가진사람들등여러집단에게좋은자원이될수있어야한다. 사전화작업과작업자 사전팀은농인인 3명의상근직원으로구성되어있으며스웨덴수어단어를기록하고온라인사전을계속해서업데이트하는일을한다. 사전화작업은장애가있는학생들을위해스톡홀름대학교가대학차원에서국가예산을통해자금을지원한다. ( 내부 ) 자문단도함께작업해왔다. 사전의규모가계속커지고있기때문에외부자문단이필요한상황이다. 따라서언어협의회와함께자문단이구성될예정이다. 데이터베이스는파일메이커프로 (FileMaker Pro) 데이터베이스파일로만들어졌고구축이계속되고있다. 스웨덴에서는사인뱅크 (Signbank) 가사용되지않는데, 가까운시일내에사인뱅크를연구도구로사용할계획이있기는하다. 현재사전화실무단은파일메이커와 ELAN을기술적으로연결할수있는방법을찾는중이다. 문장구성방법과관련해서, 작업자 ( 외부자문단위원 ) 는스웨덴어문자없이사람이수어동작을보여주는정지화면만을보게된다. 이는서로의미가다른여러수어동작간연관성을구축한다. 그후수어작업자는자신만의사용예시를문장의형태로고안해야한다. 이러한유도도좋기는하지만, 앞으로는이방식에사전편찬자가말뭉치에서관련문장을검색하고검색된문장을사용에적절한형태로변경하여사전에추가하는방식이더해질것이다. 사전의설계 주요어휘데이터베이스인스웨덴수어사전은 2008년만들어졌으며그이후인터페이스개발이계속되고있다. 스웨덴수어사전웹사이트는 이다. 기술적지원의목적은스웨덴수어사전이좀더사용자친화적이고아이폰이나안드로이드앱등기타플랫폼에서도사용가능하도록개발하는것이었다. 스웨덴수어사전에서수어동작은수형, 수향, 수위, 수동, 스웨덴어번역 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 63

62 2 부발표 과수어표기로설명된다. 사전의각항목은단어 ( 또는구절 ) 영상, 스웨덴어번역, 음운정보및음운또 는의미상동등한수어동작과의사전내교차연결 ( 동음어와동의어등, 사진 1-3) 로대표된다 (Svenskt teckenspråkslexikon, 2018). 스웨덴수어말뭉치주석자를지원하기위해 ID- 글로스가추가된다. 수어동작은세가지방식 ( 사진 4) 이나주제별로 (46개의하위분류 ) 검색가능하다 ( 가족, 스웨덴도시와지역, 의료서비스나지화등 ). 또한각수어 ( 일반및특수용어 ) 동작에 ID 번호가있다. 하지만 ID 글로스만으로는수어를완벽하게설명할수없다. 각수어동작에대한설명과번역도있다. 스웨덴수어사전에는또한첫스웨덴수어책자사전 (Österberg, 1916) 에포함된 415개의표제어와함께구수어동작이추가되었다. 빈도는특정기한내에방문자가얼마나자주사전사이트를재방문했는지를알려준다. [ 사진 1.] 느린재생, 다시재생, 같은의미를가진다른동작, 이동작의다른의미옵션과함께동작을보여준다 [ 그림 2.] 문장의예를보여준다 국립국어원국제학술대회 보이는언어의기록, 수어사전

63 수어사전편찬의과제 요한나메시 [ 그림 3.] 정지된화면사진 ( 프린트용 ) 을스웨덴수어사전용표기, ID 글로스, 동작설명, 말뭉치주석용글로스, 사전용글로스와함께보여준다 [ 그림 4.] 수형, 수향, 수위및수동으로검색한다 스웨덴수어사전은 L2 교육과자기학습의표준화된자료로사용된다. 학습목적으로사전을사용하는것은학습자, 가족및교육적환경에있어중요하다. 농인커뮤니티역시페이스북그룹덕분에사전을유용하게이용하고있다. 학습자를위한퀴즈또한만들어졌다. 트위터봇인 Ett techken varje dag ( 하루에한동작 ) 이 2018년 1월에만들어졌으며학습자들사이에서반응이좋다.[ 이트위터봇은스웨덴수어사전에서무작위로선택한수어동작을하나씩올린다 ( 그림 5) NIKL Conference Records of Visible Language: Sign Language Dictionary 65

64 2 부발표 [ 그림 5.] 하루에한동작트위터봇 수어사전을지원하는언어자원 스웨덴어수어말뭉치 현재일반대중에게공개된스웨덴수어의주요언어자원은두가지로사전과말뭉치이다 (Mesch, 2012; Mesch & Wallin, 2012, 2015; Mesch, Wallin, & Björkstrand, 2012; Mesch, Wallin, Nilsson, & Bergman, 2012). 두언어자원모두스웨덴수어를위한기록문서역할을한다. 스웨덴수어의첫말뭉치는 2004년 ECHO 프로젝트와연계되어만들어졌다 (O. Crasborn et al., 2007). 이말뭉치는스웨덴수어말뭉치 (SSLC) 로 24시간분량의영상데이터대화, 서술, 프레젠테이션으로구성되어있으며 2009~2011년수어사용자 42명으로부터수집된자료로릭스방켄스유빌레움스방크 (Riksbankens Jubileumsbank) 가자금을지원했다 (Mesch, Wallin, Nilsson, et al., 2012; Wallin & Mesch, 2018). 말뭉치구축, 특히주석작업은매우오랜시간이걸린다. 수집된 24시간분량의데이터중약절반정도에주석이달렸다 (2018년 9월기준 ). 그외스웨덴수어의말뭉치에는촉각수어말뭉치 (Mesch, 2016) 와스웨덴수어학습자말뭉치 (Schönström & Mesch, 2017) 가있다. 스웨덴촉각수어말뭉치는 8명의수어사용자로부터 2007년과 2011년수집된대화와유도로구성되어있다. 수집된 3시간 30 분분량의자료중약 40%( 수어토큰 6,000개 ) 에주석이달렸다. 스웨덴수어말뭉치와다른말뭉치의주석작업은주석자 10명 ( 농인 8명비농인 2명 ) 이비상근으로참여하고있으며여전히진행중이다. 스톡홀름대학언어학과수어부가수어사전과스웨덴수어말뭉치를고안및유지하고있지만, 처음에는서로독립적으로만들어졌다. 이로인해서두자료간기능적불일치가생겨났다. 하지만두프로젝트는작년에통합되기시작했고현재는이미서로를보완하는수준이다. 예를들어, 스웨덴수어말뭉치의수어토큰에는스웨덴수어사전 ID 번호가주석으로달리고 (Riemer Kankkonen, Björkstrand, Mesch, & Börstell, 2018) 수어동작이스웨덴수어말뭉치에포함되면스웨덴수어사전에도추가된다. 또한스웨덴 국립국어원국제학술대회 보이는언어의기록, 수어사전

65 수어사전편찬의과제 요한나메시 수어사전온라인인터페이스는동의어검색결과를스웨덴수어말뭉치어휘빈도수인터페이스를통해계산된스웨덴수어말뭉치빈도수에따라분류한다 (Börstell & Östling, 2016). 스웨덴수어말뭉치데이터빈도수도구는경쟁적인이형수어간상대적빈도수를보여준다 ( 예 : 수어동작 TIO와 TIO(Y) 10 ). 스웨덴수어말뭉치가확대되면좋은언어자원이될것이다. 스웨덴수어말뭉치는현재약 113,000개의수어토큰을포함하는데, 말뭉치데이터수집단계에서구체적인어휘변형유도작업이없었기때문에 (Stamp et al., 2014), 많은경우동의어나수어동작의변형이말뭉치데이터사용만으로는조사가불가능하다. 113,000개의수어토큰은언어의변형을체계적으로조사하기위해사용되는구화말뭉치보다는훨씬작지만수어말뭉치로는꽤큰규모다. 크라우드소싱 언어자원개선 사전, 즉스웨덴수어사전을개발하고개선하기위해우리는농인커뮤니티를참여시켜새로운수어동작을수집하고동의어와의유사성및동의어들간차이에대한데이터를수집한다. 바로우리의언어자원, 즉스웨덴수어사전을개선하기위한크라우드소싱이다. 두가지방식으로작업이진행된다 : a) 스웨덴수어사용자와수어동작사용, 어휘의차이, 그리고수어동작구성에대해에스엔에스 (SNS) 상에서토론하고 b) 스웨덴수어사용자와여러장소에서직접적으로교류하며 ( 전국농인기관및농인커뮤니티사업및활동연례참석 ) 수어와수어에대한판단을수집한다 (Riemer Kankkonen et al., 2018). Teckenspråkslexikon ( 수어사전 ) 이라는이름의페이스북그룹은 2014년 10월에시작되었으며사전팀이관리한다. 2,642명의회원이있으며 (2018년 5월기준 ) 회원들은다양한방식으로교류하면서어떤수어동작을사용해야하고수어동작변형간의차이가무엇인지등에대해토론한다. 회원들은스웨덴어나스웨덴수어영상을이용해서질문을올리거나답글을단다. 지난 2년간 (2016년 1월부터 2018년 1월까지 ) 593개의글이올라왔고 5,817건의교류가있었다. 여기에는 X를수어로어떻게하나요?(53%), Y사람에대한수어동작이있나요?(20%) 및사용방법에대한일반적인질문 (10%) 과특정수어동작의어원 (-0.2%) 등의질문이포함된다. 결론 수어사전을개발하고유지하는것은많은시간을요하는자료집약적인작업이다. 비디오파일을업데이 트하거나고장및해킹을방지하면서기술적인발전을지속하는것이중요하다. 사전은또한양질의내용과 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 67

66 2 부발표 사용자친화적인자원을유지하기위해연구및기술적지원이필요하다. 수어사전을업데이트하고개발하는방법에는언어지식, 이전영상내용등의기타자원, TV 프로그램과영상메시지, 그리고국가별수어에대한데이터를에스엔에스 ( 페이스북그룹등 ) 상에서농인커뮤니티로부터직접크라우드소싱하기등이포함된다. 수어의말뭉치규모는아직작은수준이라말뭉치데이터만으로는수어형태의변형을조사할수없다. 사전을위한크라우드소싱은언어자원을개선하는유용하고빠른방법이다. 오프라인데이터수집은사전팀에게개인적으로농인커뮤니티를만나교류할수있는기회를제공하는데이는언어기록화작업의중요한요소다. 그뿐만아니라개인과의직접적교류로수어동작이형에대해좀더질적인접근이가능할수도있다. 각수어사용자와관련된메타데이터는말뭉치등기록화작업에추가될필요가있는특정개인이나집단을대상으로한다. 말뭉치를기반으로한사인뱅크는수어동작의분류정리화를고려하는연구자들에게필요한도구다. 예를들어사인뱅크에수어동작글로스가변경되거나새롭게추가될경우, 연결되어있는 ELAN 주석파일을통해자동으로업데이트되어주석자와연구자들의작업이쉬워진다. 초기사전이나현재어휘데이터베이스 ( 스웨덴수어어휘데이터베이스 ) 는말뭉치의분류정리화를위한말뭉치프로젝트와함께사용되면충분히좋은자원이될수있다. 구어 / 문어복합의경우 ( 의학, 기술또는직업관련주제에서매우흔함.) 수어사전에표제어를어떻게넣을지는아직도명확하지않으며, 빈도수가높은생산적수어 ( 묘사어나분류사 ) 도사전에추가되어야한다. 참고문헌 Börstell, C., & Östling, R. (2016). Visualizing lects in a sign language corpus : Mining lexical variation data in lects of Swedish Sign Language. In E. Efthimiou, S. E. Fotinea, T. Hanke, J. Hochgesang, J. Kristoffersen, & J. Mesch (Eds.), Workshop Proceedings: 7th Workshop on the Representation and Processing of Sign Languages: Corpus Mining (pp ). Paris: ELRA. Crasborn, O., Mesch, J., Waters, D., Nonhebel, A., Van Der Kooij, E., Woll, B., & Bergman, B. (2007). Sharing sign language data online: Experiences from the ECHO project. International Journal of Corpus Linguistics, 12(4). McKee, R., & Vale, M. (2017). Sign language lexicography. In P. Hanks & G.-M. de Schryver (Eds.), International Handbook of Modern Lexis and Lexicography (pp. 1 22). Berlin, Heidelberg: Springer Berlin Heidelberg. Mesch, J. (2012). Swedish Sign Language Corpus. Deaf Studies Digital Journal, 3. Retrieved from gallaudet.edu/index.php?issue=4&section_id=2&entry_id= 국립국어원국제학술대회 보이는언어의기록, 수어사전

67 수어사전편찬의과제 요한나메시 Mesch, J. (2016). Dataset. Tactile Swedish Sign Language Corpus. Department of Linguistics, Stockholm University. Mesch, J., & Wallin, L. (2012). From meaning to signs and back: Lexicography and the Swedish Sign Language Corpus. In O. Crasborn, E. Efthimiou, E. Fotinea, T. Hanke, J. Kristoffersen, & J. Mesch (Eds.), Proceedings of the 5th Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon [Language Resources and Evaluation Conference (LREC)] (pp ). Paris: European Language Resources Association (ELRA). Retrieved from proceedings/lrec2012/workshops/24.proceedings_signlanguage.pdf Mesch, J., & Wallin, L. (2015). Gloss annotations in the Swedish Sign Language Corpus. International Journal of Corpus Linguistics, 20(1), Mesch, J., Wallin, L., & Björkstrand, T. (2012). Sign language resources in Sweden: Dictionary and corpus. In O. Crasborn, E. Efthimiou, E. Fotinea, T. Hanke, J. Kristoffersen, & J. Mesch (Eds.), Proceedings of the 5th Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon [Language Resources and Evaluation Conference (LREC)] (pp ). Paris: European Language Resources Association (ELRA). Retrieved from workshops/24.proceedings_signlanguage.pdf Mesch, J., Wallin, L., Nilsson, A.-L., & Bergman, B. (2012). Dataset. Swedish Sign Language Corpus project (version 1). Sign Language Section, Department of Linguistics, Stockholm University. Retrieved from Österberg, O. (1916). Teckenspråket. Uppsala: P. Alfr. Persons förlag. Riemer Kankkonen, N., Björkstrand, T., Mesch, J., & Börstell, C. (2018). Crowdsourcing for the Swedish Sign Language Dictionary. In M. Bono, E. Efthimiou, S.-E. Fotinea, T. Hanke, J. Hochgesang, J. Kristoffersen, Y. Osugi (Eds.), Proceedings of the 8th Workshop on the Representation and Processing of Sign Languages: Involving the Language Community [Language Resources and Evaluation Conference (LREC)] (pp ). Paris: European Language Resources Association (ELRA). Schönström, K., & Mesch, J. (2017). Dataset. The project From speech to sign learning Swedish Sign Language as a second language. Department of Linguistics, Stockholm University. Svenskt teckenspråkslexikon. (1997). Leksand: Sveriges Dövas Riksförbund. Wallin, L., & Mesch, J. (2018). Annoteringskonventioner för teckenspråkstexter. Version 6, januari [Annotation guidelines for sign language texts] NIKL Conference Records of Visible Language: Sign Language Dictionary 69

68 Session 2 Challenges of creating a Sign Dictionary Johanna Mesch, Stockholm University, Sweden The article will describe how the Swedish Sign Language Dictionary has developed, and why it takes a long time to establish such a dictionary. A lexicographic work of Swedish Sign Language was initiated in 1988 at Stockholm University, and it resulted, in 2001, the first dictionary online. The Swedish Sign Language Dictionary was created in 2008 and has been in development since. When the direction of the corpus construction started in 2003, and when the corpus data, thanks to the threeyears project of the Swedish Sign Language Corpus , expanded with gloss annotations, a discussion has arisen about how the Online Swedish Sign Language Dictionary should continue to be in its development and in which direction, and how to use the SSL Corpus as a source of input for new signs and lexical variation in the SSL Dictionary. Keywords: Sign language dictionary, lexicography, Swedish Sign Language Introduction A sign language dictionary elevates the status for existence of sign language(s) in many countries. It is also a necessary tool for serving the needs of learners, teachers, families and researchers of signed languages (McKee & Vale, 2017). The first Swedish Sign Language Dictionary in a book format was published in 1916 in the shadow of the oralism period (Österberg, 1916). The latest dictionary in printed book format was published by the Swedish National Deaf Association in 1997 after ten years work (Svenskt teckenspråkslexikon, 1997). The lexicographic work of Swedish Sign Language was initiated in 1988 by the Sign Language Section of the Department of Linguistics at Stockholm University. In 2001, the first dictionary resulting from the project went online, entitled the Digital version of Swedish Sign Language Dictionary and included 3,132 sign entries. This dictionary includes video files for signs taken from a printed dictionary of the Swedish National Deaf Association. It soon became clear that following criteria of the printed dictionary was making it a limited resource. It needed to be a flexible dictionary for different purposes 국립국어원국제학술대회 보이는언어의기록, 수어사전

69 Challenges of creating a Sign Dictionary Johanna Mesch A follow-up dictionary, the main lexical database, the Swedish Sign Language Dictionary was created in 2008 and has been in development since then (Mesch, Wallin, & Björkstrand, 2012). Today, it has approximately 17,800 sign entries and 4,070 sentence examples (September 2018). The lexicographic work group associated with this project documents the vocabulary of Swedish Sign Language and continuously updates the web-based dictionary. The work of the lexicographers is often based on their intuitions concerning language, but also on different sources. An important aspect of any language resource is that it is representative of the language or register that it covers. Because of this, the resources also serve as language documentation. This is perhaps even more important. A sign language dictionary should be a good resource for different purposes and different groups such as sign language learners, interpreters, sign language teachers, the deaf community, and other groups such as people with Downs syndrome and other language disorders. The lexicographic work and actors The dictionary team consists of three full-time employees, all Deaf, who document SSL vocabulary and continuously update the web-based dictionary. The work is financed by Stockholm University through national financing to support for disabled students at university level. There has been the (internal) advisory group. Because the dictionary has been growing bigger, an external advisory group is needed. With The Language Council, an advisory group will be established. The database has been created as FileMaker Pro database files and its construction is ongoing. Signbank is not used in Sweden, although it is planned to have a Signbank as research tool in near future, and the group is trying to find a possible technical connection between FileMaker and ELAN. With respect to the method for creating sentences, the actors (members of the external advisory group) are given still photos only, in which a person is presenting a sign, without any written Swedish words. This creates associations with many signs of different meanings. Then the sign actors have to devise their own usage examples in the form of sentences. This elicitation may be good, but in the future it will be combined with the possibility for lexicographers to look up related sentences in the corpus and change them to a suitable form for use in the dictionary. Dictionary design The main lexical database, the SSL Dictionary was created in 2008 and its interface has been in development 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 71

70 Session 2 since. The website for Swedish Sign Language Dictionary is Technical support aimed to develop the dictionary to become more user-friendly and usable on different platforms such in as apps for iphone and Android. In the SSL Dictionary, a sign is described in terms of its handshape, orientation, location, movement, Swedish translations and sign transcription. Each dictionary entry is represented by a video of the sign (or phrase), a Swedish translation, phonological information, and internal cross-links to phonologically or semantically equivalent signs i.e. homophones and synonyms, see Figures 1-3 (Svenskt teckenspråkslexikon, 2018). ID-glosses are added for supporting SSL corpus annotators. A sign can be searched for using these ways (see Figure 4) or by subject (46 with subcategories), e.g. family, Swedish towns and provinces, healthcare or fingerspelled signs. There are also ID numbers for signs (in general and special terms). However, ID glosses alone cannot describe sign language perfectly. There are also descriptions and translations for each sign. The old signs with 415 sign entries from the first printed dictionary in SSL (Österberg, 1916) are also added to this lexical database. osterberg-1916 Frequency refers to how often visitors return to the dictionary site within a time frame. [Figure 1.] Showing a sign with options for slow motion, play again, other signs with same meaning, this sign has also other meaning [Figure 2.] Showing some sentence examples 국립국어원국제학술대회 보이는언어의기록, 수어사전

71 Challenges of creating a Sign Dictionary Johanna Mesch [Figure 3.] Showing still photos (for printing) with sign transcription for SSL, ID gloss, sign description, gloss for corpus annotation, gloss for dictionary [Figure 4.] Searching via handshape, orientation, location and/or movement The SSL Dictionary is used as standardized learning resource for L2 teaching and self-learning. The use of the dictionary for learning purposes is important for learners, families and educational settings. The Deaf community has also found it valuable thanks to a Facebook group. A Quiz has also been created for learners, se/app/#/start. A twitter bot Ett tecken varje dag (One sign every day) was created in January 2018, and it has been well-liked by learners [ It publishes a randomly selected sign from the SSL Dictionary, see Figure 5. [Figure 5.] Twitter bot One sign every day 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 73

72 Session 2 Language resources supporting for the sign language dictionary Swedish Sign Language Corpus There are currently two main language resources of SSL publicly available, a dictionary and a corpus (Mesch, 2012; Mesch & Wallin, 2012, 2015; Mesch, Wallin, & Björkstrand, 2012; Mesch, Wallin, Nilsson, & Bergman, 2012), both of which also serve as a form of language documentation for SSL. The first corpus of Swedish Sign Language was created in 2004 in connection to the ECHO project (O. Crasborn et al., 2007). The resource is the Swedish Sign Language Corpus (SSLC), which consists of 24 hours of video data conversations, narratives, and presentations) from 42 different signers collected in the years , financed by Riksbankens Jubileumsbank (Mesch, Wallin, Nilsson, et al., 2012; Wallin & Mesch, 2018). Corpus work, especially annotation work, is very time-consuming. Approximately half of the collected 24 hours of data have been annotated (September 2018). Other corpora in SSL are the Tactile Sign Language Corpus (Mesch, 2016) and the Learner Corpus in SSL (Schönström & Mesch, 2017). The Tactile SSL Corpus consists of dialogues and elicitation from 8 signers collected in 2007 and 201. Approximately 40 % (6,000 sign tokens) of the collected four hours and thirty minutes hours have been annotated. Annotation work of the SSL corpus and other corpora is still ongoing with ten (eight deaf and two hearing) annotators working on it part-time. Although both the SSL Dictionary and the SSL Corpus are designed and maintained by the Sign Language Section at the Department of Linguistics, Stockholm University, they were initially set up independently from each other. This has led to some discrepancies between the functions of the two resources. However, the two projects have started to converge in the last year, and now they are already supporting each other. For example, sign tokens in the SSLC are annotated with the SSLD ID number (Riemer Kankkonen, Björkstrand, Mesch, & Börstell, 2018), signs are added to the SSLD as they appear in the SSLC, and the SSLD online interface sorts synonyms search hits according to SSLC frequencies calculated through the SSLC lexical frequency interface (Börstell & Östling, 2016). This frequency tool for the SSLC data shows the relative frequencies of two competing sign variants, e.g. the signs TIO and TIO(Y) ten. It will be a good resource when the SSLC has been expanding. The SSLC contains currently around 113,000 sign tokens, and, as there was no explicit lexical variation elicitation task during the collection of the corpus data (Stamp et al., 2014), many sign synonyms or form variations cannot yet be investigated solely with the use of corpus data. A set of 113,000 sign tokens is fairly large for a sign language corpus, although it remains much small when compared to spoken language corpora used to investigate variation systematically 국립국어원국제학술대회 보이는언어의기록, 수어사전

73 Challenges of creating a Sign Dictionary Johanna Mesch Crowdsourcing - Improving language resources To develop and improve the dictionary, specifically the SSLD, we involve the Deaf community to collect new signs and to gather data on familiarity with and variation within sign synonyms that is, crowdsourcing to improve our language resources, specifically the SSLD. This works in two ways: a) discussing sign usage, lexical variation, and sign formation with SSL signers on social media; and b) interacting with SSL signers directly at various venues (the annual participation in Deaf events of national Deaf organizations and Deaf community business and activities) collecting signs and judgments about signs (Riemer Kankkonen et al., 2018). A Facebook group entitled Teckenspråkslexikon ( sign language dictionary ) was started in October 2014 and is administrated by the dictionary team. There are 2,642 members (in May 2018), who interact in different ways, discussing which signs are to be used, or the difference between sign variants. They can write questions or comments in Swedish or in SSL as video comments. In the last two years (i.e., January 2016 to January 2018), there have been 593 posts and 5,817 interactions. This includes questions about signs, such as How do you sign X? (53%), Is there a sign for person Y? (20%), general questions about the usage (10%) and etymology (<0,2%) of specific signs. Conclusions Developing and maintaining the sign language dictionary is time-consuming and resource-intensive. It is necessary to keep up with technical development, for example by updating video files, and protecting against crashes and hackers. The dictionary also needs research and technical support to maintain a good quality and user-friendly resource. The methods for updating and developing the sign language dictionary include language knowledge, other sources such as old video stories, TV programmes and video messages, and crowdsourcing data about the national sign language(s) directly from the Deaf community using social media, e.g. Facebook group. The corpus size for signed languages is still small, so form variations cannot yet be investigated solely with the use of corpus data. Crowdsourcing for the dictionary is a useful and rapid method for enhancing the language resource. Data collected through offline methods may provide a more qualitative approach to sign variants, interacting with individuals directly aside from giving the dictionary team an opportunity to personally meet and interact with the Deaf community, which is an important aspect of any language documentation work. Metadata about each signer target specific individuals or groups that are needed to add to the documentation work, for example, from the corpus. The corpus-based Signbank is a necessary tool for researchers considering sign lemmatization. When changes or additions of new sign glosses, for example, have been made to the Signbank, they are automatically updated through linked ELAN annotation files, which makes annotators and researchers work easier. Early dictionaries or current lexical databases, e.g. the SSL lexical database, are good enough sources, if they are used together with the corpus projects for lemmatization of a corpus. How to put a 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 75

74 Session 2 headword on a sign dictionary is still not clear when thinking about compounds from spoken/written language (very common in subject areas such as medicine, technical or vocational words), and the high frequency part of productive signs (depicting or classifier signs) should be added to the dictionary. References Börstell, C., & Östling, R. (2016). Visualizing lects in a sign language corpus : Mining lexical variation data in lects of Swedish Sign Language. In E. Efthimiou, S. E. Fotinea, T. Hanke, J. Hochgesang, J. Kristoffersen, & J. Mesch (Eds.), Workshop Proceedings: 7th Workshop on the Representation and Processing of Sign Languages: Corpus Mining (pp ). Paris: ELRA. Crasborn, O., Mesch, J., Waters, D., Nonhebel, A., Van Der Kooij, E., Woll, B., & Bergman, B. (2007). Sharing sign language data online: Experiences from the ECHO project. International Journal of Corpus Linguistics, 12(4). McKee, R., & Vale, M. (2017). Sign language lexicography. In P. Hanks & G.-M. de Schryver (Eds.), International Handbook of Modern Lexis and Lexicography (pp. 1 22). Berlin, Heidelberg: Springer Berlin Heidelberg. Mesch, J. (2012). Swedish Sign Language Corpus. Deaf Studies Digital Journal, 3. Retrieved from gallaudet.edu/index.php?issue=4&section_id=2&entry_id=128 Mesch, J. (2016). Dataset. Tactile Swedish Sign Language Corpus. Department of Linguistics, Stockholm University. Mesch, J., & Wallin, L. (2012). From meaning to signs and back: Lexicography and the Swedish Sign Language Corpus. In O. Crasborn, E. Efthimiou, E. Fotinea, T. Hanke, J. Kristoffersen, & J. Mesch (Eds.), Proceedings of the 5th Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon [Language Resources and Evaluation Conference (LREC)] (pp ). Paris: European Language Resources Association (ELRA). Retrieved from proceedings/lrec2012/workshops/24.proceedings_signlanguage.pdf Mesch, J., & Wallin, L. (2015). Gloss annotations in the Swedish Sign Language Corpus. International Journal of Corpus Linguistics, 20(1), Mesch, J., Wallin, L., & Björkstrand, T. (2012). Sign language resources in Sweden: Dictionary and corpus. In O. Crasborn, E. Efthimiou, E. Fotinea, T. Hanke, J. Kristoffersen, & J. Mesch (Eds.), Proceedings of the 5th Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon [Language Resources and Evaluation Conference (LREC)] (pp ). Paris: European Language Resources Association (ELRA). Retrieved from 국립국어원국제학술대회 보이는언어의기록, 수어사전

75 Challenges of creating a Sign Dictionary Johanna Mesch workshops/24.proceedings_signlanguage.pdf Mesch, J., Wallin, L., Nilsson, A.-L., & Bergman, B. (2012). Dataset. Swedish Sign Language Corpus project (version 1). Sign Language Section, Department of Linguistics, Stockholm University. Retrieved from Österberg, O. (1916). Teckenspråket. Uppsala: P. Alfr. Persons förlag. Riemer Kankkonen, N., Björkstrand, T., Mesch, J., & Börstell, C. (2018). Crowdsourcing for the Swedish Sign Language Dictionary. In M. Bono, E. Efthimiou, S.-E. Fotinea, T. Hanke, J. Hochgesang, J. Kristoffersen, Y. Osugi (Eds.), Proceedings of the 8th Workshop on the Representation and Processing of Sign Languages: Involving the Language Community [Language Resources and Evaluation Conference (LREC)] (pp ). Paris: European Language Resources Association (ELRA). Schönström, K., & Mesch, J. (2017). Dataset. The project From speech to sign learning Swedish Sign Language as a second language. Department of Linguistics, Stockholm University. Svenskt teckenspråkslexikon. (1997). Leksand: Sveriges Dövas Riksförbund. Wallin, L., & Mesch, J. (2018). Annoteringskonventioner för teckenspråkstexter. Version 6, januari [Annotation guidelines for sign language texts] NIKL Conference Records of Visible Language: Sign Language Dictionary 77

76

77 주제 4 대한민국수어사전의현황과발전방향 최혜원, 이현화한국, 국립국어원 Hyewon Choi, Hyunhwa Lee National Institute of Korean Language, Korea

78 2 부발표 대한민국수어사전의현황과발전방향 최혜원 이현화한국, 국립국어원 1. 머리말 한국수어언어법 (2016) 이시행된지 2년여의시간이흘렀다. 한국사회에서한국수어가언어로인정되고한국수어의사용권리를보장받기위한농인의오랜염원을담은 한국수어언어법 은한국수어의발전을위한다양한제도와사업활성화의근거가되었다. 2016년 한국수어언어법 이제정된이래대한민국은한국수어의언어로서의발전을위해다양한연구와사업을시행하고있다. 5년마다한국수어발전기본계획을체계적으로수립하고, 한국수어사용환경을개선하고자한국수어의사용실태조사, 한국수어말뭉치구축, 문법에관한기초연구등을본격적으로가동했다. 또한한국수어의교육및보급을위해한국수어교원자격제도를시행하면서대상별수어교재를개발하고한국수어문화학교프로그램을운영하고있다. 이와함께기존의 한국수어사전 을정비하는과정에서한언어의총체를담은한국수어사전으로의도약을위한발전적인방향을모색하고있다. 수어사전의편찬은수어의독립적지위를알리기위한가장기초적인작업이다. 음성언어의종속적인언어, 보조적인의사소통수단이라는통념을딛고언어로서의수어를기술한수어사전은농인들이독립된언어공동체를이루고있다는증거가된다. 실제로도여러나라에서언어사전으로서의수어사전을편찬하여농인공동체의언어적, 문화적지위향상의계기를마련하였다. 이때, 수어사전의형태는그간의음성언어중심의관점으로음성언어에대응하는수어를보여주는단순한대역어형태의사전을뛰어넘어야함은물론이다. 조선이일본에주권을빼앗긴일제강점기시대, 국어학자들을중심으로한글로된최초의조선어사전을편찬하려하였으나결국세상에먼저나오게된것은조선총독부가편찬한 조선어사전 (1920) 이었다. 사전명칭과는달리이사전은조선어를대상화한일본어중심의사전일뿐이었다. 1) 조선어가독립국가의언어가아닌상태에서편찬된사전은일본어표제어와이에대응하는한글표기, 그리고일본어뜻풀이 1) 이후문세영이라는국어학자의치열한노력으로 조선어사전 (1938) 이라는한글로표기된한국어사전이만들어져우리민족에게큰희망을주었다 국립국어원국제학술대회 보이는언어의기록, 수어사전

79 대한민국수어사전의현황과발전방향 최혜원, 이현화 로구성되어, 조선인의일본어교육, 일본인의조선어교육 이라는식민지지배정책에활용되었다 ( 최경봉, 2014). 위의사례와마찬가지로대한민국에서그간만들어진한국수어사전들도진정한의미의언어사전으로서틀을갖추지못하였다. 한국수어에대한이해가깊지않은청인교육자들이한국어텍스트로된교과목을가르치기위한수단으로만든대역어형태의수어사전은그태생적한계를극복하지못한채현재까지답습되고있는실정이다. 농교육을목적으로한사전의출발점은서구 2) 에서도동일하지만 90년대전후유럽을중심으로언어학적접근을통해수어의자연언어로서의체계를담은진일보한수어사전을편찬하고있다. 우리나라는 한국수어언어법 을통해한국수어를언어로선언하였지만, 한국사회에서명실공히언어로서인정되기위해서는선언을뛰어넘어한국수어의언어로서의체계를밝히보이는사전이필요하다. 수어의무한한가능성을보이고한국농사회의정체성과자부심을뒷받침할수있는사전편찬을위해이번학술행사는매우의미깊은역할을할것이라기대한다. 이글에서는그간한국에서편찬된수어사전들의현황과문제점을살펴보고향후농인이실제사용하는한국수어자료를충실히담은새로운한국수어사전편찬을위해국립국어원에서수립한중장기계획을밝히도록하겠다. 2. 한국수어사전의편찬현황 수어사전이없던시기에수어어휘목록은주로수어보급을위해만든수어교재를통해볼수있었다. 1982년에야한국에서는처음으로수어사전이편찬되었는데 표준수어사전 이라는사전의명칭에서도확인할수있듯이이사전은 수어표준화 3) 에기여하기위한목적에서출발하였다 ( 이준우, 남기현, 조준모, 2018). 이목적에따라농학교교사들을중심으로구성된 한국표준수어사전편찬위원회 는일상생활수어어휘, 초 중등국어교과서의국어어휘에대응하는수어어휘를사전에담아냈다. 아래의표에서보듯이한국에서는그간 표준수어사전 (1982) 을시작으로총 6종의수어사전이편찬되었다. 4) 연번사전 / 편찬기관 ( 편찬자 ) 출판연도표제어미시구조 1 표준수어사전 표준수어사전편찬위원회 ,494 개 2) 1989 년호주수어사전에서도영어의문법형태 ( 과거형분사인 -ed ) 에대응시킨수어표현의흔적이남아있다. 3) 여기서말하는 표준화 는일반적인의미의언어표준화와는다르다. 광범위한수어자료의수집과정이없었기때문에지역적변이형등여러형태로나타나는어형가운데가장일반적이면서널리통용되는어휘를대표형으로선정하는방식은아니었다. 한국어대응어가없는것은새로만들어내고, 동형이의어는변별을위해보조사인을첨가하는등음성언어의관점에서수어를 규격화 하는작업이대거포함되었다. 4) 어원사전, 전문용어사전등은목록에서제외하였다 NIKL Conference Records of Visible Language: Sign Language Dictionary 81

80 2 부발표 연번사전 / 편찬기관 ( 편찬자 ) 출판연도표제어미시구조 2 수어사전 단국대학교출판부 ,966 개 3 표준수어 김승국 ,966 개 4 한국수어사전 국립국어원 한국농아인협회 ,800 여개 5 한국수어사전 김삼찬외 ,818 개 6 한국수어사전 국립국어원 ( 온라인 ) ,793 개 < 표 1> 국내에서편찬된한국수어사전 위에서보듯이한글자모가나다순으로배열된 6개사전은한국어표제어와이에대응하는수어를담은대역어사전의형태이다. 대역어조차도실제사용되는수어만담은것이아니라한국어어휘를표현하는데에필요하다는이유로수어표현에는없는것을새로만들어올린경우가적지않았다. 이러한결과로과도한수어표제어가산출되었고, 그중에는한국어어미와조사, 합성어에대응하는다수의인위적인수어가포함되었다. 5) 일반적으로대역어사전은 A 언어표제어의여러의미를 B 언어의어휘나대체표현으로다양하게제시한다. 예를들어표제어가한국어 상하다 일경우 상하다 의여러의미와이에대응하는한국수어대역어정보를제시해야한다. 5) 1998 년네덜란드정부는네덜란드수어의법적지위를인정하기위한선결과제로네덜란드기본어휘표준화 (STABOL, Standardisation of the Basic Lexicon of NGT) 를내세웠다. 농사회에수어연구자들의거센반대에부딪히기도한이프로젝트에서표준화된어휘수는약 14,000 개를기록한다. 이중 33% 가의료나법률, 학교교과서용어등새로운어휘이다.(Roland Pfau, Markus Steinbach & Bencie Woll, 2012) 국립국어원국제학술대회 보이는언어의기록, 수어사전

81 대한민국수어사전의현황과발전방향 최혜원, 이현화 한국어표제어 의미 / 한국수어대역어 다치다 야위다 망가지다 상하다 상하다 < 표 2> 한국어 상하다 에대응되는여러개의수어표현 그러나 한국수어사전 은한국어표제어의대응수어를하나로국한하여한국어표제어의뜻을한국수어단어하나로제시하거나 (1, 5), 한국어표제어의사전뜻풀이를거의단일하게압축하여보이고있다 (4, 5, 6). 이는사전이용자에게한국어에하나의수어가대응되고, 한국어와한국수어의의미갈래가일치한다는한국수어에대한잘못된이해를불러일으킬수있다. 상하다 ' 상하다 ' < 그림 1> 한국어표제어 상하다 와글로스가일치하는한국수어의제시 대역어사전의틀에서기술되면서사전에서다루는수어정보는수어의형태적특징에한정될뿐그외수어의언어적특징은기술되지못하였다. 이마저도한국어에한국수어를무리하게연결시키다보니한국수어가왜곡되는현상을초래하였다. 예를들어 5의 목적지나일정한수준에다다름. 의뜻을지니는 도달 에대응하는수어의동작을보자. 수향이수직방향으로위로향해있으나 목적지 는반드시높은곳에있지않을수있다. 목적지 의위치특성에따라 도달 을표현하기위해수어화자는수향의변화를다양하게주는데사전에서는이를반영하지못하고있다. 또한구체화된표현이많은한국수어의특성상한국어표제어의뜻풀이와수어가대응되더라도한국어용례와는불일치하는경우가발견된다. 한국어 가장자리 ( 둘레나끝에해당되는부분 ) 에대응되는수어는한국어의일반적인뜻에서크게벗어나지않지만 눈가장자리가빨갛게되었다. 와같은용례를표현할때는부적절한표현이된다 NIKL Conference Records of Visible Language: Sign Language Dictionary 83

82 2 부발표 < 그림 2> 표제어 가장자리 에대응되는수어표현 한국수어사전이처음만들어진시기나사전의총수, 표제어수는다른어느나라에비교해봐도크게뒤지지않는다. 6) 그러나사전콘텐츠의충실성은이처럼별다른개선없이답보상태에머물고있다. 7) 이는 2005년 한국수어사전 까지진행된사전편찬작업이 수어표준화 의기본취지에서크게벗어나지못하였기때문이다. 이러한한계는 2005년 한국수어사전 의형태를그대로반영한, 현재까지서비스되고있는온라인형태의사전인국립국어원의 한국수어사전 8) (2015) 에서도마찬가지이다. 3. 한국수어사전의발전방향 앞서살펴본바대로한국수어표준화사업의결과물인 한국수어사전 은한국수어자료를모아연구하고수어연구방법론에대한환기를시켰다는점에서의의를찾을수있다. 그러나구축당시의시대적한계로인하여이사전은사용자들에게지속적인비판을받아왔다. 이에국립국어원은그간의비판을겸허히수용하고기존사전의한계를극복한새로운한국수어사전의편찬을계획하고있다. 효과적인사전설계는사용자의목적과능력의현실적인이해에서부터시작된다고한 Atkins & Rundell(2008) 의주장처럼국립국어원은 한국수어사전 개선방향에대한올바른설계를위하여여러사업을통해농사회의의견을수렴하였다. 9) 2014년에구성된 한국수어연구자문위원회 에서는전국의한국수어사용자 553명을대상으로 수어전문용어제정및표준화수요조사 를진행하였다. 이조사에따르면참여자들은수어표준화가추진된전문분야수어를사용하지않는이유로 실제사용되는수어와다름 (28.1%) 에제일많이응답하였다. 수형기반한국수어사전구축 ( 국립국어원, 2016) 사업에서는농인 15명을대상으로초점집단인터뷰 6) 초기의뉴질랜드수어사전 (The Dictionary of New Zealand Sign Language, 1986) 은 1,200 개, 호주수어사전 (Auslan Dictionary, 1989) 은 3,500 개, 영국수어사전 (Dictionary of british sign language, 1992) 은 1,800 개의표제어가수록되어있다. 7) 1960 년대스토키를중심으로한갤러뎃대학연구자들의미국수어에대한언어학적연구방법론은 1970 년대중반영국과유럽전역의연구자들에게영향을주었고, 1980 년대이후수어는전세계적인연구분야의중심에섰다. 8) 한국수어사전 (2015) 에는 한국수어사전 (2005) 의어휘목록과함께이후동일한사전편찬방향하에연차별작업으로추가된 6 천여개의어휘가포함되어있다. 9) 이글에서는조사결과중에서기술에필요한의견일부를선택적으로제시하였다 국립국어원국제학술대회 보이는언어의기록, 수어사전

83 대한민국수어사전의현황과발전방향 최혜원, 이현화 를실시하였는데그결과기존사전이 농사회에서자주사용되는수어가부재 한상태이며 한국어대응수어표현이중심이되다보니농인들로부터외면 받고있다고하였다. 그리고이사전이 수어통역사들의교육용으로활용 되고 청인들의수어학습용으로활용 되고있지만 농인조차이해하기어려운수어사전 이라 농인과청인수어통역사간소통의문제가발생 한다고하였다. 농인들은현사전을 필요할때이용하지만큰도움을주지않는수어사전 이며 농인이이해하기어려운표현의한국어정보 로한국어를이해하기위해사전을이용하는것에도제한이있는것으로평가하고있었다. 이어 2017년진행된 한국수어사전수어뜻풀이와용례구축및정비 ( 국립국어원 ) 사업을통해수어학계와농사회의수어관련전문가 30인을대상으로설문조사를실시하였다. 조사결과 한국수어사전 의단점으로거시구조측면에서는낮은접근성과인위적으로만들어진수어가등재되어있다는점이주로지적되었고, 미시구조측면에서는정보항목간의낮은일치도와수어를활용하기위해반드시필요한용례, 문법정보등의부재가지적되었다. 더불어 2016년과 2017년에각각진행된두조사에서는 한국수어사전 의발전방향에대해제언을받기도하였는데주요내용은다음과같다. ㅇ수형기반한국수어사전구축 ( 국립국어원, 2016) - 농인사용자중심의새로운사전제작 - 한국어정보를농인의관용적수어표현으로 10) 쉽게풀이해서설명 - 수어사전의목적과이용대상에대한명확한설정필요 - 접근성향상을위한방안ㅇ한국수어사전수어뜻풀이와용례구축및정비 ( 국립국어원, 2017) - 농인과청인이모두사용할수있는사전 - 현사전에서인위적인표제어의삭제및정비 - 농인이실제로사용하는수어를수록하고수어관련정보를충실히담은사전 - 수어의다양한변이형을담은사전 - 사전의접근성을높이기위한표제어분류및검색방법 ( 움직임인식등 ) 개발 이상의결과로볼때현재 한국수어사전 의이용자는한국수어학습자와한국어학습자로나뉘며각그룹은다른목적으로사전을이용하고있다. 또한이사전이한국수어표현을인위적으로만들어사전에등재하였으나이것이농사회에받아들여지지못해사전으로서의역할을하기에부족한상황임을알수있다. 따라서이러한의견을바탕으로 한국수어사전 을정비하여사전의일치도와완성도를높이는한편이용자들이목적에따라사전을이용할수있도록새로운사전을편찬할필요가있다. 10) Cokely, D. & Baker-Shenk, K, C. (1980) 는미국수어 (ASL) 의설명에서 관용어 (idioms) 로표현된것들의대부분은사실관용어가아니고관용어정의에맞지않는다고하며단순히몇개의다른용법을가진수어들이라고하였다. 이정옥, 이준우 (2005) 도한국의이와같은현상을정리하며 자연수어 혹은 농식수어 라는명칭으로관용표현을받아들이고있음을지적하고있다. 따라서본고에서인용한조사의결과도같은맥락으로해석할수있다 NIKL Conference Records of Visible Language: Sign Language Dictionary 85

84 2 부발표 Hartmann(1983) 은외국어를학습하고교육하는보조도구로서이언어사전이유용하게사용될수있다고하였다. 또한한국수어기술사전편찬의필요성에대해논의한이현화 (2017) 는한국수어사전이인위적으로가공된자료가아닌한국농인의자연스러운발화를담고있는실제자료에근거하여표제어를추출하고이를바탕으로편찬되어야한다고하였다. 이러한논의들을바탕으로볼때한국수어사전의편찬방향은다음두가지로귀결된다. 첫째, 사전은사용목적을분명히하여 한국수어-한국어사전, 한국어-한국수어사전 으로이원화구축되어야한다. 둘째, 변이가많은수어의특성을담은사전은기술언어학에근거하여야하며따라서한국수어의실증적인자료를모은한국수어말뭉치를바탕으로한사전편찬이이루어져야한다 양방향이중언어사전중장기구축계획 한국수어사전 의사용자와전문가집단의의견과제언등을종합적으로고려하여국립국어원이수립한 새로운한국수어사전의중장기구축계획은다음과같다 양방향이중언어사전의구조 (1) 거시구조 한국수어-한국어사전ㅇ표제어의선정 - 한국수어말뭉치 ( 15~) 와추가로구축할사전용말뭉치 ( 예정 ) 를바탕으로표제어선정 - 한국수어의고정된어휘부 (frozen lexicon) 를중심으로생활용어, 지역어, 신어등수록ㅇ표제어표기방법 - 한국수어의형태를기록할수있는수어문자 (SignWriting) 로수어를표기하여한국수어가한국어와다른독자적인언어라는인식제고 - 글로스 (gloss) 사용으로인해서발생하는여러문제점개선ㅇ표제어의배열 - 수형기반한국수어사전구축 ( 국립국어원, 2016) 의수어소분류를참조하여배열 한국어-한국수어사전ㅇ표제어의선정 - 한국농인의한국어학습에필요한표제어 10,000개평정 - 한국수어-한국어사전에한국어대응표현으로수록된한국어표제어를중심으로수록ㅇ표제어표기방법 - 표제어는한글로표기하되한국어표제어에대응하는한국수어표현은섬네일 (Thumbnail) 로제시 국립국어원국제학술대회 보이는언어의기록, 수어사전

85 대한민국수어사전의현황과발전방향 최혜원, 이현화 ㅇ표제어의배열 - 가나다순 (2) 미시구조 한국수어-한국어사전ㅇ형식항목 - ( 표제어 ) 기본형과변이형모두확인할수있도록영상으로제시 - ( 마우스제스처 ) 수어를할때필수적으로나타나는마우스제스처를그림으로제시 - ( 수어문자 ) 표제어의형태를수어문자로기록 - ( 수형그림 ) 수어소검색시활용할수있도록수형그림제시 ㅇ내용항목 - ( 대응표현 ) 한국수어에대응하는복수의한국어표현을제시 - ( 뜻풀이 ) 한국수어말뭉치를바탕으로추출한한국수어의의미를한국어로쉽게이해할수있도록기술 - ( 용례 ) 표제어의다양한활용예를수어동영상으로제시 제시된용례를수어문자로표기하고이를클릭할경우해당표제어로이동할수있도록연동기능구현 정확하고자연스러운한국어로번역 - ( 관련어 ) 표제어와일정한의미관계를맺는한국수어를제시 - ( 형태는비슷하지만의미가다른수어 ) 한국수어학습자들이빈번하게나타내는학습오류유형으로, 수어학습의효과를높이기위해관련정보제공 - ( 문법정보 ) 수어의복수형, 부정형, 활용형등문법정보제시 - ( 수어지도 ) 지역변이가매우발달한수어의특성에따라수어지도를통해지역어정보제공 < 그림 3> 한국수어 - 한국어사전정보항목 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 87

86 2 부발표 < 그림 4> 한국수어 - 한국어사전시범화면 한국어-한국수어사전ㅇ형식항목 - ( 표제어 ) 1만개의규모로표제어제시 - ( 활용 ) 표제어의활용형을제시하여한국어학습효과제고 - ( 품사 ) 한국어표제어의품사제시 국립국어원국제학술대회 보이는언어의기록, 수어사전

87 대한민국수어사전의현황과발전방향 최혜원, 이현화 ㅇ내용항목 - ( 대응표현 ) 표제어에대응하는복수의한국수어표현을섬네일 (Thumbnail) 로제시 - ( 뜻풀이 ) 한국어기초사전의뜻풀이제시 - ( 용례 ) 한국어기초사전의용례제시 - ( 한국수어로풀이 ) 번역이아닌농인의이해를기반으로하여한국어표제어의뜻풀이와용례를하나의영상안에서설명 - ( 관련어 ) 한국어표제어와유의, 반의관계를맺고있는단어를관련어로제시 < 그림 3> 한국수어 - 한국어사전정보항목 < 그림 6> 한국어 - 한국수어사전시범화면 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 89

88 2 부발표 양방향이중언어사전구축계획 구분 1 단계 ( 20~ 22) 2 단계 ( 23~ 25) 3 단계 ( 26~ 28) 4 단계 ( 29~ 31) 한국수어말뭉치구축 한국수어자료수집 한국수어자료수집및가공 한국수어말뭉치구축 - 서울외지역말뭉치구축 - 기본전사 - 의미전사 사전용말뭉치기초연구및자료구축 - 의미전사 - 변이형전사 - 문법정보전사 - 마우스제스처전사 사전용말뭉치구축 - 기본전사 - 의미전사 한국수어말뭉치구축 - 기본전사 ( 계속 ) 한국수어말뭉치구축 - 기본전사 ( 계속 ) 멀티미디어자료구축 한국수어 - 한국어사전편찬기개발및기능고도화 한국수어지식관리시스템구축 한국수어자료통합관리시스템고도화 한국수어 - 한국어용례구축시스템개발및기능고도화 한국수어 - 한국어사전서비스시스템구축및기능고도화 - 한국어어휘망활용검색도입및수형검색기능고도화 - 사용자화면편집기능개발 한국어 - 한국수어사전편찬기개발및기능고도화 한국어 - 한국수어사전서비스시스템구축및기능고도화 한국수어뜻풀이영상관리시스템개발및기능고도화 한국어 - 한국수어사전서비스시스템구축및기능고도화 ( 이원화된사전간연결 ) - 모바일웹인터페이스구축 < 한국수어 - 한국어사전 > 양방향 사전 편찬 < 한국수어 - 한국어사전 > - 사전시범제작 - 한국수어표제어구축및지침마련 - 한국수어의미구축및지침마련 - 한국수어뜻풀이구축및지침마련 - 한국수어용례구축및지침마련 - 한국어대응표현구축및지침마련 - 한국수어뜻풀이구축및지침마련 - 수어동영상제작및감수 < 한국수어 - 한국어사전 > - 한국수어표제어추가구축 - 한국수어변이형구축및지침마련 - 수어문자구축및지침마련 - 수형정보구축및지침마련 - 한국수어의미구축확장 - 한국수어뜻풀이확장구축 - 한국수어용례확장구축 - 한국어대응표현구축 - 관련어구축및지침마련 - 수어동영상제작및감수 - 문법정보구축및지침마련 - 수어지도구축및지침마련 - 마우스제스처정보구축및지침마련 - 사용성평가및오류정비 - 수어문자교육및보급 - 오류정비 < 한국어 - 한국수어사전 > - 한국어표제어 1 만개평정 - 한국수어뜻풀이, 용례선정및지침마련 - 한국어의뜻풀이와용례를한국수어로풀이 - 한국수어대응표현구축및지침마련 < 한국수어 - 한국어사전 > - 수어문자교육및보급 - 오류정비 < 한국어 - 한국수어사전 > - 한국어의뜻풀이와용례를한국수어로풀이 - 한국수어로한국어의뜻풀이확장구축 - 한국어의용례수어번역확장구축 - 한국수어대응표현구축 - 한국수어대응표현동영상제작 - 관련어구축 < 표 3> 양방향이중언어사전중장기구축계획 국립국어원국제학술대회 보이는언어의기록, 수어사전

89 대한민국수어사전의현황과발전방향 최혜원, 이현화 오랜시간과대규모의작업범위로농사회의자원이총투입되는이계획에따르면새로운한국수어사전 은 2031 년에완성될예정이다. 향후편찬될사전은다음과같은부분에서이전의사전과는차별화된모습 으로도약할것이라고기대해볼수있다. ㅇ이용자의사용목적에따른사전이원화편찬수어사전을편찬한대부분의나라들이이중언어사전을편찬하였지만현실적인제한 ( 사업기간, 예산등 ) 으로인해하나의사전으로두목적을이루는방향을취하고있다. 이러한사전은주로수어사용자에게는수어에해당하는음성언어대역어를제공하고, 수어학습자에게는수어와관련된정보를제공한다. 물론이것은두가지목적을동시에이룬다는장점이있으나여전히하나의사전으로모든정보를제공하기에는제한적일수밖에없다. 따라서학습자사전으로서의효용성을높이기위해서는두목적을분리한사전의편찬이필요하다. 국립국어원에서계획하고있는사전은목적에따라이원화한것으로 한국수어-한국어사전 은한국수어사용자를위한표현용사전이며, 한국수어학습자를위한이해용사전이다. 반면에 한국어-한국수어 사전은한국어학습자를위한이해용사전이며한국수어학습자를위한표현용사전이다. ㅇ증거가있는사전지금까지한국에서편찬된 6종의수어사전은대체로비슷한방법으로자료를수집하였는데그방법은다음과같다. 첫째, 한국어표제어목록을선정하고이에대응하는한국수어를수집하거나수어표현이존재하지않는경우에는수어를만든다. 둘째, 기존에나와있는수어교재또는사전에서수어단어를표집한다. 이렇게수집한수어를사전으로엮었지만이용자입장에서는어느수어가만들어진것인지, 어느수어가농인이실제로사용하는것인지등을확인할수없어혼란이가중되었고결국사전의신뢰도가하락하였다. 이에수어를기술하는사전으로편찬방향을설정하고말뭉치에근거하여표제어등사전정보를추출하고자한다. 수어를정의하는부분에서중요한것은수어를제1언어로사용하는사람 (native signer) 이해당수어를맥락안에서어떻게사용하는지를보는것이다 (Johnston,1998). 따라서새로운사전이이용자에게단어의적절한용법을가르쳐줄수있도록용례를제공하고자한다. 용례는말뭉치에서추출한용례를부분적으로가공한가공례와농인의직관에의해제작한작성례로구분한다. 이를제시할경우그출처를표시하여사전이용자들이참고할수있도록할예정이다. ㅇ변이수용 Johnston(2003) 은수어에관해서, 어휘적변이 (lexical variation, 다양한수어사용자들이같은의미를완전히다른수어로나타내는것 ) 는몇몇의수어사전편찬자들에게주요한문제가된다고하였다. 그러나이러한변이를없애기위해추진하는수어표준화가얼마나실현가능성이낮은지여러이유를근거로제시하였다. 역사적으로우리나라뿐만아니라다른나라에서도시도되었던음성언어중심의수어표준화는농사회로부터외면을받아왔다. 그러므로이러한변이를없애려는시도보다변이를모두수용하고기록하 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 91

90 2 부발표 여사전이수어언어의총체적자원이되도록해야할것이다. ㅇ수어문자 (SignWriting) 도입수어는문자체계가없는언어로이것은사전편찬을더어렵게만드는요인이된다. 수어사전편찬자들은이러한어려움을극복하고자여러접근을하고있는데 Kristoffersen & Troelsgård(2010) 에따르면편찬된수어사전들은수어표제어를나타내기위해동영상, 그림, 사진, 형태표기 (HamNoSys, SignWriting 등 ), 글로스 (glosses), 숫자, 대역어, 어떻게수어가산출되는지글로묘사하기와같은다양한방법을사용하고있다. 대부분은글로스를사용하는데이것은수어의의미를나타낸음성언어의어휘이며, 수어를기억하기위해사용된다. 음성언어의어휘를하나의라벨처럼수어표제어에부여하고관리하는것은사전편찬자와이용자모두에게상당한편의를제공하지만해당수어가음성언어어휘의의미와완전히동일하다거나혹은그어휘가가지고있는여러의미중에서수어는단하나의의미만을가지고있다는오해를일으키기도한다. 또한음성언어라는라벨을보고수어에접근하는것은결국음성언어를통해수어를인지하도록해서수어학습에방해요소가된다. 수어를정확하게기술하고, 학습자로하여금이를시각적으로인식할수있도록하기위해서향후나올사전에서는수어문자 (SignWriting) 를사용해수어를표기하고자한다. 이미미국, 호주, 벨기에, 브라질, 캐나다등여러나라에서수어문자를사용하고있으며 11) 체코수어사전 12) 에서는수어문자를사용하여표제어를표기하고있다. 수어문자는한국의국가사업으로는처음도입되는것으로이를계기로하여수어로생산되는많은정보가기록, 전승되는발판을마련하게될것이다. ㅇ번역이아닌설명한국수어가하나의언어로인정받기이전의시기에는이에대한이해부족으로한국어에한국수어어휘만대응시킨일대일번역이주를이루었다. 그러나 청각장애학생을위한국어교육용한국어-수어동영상사전개발 ( 윤점룡외, 2009) 에서지적한바처럼시각언어인수어는청각음성언어인한국어와는다른문법적특성을가지고있으므로단어를일대일로대응시켜번역하는것만으로는의미를제대로전달할수없다. 이와같은오류를피하기위하여국립국어원은 2017년에한국어의뜻풀이를한국수어로시범번역하였고 13) 문헌고찰과전문가의견수렴을통해지침을수립하였다. 그러나번역시한국수어가한국어에강하게견인되고, 한국어와한국수어간의설명방식차이가상당하여수어사용자들이번역문을이해하지못하는경우가발생하였다. 이와같은경험에근거하여수어사용자가사전의뜻풀이를직관적으로이해할수있도록번역이아닌설명방식의뜻풀이를제공하고자한다. 또한대부분의경우예시를들어설명하는농인의어법을충실히따라뜻풀이와용례를분리하지않고하나의흐름으로제작하여상호보완적역할을하도록하며이를통해수어사용자의접근성과활용도를높일것이다. 11) 12) 13) 한국수어사전수어뜻풀이와용례구축및정비사업 국립국어원국제학술대회 보이는언어의기록, 수어사전

91 대한민국수어사전의현황과발전방향 최혜원, 이현화 4. 결론 지금까지한국수어사전의편찬현황을개관하고발전방향을제시하였다. 현시점에서그간편찬된한국수어사전은진정한의미의언어사전으로서틀을갖추지못하였다고할수있다. 그러나국립국어원은 한국수어언어법 제정 (2016) 이후에관련정책과연구를위한기틀을마련하고있다. 더불어한국수어사전이언어지식의총체로거듭날수있도록다각도로접근하고있다. 이러한계획이적절히실현되기위해서는대규모의인력이필요하나그간농사회에서연구인력양성과그에대한투자가충분하지못했다. 이러한제한으로사전구축초기에는다소시행착오를겪을수도있겠으나지속적으로연구인력양성이병행된다면농인의언어그리고농인과청인이공존하는방향을담은사전에한발더다가갈수있을것이라기대해본다. 참고문헌 김삼찬, 이선호외 36인 (2005), 한국수어사전, 형설출판사. 김승국 (1993), 표준수어, 오성출판사, 서울. 국립국어원 (2014), 한국수어연구자문위원회회의결과보고서, 국립국어원. 국립국어원 (2016), 수형기반한국수어사전구축, 국립국어원. 국립국어원 (2017), 한국수어사전수어뜻풀이와용례구축및정비, 국립국어원. 국립국어원, 한국농아인협회 (2005), 한국수어사전 김승국 (1983), 수어사전, 단국대학교출판부윤점룡, 정희선, 김선영, 원성옥, 김지숙, 장은숙, 김경진, 임진이, 하길종, 양국종 (2009), 청각장애학생을위한국어교육용한국어-수어동영상사전개발, 한국연구재단. 이정옥, 이준우 (2005), 한국수어관용표현에관한탐색적연구, 특수교육저널 : 이론과실천 제6권제3 호, 한국특수교육문제연구소, 쪽이준우, 남기현, 조준모 (2018), 한국수어사전편찬방향을위한제언, 특수교육저널 : 이론과실천 제19 권제1호, 한국특수교육문제연구소, 쪽이현화 (2017), 한국수어정비사업 : 한국수어사전 ( 한국수어의기록 ), 새국어생활, 제27권제2호, 67-83쪽 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 93

92 2 부발표 최경봉 (2014), 원고로남은최초의우리말사전, 말모이, 새국어생활 제24권제3호, 국립국어원, 표준수어사전편찬위원회 (1982), 표준수어사전, 금옥학술문화재단허일 (2003), 한국농인의언어사전편찬에있어서의쟁점연구, 제5회한국수어학회학술대회. Atkins, B.T.S. & M. Rundell. (2008), The Oxford Guide to Practical Lexicography, Oxford/New York: Oxford University Press. Cokely, D, & Baker-Shenk, K, C. (1980), American Sign Language: A Teacher s Resource Text on grammar and Culture, Gallaudet University Press. Johnston, T.(1998). Signs of Australia. A new dictionary of Auslan(Australian Sign Language) (the sign language of the Australian deaf community), North Rocks, NSW, Australia: North Rocks Press, Royal Institute for Deaf and Blind Children. Johnston, T.(2003). Language Standardization and signed Language Dictionaries, Sign Language Studies, vol 3, Published by Gallaudet University Press, Kristoffersen, J. H., & Troelsgård, T. (2010), Compiling a Sign Language Dictionary: Some of the Problems faced by the Sign Language Lexicographer In M. Mertzani(Ed.), Sign Language Teaching and Learning: Papers from 1st Symposium in Applied Sign Linguistics(Applied Sign Linguistics), Bristol: University of Bristol, Centre for Deaf Studies. Roland Pfau & Markus Steinbach & Bencie Woll(2012), Sign Language: An International Handbook, de Gruyter Mouton R.R.K. Hartmann (1983), 사전편찬의원리와실제 ( 서태길, 김양진, 도원영, 이상혁, 권오희, 옮김 ). 서울 : 제이앤씨.( 원서출판 1983) 국립국어원한국수어사전 ( 수어문자누리집 ( 한국민족문화대백과사전 ( 체코수어사전 ( 국립국어원국제학술대회 보이는언어의기록, 수어사전

93 Current Status and Development Direction of Korean Sign Language Dictionary Hyewon Choi, Hyunhwa Lee Current Status and Development Direction of Korean Sign Language Dictionary Hyewon Choi, Hyunhwa Lee National Institute of Korean Language, Korea 1. Introduction It has been approximately 2 years since the <Korean Sign Language Act> (2016) was implemented. The <Korean Sign Language Act>, embodying a long-standing desire of the deaf who had hoped Korean sign language to be recognized as a language in Korean society and to be granted with the right to use Korean sign language, became the basis for various systems and projects that support the development of Korean sign language. Since the enactment of the <Korean Sign Language Act> in 2016, the Republic of Korea has carried out a wide range of research and projects for the development of Korean sign language including systematic establishment of Korean sign language development plan in every 5 years; research on the use of Korean sign language to improve the environment for the use of Korean sign language; creation of Korean sign language corpus; and full-scale basic research on grammar. In addition, a qualification system for Korean sign language teachers has been put in place along with development of sign language textbooks for different targets and Korean sign language cultural school programs for education and distribution of Korean sign language. The Republic of Korea is also seeking for the ways to improve Korean Sign Language Dictionary to a dictionary that conveys the summation of the language. The compilation of a sign language dictionary is the most fundamental project to publicize the independent status of the sign language. A sign language dictionary that describes sign language as a language, against a common notion that sign language is a sub-language of a spoken language and a subsidiary communication means, proves that the deaf form an independent linguistic community. In fact, many countries have provided opportunities to enhance the status of the deaf community culturally and linguistically by publishing a sign language dictionary. Of course, the form of a sign language dictionary in this case should be more than just a translation dictionary that simply presents sign language equivalents to spoken language words from a spoken language-oriented perspective. When Joseon, an old Korean dynastic kingdom, was a colony of Japan, Korean linguists tried to compile the 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 95

94 Session 2 first Joseon language dictionary in Korean alphabet. However, Joseon Language Dictionary (1920) became the first dictionary published by the Government-General of Joseon. Unlike the title, the dictionary was centered around Japanese language, targeting Joseon language. The Joseon language 1) dictionary, which was compiled while Joseon was not an independent country, consisted of Japanese entries, Korean equivalents in Korean alphabet and definition of Japanese entries. The dictionary was used for a colonization policy to teach Japanese language to Koreans and Korean language to Japanese (Choi Kyeong-bong, 2014). Just as the case above, Korean sign language dictionaries published in the Republic of Korea so far did not have the right frame as a true language dictionary either. Sign language dictionaries in the form of a target language dictionary were made by hearing teachers without deep understanding of Korean sign language as a means to teach a curriculum in Korean texts. This type of dictionaries still continue to follow the old format without being able to overcome their inherent limitations. Sign language dictionaries in western countries 2) had a similar starting as they were for education of the deaf. However, more advanced sign language dictionaries began to be compiled mostly in Europe around the 1990s to present the system of sign language as a natural language from a linguistic approach. The Republic of Korea also announced the Korean sign language as a language through the <Korean Sign Language Act>. Nevertheless, to ensure that a sign language is recognized as a legitimate language in Korean society, the country needs to go beyond publicizing and should compile a dictionary that clearly shows the system of Korean sign language as a language. This conference is expected to play a pivotal role in presenting infinite potential of sign language and creating a dictionary that supports the identity and pride of the Korean deaf community. This article will look into the current status and challenges of the sign language dictionaries published in the Republic of Korea and the mid- & long-term plan established by the National Institute of Korean Language for the compilation of a new Korean sign language dictionary that faithfully includes actual Korean sign language data used by the deaf. 2. Current Status of Korean Sign Language Dictionary When there was no sign language dictionary, a list of sign language words was included in sign language textbooks created for the distribution of sign language. It was not until 1982 that the first sign language dictionary was published in the Republic of Korea under the title of Standard Sign Language Dictionary. As the title 1) After the publication, a Koren dictionary titled as Joseon Language Dictionary (1938) in Korean alphabet was created thanks to the hard work by Moon Se-young, a Korean linguist, and it gave a great hope to the Koreans. 2) An Australian sign language dictionary from 1989 has a trace of applying English language grammar ( -ed for past particle) to sign language expressions. 3) The 'standardization' here is different from general standardization of a language. As there was no collection of a broad range of sign language documents, the standardization was not about selecting the most widely used form as the standard from variations. Rather, it focused on regularizing sign language from a perspective of spoken language by creating new sign language equivalents of Korean language if there was no equivalent and adding supplementary signs to distinguish same signs with different meanings 국립국어원국제학술대회 보이는언어의기록, 수어사전

95 Current Status and Development Direction of Korean Sign Language Dictionary Hyewon Choi, Hyunhwa Lee suggests, this dictionary was created to contribute to standardizing sign language 3) (Lee Jun-woo, Nam Ki-hyun & Cho Jun-mo, 2018). The Korean Standard Sign Language Dictionary Compilation Committee consisting of school teachers was formed to this end and the committee compiled the dictionary which contained sign language words for daily life and the sign language equivalents of Korean words used in elementary and middle school textbooks. As seen in the table below, a total of 6 sign language dictionaries have been published in the Republic of Korea starting from <Standard Sign Language Dictionary> (1982). 4) No. Dictionary/ Institution(compilerr) Year No. of entries Micro structure 1 Standard Sign Language Dictionary Standard Sign Language Dictionary Compilation Committee ,494 2 Sign Language Dictionary Dankook University Press ,966 3 Standard Sign Language Kim Seung-kuk ,966 4 Korean Sign Language Dictionary NIKL Korea 2005 Around 6,800 5 Korean Sign Language Dictionary Kim Sam-chan and others ,818 6 Korean Sign Language Dictionary National Institute of Korean Language (online) ,793 <Table 1> Korean sign language dictionaries published in the Republic of Korea 4) Dictionaries for etymology and terminology are excluded NIKL Conference Records of Visible Language: Sign Language Dictionary 97

96 Session 2 As seen above, the 6 dictionaries presented in Korean alphabetical order are translation dictionaries that contain Korean entries and sign language equivalents. Even the translation in these dictionaries included artificially created signs in addition to actually used sign language simply because it was necessary in expressing Korean words. As a result, an excessive number of sign language entries were compiled including many artificial sign language expressions corresponding to suffixes, prepositions and compound words of Korean language. 5) A translation dictionary in general provides various meanings of an A language entry in B language as words or alternative expressions. For instance, if an entry is Sanghada (go bad) in Korean, a translation dictionary should propose different meanings of Sanghada along with the information on Korean sign language translations. Korean language entry Meaning / Korean sign language translation Get hurt Become thin Get ruined Go stale Go bad <Table 2> Various sign language expressions equivalent to Sanghada in Korean language However, Korean sign language dictionaries either limit the sign language equivalent of a Korean language entry to one and suggest only one Korean sign for a Korea entry (1,5) or simplify the definition of a Korean language entry (4,5,6). This could cause a misunderstanding about Korean sign language among the dictionary users that one sign is the equivalent of one Korean word and that Korean language and Korean sign language share the identical semantic domain. Sanghada Sanghada <Picture 1> Korean language entry Sanghada and Korean sign language with the same gloss 5) The government of the Netherland in 1998 has set Standardisation of the Basic Lexicon of NGT (STABOL) as a priority task to recognize the legal status of Dutch sign language. The project, opposed greatly by sign language researchers in the deaf community, standardized approximately 14,000 words. 33% of them were new words including terminologies for medical and legal sectors as well as for school textbooks. (Roland Pfau, Markus Steinbach & Bencie Woll, 2012) 국립국어원국제학술대회 보이는언어의기록, 수어사전

97 Current Status and Development Direction of Korean Sign Language Dictionary Hyewon Choi, Hyunhwa Lee As these dictionaries were based on the frame of a translation dictionary, the information on sign language was limited to the morphological features of sign language rather than embracing linguistic characteristics. Furthermore, the unnatural linkages of Korean sign language with Korean language in these dictionaries resulted in distorting Korean sign language. For instance, the sign language for Dodal, meaning reaching a destination or a certain level in Korean in dictionary 5 shows that the hand orientation is vertically upward although destination is not always above. Sign language users change the hand orientation to express Dodal depending on the locational features of a destination, but the dictionary does not reflect such usage. Also, due to the characteristics of Korean sign language that has a great number of detailed expressions, a sign language gesture may not correspond to Korean language examples even if the gesture is equivalent to the definition of the Korean entry provided in the dictionary. For example, the sign language for Gajangjari (circumference or edge) in the dictionary is not particularly wrong in general meaning, but is inadequate to be used in an expression as, The edge of an eye is red. <Picture 2> A sign language expression equivalent to an entry Gajangjari Korean sign language dictionaries are not far behind compared to the dictionaries of other countries in terms of the time it was first compiled, the number of dictionaries and the number of entries. 6) Nevertheless, Korean sign language dictionaries are in a stalemate without much improvement when it comes to how faithfully the content reflects the actual language. 7) This is because the compilation process that had continued up to the creation of Korean Sign Language Dictionary in 2005 did not develop much from the basic intention of the standardization of sign language. This limitation is also seen in Korean Sign Language Dictionary 8) (2015) compiled by the National Institute of Korean Language, which is still in service online, as it reflects the format of Korean Sign Language Dictionary created in ) The first sign language dictionary in New Zealand(The Dictionary of New Zealand Sign Language, 1986) included 1,200 entries while the first sign language dictionaries in Australia(Auslan Dictionary, 1989) and Britain(Dictionary of british sign language, 1992) contained 3,500 and 1,800 entries, respectively. 7) The linguistic research methodology for American sign language by researchers of Gallaudet University led by Stokoe in the 1960 affected the researchers in the UK and Europe in the mid-1970s, and sign language became the major subject of research after the 1980s. 8) Korean Sign Language Dictionary (2015) includes the lexicon from Korean Sign Language Dictionary (2005) and 6,000 other words that were added each year under the same direction of compiling dictionary NIKL Conference Records of Visible Language: Sign Language Dictionary 99

98 Session 2 3. Development Direction of Korean Sign Language Dictionary As has been described, Korean Sign Language Dictionary, which was a result of the Korean sign language standardization project, is meaningful in that it collected and studied Korean sign language materials and called attention to the methodologies of sign language research. However, because of the limitations of the time when it was published, the dictionary has continued to receive criticism from its users. The National Institute of Korean Language has humbly accepted the criticisms and is now planning to compile a new Korean sign language dictionary that overcomes the limitations of the previous dictionary. According to Atkins & Rundell(2008), designing an effective dictionary begins from the realistic understanding of the objectives and capabilities of users. As such, the National Institute of Korean Language has collected opinions from the deaf community through various projects to set the right direction for the improvement of Korean Sign Language Dictionary 9). The Korean Sign Language Research Advisory Committee, established in 2014, conducted a Survey on the demand for enactment and standardization of sign language terminology. According to the survey, respondents answered that the biggest reason for not using standardized sign language terminologies was because it was different from the sign language actually being used (28.1%). The Creation of Korean sign language dictionary based on hand orientation project (National Institute of Korean Language, 2016) performed a targeted group interview with 15 deaf, and the result showed that the existing dictionaries were lacking sign language expressions frequently used in the deaf community and disregarded by the deaf as they are centered around sign language expressions that are equivalents to Korean language. The interviewees also answered that the existing dictionaries were used for educating language interpreters and teaching sign language to hearing people, but were difficult even for the deaf to understand which could cause communication problems between the deaf and hearing sign language interpreters. The deaf saw the existing dictionaries as dictionaries they use when necessary but not of a great help and evaluated that even using the dictionaries to understand Korean language was limiting because the information provided in Korean language was difficult for the deaf to understand. The Establishment and modification of definitions and examples of sign language in Korean sign language dictionaries project (National Institute of Sign Language) in 2017 performed a survey targeting 30 sign language experts in the sign language academia and deaf community. The result suggested that the drawbacks of Korean Sign Language Dictionary were poor accessibility and artificially created sign language expression from a macro-structure perspective. For a micro structure, poor matching of information items and lack of examples and information on grammar that are absolutely necessary for the use of sign language. The two studies performed each in 2016 and 2017 also collected suggestions for the development direction of Korean Sign Language Dictionary, and the major details are as follows. 9) This article selectively addresses the opinions that are necessary from the survey results 국립국어원국제학술대회 보이는언어의기록, 수어사전

99 Current Status and Development Direction of Korean Sign Language Dictionary Hyewon Choi, Hyunhwa Lee Creation of Korean sign language dictionary based on hand orientation (National Institute of Korean Language, 2016) - Creation of a new dictionary centered around the deaf users - Easy explanation of Korean language information in sign language expressions commonly used by the deaf 10) - Clear designation of the objectives and target users of a sign language dictionary - Measures to improve accessibility Establishment and modification of definitions and examples of sign language in Korean sign language dictionaries (National Institute of Korean Language, 2017) - Dictionaries that can be used by both the deaf and hearing people - Deletion and modification of artificial entries in existing dictionaries - A dictionary that contains sign language actually used by the deaf and faithfully includes sign language-related information - A dictionary that contains diverse variations of sign language - Development of methods to categorize entries and search for words (motion recognition, etc.) in order to enhance accessibility of dictionary Considering the results described above, the users of the current Korean Sign Language Dictionary can be divided into Korean sign language learners and Korean language learners, and the two groups are using the dictionary with different objectives. Also, the dictionary created and included artificial Korean sign language expressions, but these were not accepted by the deaf community. As such, the dictionary can not perform its role as a dictionary. It is necessary to revamp Korean Sign Language Dictionary based on these opinions to ensure that it reflects the actual language better and it is more complete while compiling a new dictionary so that the users can use a dictionary according to their objectives. According to Hartmann(1983), a bilingual dictionary can be useful as a supportive tool to learn and teach a foreign language. In addition, Lee Hyun-hwa (2017), who discussed the need for publishing a descriptive dictionary for Korean sign language, argued that a Korean sign language dictionary should be compiled based on the entries extracted from the materials that contain natural dialogue of the Korean deaf rather than artificiallymade materials. These suggestions lead to a conclusion that the direction of a Korean sign language dictionary should follow the two below. First, the Korean sign language dictionary should have a clear objective and should be dualized into Korean sign language-korean language dictionary and Korean language-korean sign language dictionary. 5) Cokely, D. & Baker-Shenk, K, C. (1980) noted, in explaining American sign language (ASL), that most of expressions that are described as idioms are actually not idioms and do not fit the definition of idiom and that they are simply sign language gestures that have a few different uses. Lee Jeong-ok & Lee Jun-woo (2005) also summarize the same case in Korea, pointing out that idiomatic expressions are accepted as natural sign language or sign language of the deaf. The result of the survey mentioned in this article can be interpreted within the same context NIKL Conference Records of Visible Language: Sign Language Dictionary 101

100 Session 2 Second, the dictionary that embodies the characteristics of a sign language with a wide range of variations should be based on descriptive linguistics, and, as such, it should be centered around the Korean sign language corpus that has a collection of empirical materials of Korean sign language Mid- & long-term plan to establish a two-way bilingual dictionary The mid- & long-term plan to publish a new Korean sign language dictionary established by the National Institute of Korean Language in consideration of the opinions and suggestions by the users of Korean Sign Language Dictionary and expert groups is as follows Structure of two-way bilingual dictionary (1) Macro structure Korean sign language Korean language dictionary Entry selection - Select entries from the Korean sign language corpus ( 15~) and an additional corpus to be established exclusively for dictionary (tentative) - Include daily vocabularies, dialects and new words focusing on the frozen lexicon of Korean sign language Notation of entry - Notate sign language in sign writing (SignWriting) that could document the form of Korean sign language to raise an awareness that Korean sign language is an independent language from Korean language - Address various issues caused by the use of gloss Entry arrangement - Refer to the classification of chereme in Creation of Korean sign language dictionary based on hand orientation (National Institute of Korean Language, 2016) for arrangement Korean language Korean sign language dictionary Entry selection - Evaluate 10,000 entries necessary for the Korean deaf to learn Korean language - Focus on Korean entries included in the Korean sign language-korean language dictionary as Korean language equivalents Notation of entry - Notate entries in Korean letters (Hangul) but suggest Korean sign language expressions equivalent to Korean language entries in thumbnails Arrangement of entry - In Korean alphabetical order 국립국어원국제학술대회 보이는언어의기록, 수어사전

101 Current Status and Development Direction of Korean Sign Language Dictionary Hyewon Choi, Hyunhwa Lee (2) Micro structure Korean sign language - Korean language dictionary Format items - (Entry) Provide via a video to show both default and variations - (Mouth gesture) Present mouth gestures, essential for sign language, in a picture - (Sign writing) Document the form of entries in sign writing - (Hand shape picture) Provide pictures of hand shape to be used for searching for chereme Content items - (Equivalent expression) Propose multiple Korean language expressions that are the equivalents of a Korean sign language expression - (Definition) Describe the meaning of Korean sign language extracted from the Korean sign language corpus in easy Korean language - (Example) Suggest various examples of entries via a video Notate the provided examples in sign writing and link the notation with the entry concerned so that users can click the link to find the entry Translate examples into accurate and natural Korean language - (Relevant words) Suggest Korean sign language gestures that are semantically related to an entry - (Morphologically similar signs with different meanings) Provide information on the signs that are frequently mistaken by Korean sign language learners for more effective sign language learning - (Information on grammar) Present information on grammar such as plural, negative, conjugation, etc. of sign language - (Sign language map) Include information on dialects via sign language map considering the characteristics of a sign language that has a wide range of regional variations. KSL-KL dictionary information items Format items Entry Mouth gesture Sign writing Picture of hand orientation Content items KL equivalentsmouth Definition Examples Relevant words Information on grammar Similar signs with different meanings Multi-media materials Sign language map <Picture 3> Korean sign language-korean language dictionary information items 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 103

102 Session 2 <Picture 4> A demonstration of Korean sign language-korean language dictionary Korean language-korean sign language dictionary Format items - (Entry) Suggest around 10,000 entries - (Usage) Propose usages of entries for more effective Korean language study - (Word class) Provide word class of Korean language entries 국립국어원국제학술대회 보이는언어의기록, 수어사전

103 Current Status and Development Direction of Korean Sign Language Dictionary Hyewon Choi, Hyunhwa Lee Content items - (Equivalent expression) Propose multiple Korean sign language expressions that are the equivalents of an entry in thumbnail - (Definition) Present definitions from a basic Korean language dictionary - (Examples) Provide examples from a basic Korean language dictionary - (Explanation in Korean sign language) Explain the definitions and examples of Korean language entries via a video based on the understanding by the deaf rather than translation - (Relevant words) Suggest vocabularies that are synonyms or antonyms of Korean language entries as relevant words KL-KSL dictionary information items Format items Entry Usage Word class Content items KSL equivalents Definition Example Explanation in KSL Relevant words <Picture 5> Korean language-korean sign language dictionary information items <Picture 6> A demonstration of Korean Language-Korean Sign Language Dictionary 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 105

104 Session Publication plan for two-way bilingual dictionary Phase 1 ( 20~ 22) Phase 2 ( 23~ 25) Phase 3 ( 26~ 28) Phase 4 ( 29~ 31) Collect KSL materials Collect and modify Korean sign language materials Establish KSL corpus - Create corpus for Seoul and other regions - Basic transcript - Semantic transcript Conduct research for the corpus for dictionary and accumulate materials Establish KSL corpus - Semantic transcript - Transcript of variations - Transcript of grammatical information - Mouth gesture transcript Create corpus for dictionary - Basic transcript - Semantic transcript Accumulate multi-media materials Create KSL corpus - Basic transcript (continued) Create KSL corpus - Basic transcript (continued) Establish KSL Knowledge management system Improve KSL material integration and management system Develop a system to build examples of KSL KL and enhance its functions Develop KSL-KL dictionary compilation system and enhance its functions Establish KSL-KL dictionary service system and enhance its functions - Introduce search functions utilizing Korean language lexical network and enhance search functions based on hand orientation - Develop user interface editing function - Establish mobile web interface Develop KL-KSL dictionary compilation system and enhance its functions Establish KL-KSL dictionary service and enhance its functions Develop a management system for KSL definition videos and enhance its functions Establish KL-KSL dictionary service system and enhance its functions (connect the two dictionaries) <KSL-KL dictionary> Publish two-way dictionaries <KSL-KL dictionary> - Create a dictionary for trial - Establish KSL entries and guidance - Establish KSL meanings and guidance - Establish KSL definitions and guidance - Establish KSL examples and guidance -Establish equivalent KL expressions and guidance - Establish KSL definitions and guidance - Create and edit sign language videos <KSL-KL dictionary> - Establish additional KSL ' entries - Establish KSL variations and guidance - Establish sign writing and guidance - Establish information on hand shape and guidance - Establish and expand meaning of KSL - Expand KSL definitions - Expand KSL examples -Establish equivalent KL expressions - Establish relevant words and guidance - Create and edit sign language videos - Establish grammatical information and guidance - Establish sign language map and guidance - Establish information on mouth gesture and guidance - Evaluate usability and correct errors - Teach and distribute sign writing - Correct errors <KL-KSL dictionary> - Evaluate 10,000 KL entries - Select KSL definitions and examples and establish guidance - Explain definitions and examples of KL in KSL - Establish equivalent KSL expressions and guidance - Create relevant words <KSL-KL dictionary> - Teach and distribute sign writing - Correct errors <KL-KSL dictionary> - Explain definitions and examples of KL in KSL - Expand KL definitions in KSL - Expand KSL translations of KL examples - Establish equivalent KSL expressions - Create videos for equivalent KSL expressions <Table 3> Mid- & long-term plan for compilation of two-way bilingual dictionary 국립국어원국제학술대회 보이는언어의기록, 수어사전

105 Current Status and Development Direction of Korean Sign Language Dictionary Hyewon Choi, Hyunhwa Lee The plan, a long-term large-scale project resourced by the deaf community, states that the new Korean sign language dictionary will be completed in The dictionary that will be published in the future can be expected to be differentiated and more advanced from the previous dictionaries with the following reasons. Dualized compilation of dictionary according to objectives of users The majority of the countries that have published sign language dictionaries have bilingual dictionaries, but these dictionaries attempt to fulfill two different objectives with one dictionary due to realistic limitations (project period and budget). Such dictionaries provide sign language users with the spoken language equivalents of sign language and sign language learners with sign language-related information. This satisfies two objectives simultaneous, but there are limitations in providing all information with only one dictionary. Therefore, to make it more useful as a dictionary for learners, two separate dictionaries with different objectives should be compiled. The dictionary planned by the National Institute of Korean Language is dual dictionaries - Korean sign language-korean language dictionary supports Korean sign language users for expression and Korean sign language learners for understanding. Meanwhile, Korean language-korean sign language dictionary is to support Korean language learners for understanding and Korean sign language learners for expression. Dictionary with evidence The 6 dictionaries published in Korea collected materials in a similar method. First, a list of Korean entires was selected and Korean sign language equivalents were collected or created if there were none. Second, sign language words were sampled from existing sign language textbooks or dictionaries. The collected sign language words were included in a dictionary, but users could not tell which words were artificially created and which were actually used by the deaf. This has led to more confusion and undermined the credibility of a dictionary. Therefore, the National Institute of Korean Language is to compile a dictionary that describes sign language and extract information such as entries based on a corpus. One important thing in defining a sign language is to observe how native signer who use the sign language as the first language use the sign language concerned in context (Johnson, 1998). Therefore, the new dictionary is to provide examples so as to teach adequate usage of a word to the users. Examples will be divided into adapted examples that are extracted from a corpus then partially processed and editorial examples that are intuitively created by the deaf. Examples will have their original sources marked so that the users can refer to them. Acceptance of variations According to Johnston(2003), lexical variation (expression of a same meaning in completely different sign language gestures by sign language users) is a major issue for some sign language dictionary compilers. However, it has been proposed that standardization of sign language that aims to eliminate the variation is unrealistic due to a number of reasons. Historically, sign language standardization based on a spoken language 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 107

106 Session 2 has been disregarded by the deaf community not only in Korea but in other countries. As such, it is more desirable to accept and document all variations rather than remove them to ensure that a dictionary can be a comprehensive resource for sign language. Introduction of SignWriting Sign language is a language without a writing system, which makes it more difficult to publish a dictionary. Compilers of sign language dictionaries are approaching this issue in various ways to overcome the difficulty. According to Kristoffersen & Troelsgård (2010), the existing sign language dictionaries use different ways to present sign language entries such as video, picture, photo, transcript of form (HanNoSys, SignWriting, etc.), gloss, number, translation and written description of how a sign language gesture is created. In most cases, these dictionaries use gloss, which is a spoken language word that represents the meaning of a sign and is used to remember sign language. Assigning a spoken language word to a sign language entry as if a label offers a great convenience to both compilers and users, but it can cause a misunderstanding that the sign language entry concerned is completely identical to the meaning of the spoken language word or that the sign language entry only means one particular meaning out of the various meanings of the spoken language word. Also, approaching a sign language based on the label of a spoken language makes learners recognize the sign language through the spoken language which impedes sign language learning. For accurate description of sign language and visual recognition of the description by learners, the new Korean sign language dictionary will be using SignWriting to transcribe the sign language. It is already being used in many countries including the US, Australia, Belgium, Brazil and Canada 11), and the Czech sign language dictionary 12) presents entries in SignWriting. SignWriting will be introduced for the first time in Korea s national project and it will lay a foundation for a wide range of information created in sign language to be documented and passed down. Explanation, not translation When Korean sign language was not recognized as a language, Korean language words were paired with the sign language lexicon for one-on-one translation due to the lack of understanding. However, as mentioned in Development of a Korean language sign language video dictionary for Korean language education of students with hearing impairment (Yoon Jeom-ryong and others, 2009), Sign language, which is a visual language, has different grammatical characteristics from Korean language, which is a spoken language. Therefore, one-on-one translation of the lexicon can not fully convey the meaning of the sign language. In order to avoid such errors, the 11) 12) 13) Establishment and modification project of sign language definitions and examples in Korean sign language 국립국어원국제학술대회 보이는언어의기록, 수어사전

107 Current Status and Development Direction of Korean Sign Language Dictionary Hyewon Choi, Hyunhwa Lee National Institute of Korean Language translated the definition of Korean language into Korean sign language in 2017 as a trial 13) and established a guidance after studying documents and collecting opinions from experts. However, the result showed that Korean sign language was strongly led by Korean language in translation, and there were some cases where sign language users could not understand translation due to a significant disparity in the ways Korean language and Korean sign language explain meanings. Based on this experience, the new dictionary will contain definitions that are explanations, not translations, so that sign language users can intuitively understand definitions in the dictionary. Also, it will ensure that definitions and examples are provided coherently, rather than separately, in order to faithfully follow the way the deaf present explanation with examples and enhance its accessibility and usability for sign language users. 4. Conclusion This article has looked into the status of Korean sign language dictionary compilation and proposed its development direction. The Korean sign language dictionaries that have been published up to date are not equipped with the appropriate frame as true language dictionaries. However, the National Institute of Korean Language has established a foundation for relevant policies and research since the enactment of the <Korean Sign Language Act> (2016). The institute is in need of a large-scale manpower to materialize these plans, but the deaf community has not been able to generate a sufficient number of researchers or make investment in fostering researchers. Due to such limitations, the project to compile a dictionary may come across some difficulties in the beginning. Nevertheless, continued fostering of researchers in the future will allow us to take a step closer to the dictionary that embodies the real language of the deaf for coexistence of the deaf and hearing people. Reference Kim Sam-chan, Lee Seon-ho & 36 others (2005), Korean Sign Language Dictionary, Hyungseul Publishing Co. Kim Seung-kuk (1993), Standard Sign Language, Osung Publishing House, Seoul. National Institute of Korean Language(2014), Korean Sign Language Research Advisory Committee Report, National Institute of Korean Language. National Institute of Korean Language(2016), Creation of Korean sign language dictionary based on hand orientation, National Institute of Korean Language NIKL Conference Records of Visible Language: Sign Language Dictionary 109

108 Session 2 National Institute of Korean Language(2017), Establishment and modification of definitions and examples of sign language in Korean sign language dictionaries, National Institute of Korean Language. National Institute of Korean Language, Korea Association of the Deaf (2005), Korean Sign Language Dictionary Kim Seung-kuk (1983), Sign Language Dictionary, Dankook Press Yoon Jeom-ryong, Jeong Hee-seon, Kim Seon-young, Won Seong-ok, Kim Ji-sook, Jang Eun-sook, Kim Kyeong-jin, Lim Jini, Ha Gil-jong & Yang Kook-jong (2009), Development of a Korean language sign language video dictionary for Korean language education of students with hearing impairment, National Research Foundation of Korea. Lee Jeong-ok & Lee Jun-woo (2005), An Exploratory Study on the Idiomatic Expressions of Korean Sign Language, Special Education Journal: Theory and Practice Vol. 6, No. 3, Research Institute of the Korea Special Education, p Lee Jun-woo, Nam Ki-hyun & Cho Jun-mo (2018), A Proposal for the Direction of Korean Sign Language Compilation, Special Education Journal: Theory and Practice Vol. 19, No. 1, Research Institute of the Korea Special Education, p Lee Hyun-hwa (2017), Korean Sign Language Alignment Project : Korean Sign Language Dictionary (Documentation of Korean Sign Language), Sae kugo saenghwal, Vol. 27, No. 2, p Choi Kyeong-bong (2014), The First Korean Language Dictionary in Manuscript, Malmoi, Sae kugo saenghwal Vol. 24, No. 3, National Institute of Korean Language, Standard Sign Language Dictionary Compilation Committee (1982), Standard Sign Language Dictionary, Kumok Research and Culture Foundation Heo Il (2003), A Study on Challenges in Compiling Language Dictionary for the Korean Deaf, 5th Symposium of Korea Sign Language Society Atkins, B.T.S. & M. Rundell. (2008), The Oxford Guide to Practical Lexicography, Oxford/New York: Oxford University Press. Cokely, D, & Baker-Shenk, K, C. (1980), American Sign Language: A Teacher s Resource Text on grammar and Culture, Gallaudet University Press. Johnston, T.(1998). Signs of Australia. A new dictionary of Auslan(Australian Sign Language)(the sign language of the Australian deaf community), North Rocks, NSW, Australia: North Rocks Press, Royal Institute for Deaf and Blind Children. Johnston, T.(2003). Language Standardization and signed Language Dictionaries, Sign Language Studies, vol 3, Published by Gallaudet University Press, 국립국어원국제학술대회 보이는언어의기록, 수어사전

109 Current Status and Development Direction of Korean Sign Language Dictionary Hyewon Choi, Hyunhwa Lee Kristoffersen, J. H., & Troelsgård, T. (2010), Compiling a Sign Language Dictionary: Some of the Problems faced by the Sign Language Lexicographer In M. Mertzani(Ed.), Sign Language Teaching and Learning: Papers from 1st Symposium in Applied Sign Linguistics(Applied Sign Linguistics), Bristol: University of Bristol, Centre for Deaf Studies. Roland Pfau & Markus Steinbach & Bencie Woll(2012), Sign Language: An International Handbook, de Gruyter Mouton R.R.K. Hartmann (1983), Lexicography : principles and practice (translated by Seo Tae-gil, Kim Yang-jin, Doh Won-young, Lee Sang-hyuk, Kwon Oh-hee). Seoul:J&C.(Wonseo Publishing Co. 1983) National Institute of Korean Language s Korean Sign Language Dictionary ( SignWriting website ( Encyclopedia of Korean Culture ( Czech sign language dictionary ( NIKL Conference Records of Visible Language: Sign Language Dictionary 111

110

111 주제 5 미래의수어사전 토마스항케독일, 함부르크대학교 Thomas Hanke Hamburg University, Germany

112 3 부발표 미래의수어사전 토마스항케독일, 함부르크대학교 서론 본발표가수어사전에적용되는어떠한사이버기술또는공상과학에나올법한내용에관해소개할것으로예측했다면실망할수있을것이다. 대신수어사전설계단계에서현재의상황을발전시킬수있는사용가능한기술및곧적용할수있을미래의기술에대해소개하고자한다. 본주제는현재필자가직접참여하고있는프로젝트를기준으로작성하게되었다. 필자는 2009년부터독일함부르크과학및인문학회 (Academy of Sciences and Humanities in Hamburg) 의독일수어 (DGS) 말뭉치구축을위한장기연구프로젝트를진행하고있으며연구기간은 2023년까지로말뭉치중심의독일수어사전을편찬할예정이다. 현재 6년정도의연구기간이남았다. 따라서본발표에서다루고자하는 미래 핵심기술은 컴퓨터비전 (Computer Vision) 과 아바타기술 (Avatar Technology) 이다. 미래의수어사전과관련성이있다고판단되는그외기술은이미사용이가능하며적용만하면된다. 필자가진행하고있는프로젝트를소개한후 컴퓨터비전 과 아바타기술 이수어사전에있어필수기술이라고생각하는이유및적용방법에대해설명하도록한다. 그리고눈으로확인할수있는사전을제공하기위해기존의 표준웹기술 과 컴퓨터비전 / 아바타기술 을결합하는방법에관해설명하는것으로발표를맺고자한다. DGS-KORPUS: 말뭉치에서사전으로, 그긴여정에관하여 DGS-Korpus 는함부르크과학및인문학회와함부르크대학교의장기프로젝트이다. 연구기간은 2009 년에시작해 2023 까지진행될예정이다. 이연구프로젝트의목적은두가지로, 독일수어 (DGS) 참 조말뭉치구축및구축된말뭉치데이터를기반으로한현대 DGS 사전편찬이다. 또한프로젝트의모든 국립국어원국제학술대회 보이는언어의기록, 수어사전

113 미래의수어사전 토마스항케 단계에서농공동체 (language community) 의참여를유도하는것이중요한관심사이다. 수어를위한언어학적말뭉치를구축할때언어제공자 (informant) 로부터데이터를수집하고주석을다는방법외에는현재다른방법이없다고판단된다. 인터넷에서수집할수있는 DGS 영상데이터가방대하고일부는공영방송사에서제공하지만이러한출처로부터데이터를추출하여수어사용자및수어가사용되는문맥을대표할수있는표본데이터로활용하기에는거의불가능하며법적및윤리적으로어려움이많다. 데이터수집지역 과거프로젝트경험을기반으로독일전역에걸쳐여러지역에이동식스튜디오를마련하기로결정했다. 이는각촬영지역과해당지역출신의참여자를통해 현지 의분위기를담는동시에전사에필요한고품질의촬영을하기위해내린결정이었다. 물론촬영지역으로선정된장소는앞서언급한 현지성 과언어제공자들의이동시간및이행계획모두를고려하여결정했다. 이에따라, 현재및이전청각장애학교들의통학범위, 교육환경등을결정하는주 ( 州 ) 경계 ( 연방주, Bundesland), 특히옛동 서독국경, 방언경계선으로예상되는경계뿐만아니라촬영장소까지의이동시간등현실적요소를고려하여 13곳의데이터수집지역을선정했다. 선정지역은언어제공자모집을위해또다시최대 5개소지역으로구분했다. 대도시권은도시 농촌혼합지역또는농촌지역과달리그자체로소지역으로구분했다. 농인인구에관한인구조사가충분치않아지역별언어제공자수 ( 및소지역별분포 ) 목표를해당지역전체인구수기준으로결정했으며입증되지는않았으나농인들이대도시에거주하는것을선호한다는경험을바탕으로대도시에는 2배적용하여결정했다. 또한각지역별로최소 16명 ( 남녀최소각2명 *4개연령그룹 ) 의언어제공자모집기준을적용하여총 328명의목표참여자수를산출했다. 실제로는 330명의참여자를촬영했다. 언어제공자선별기준 수어말뭉치구축에이상적인언어제공자는누구인가라는질문은여전히답을찾지못한문제이다. 특히농인부모를둔농인자녀 ( 2세대이상 ) 만을대상으로할지혹은이집단을적어도우선시할것인지하는문제또한정해진것은없다. 연구계내에서심도깊은논의후, 본프로젝트에서는주관적으로나마수어를의사소통의주요수단으로사용하는사람들을대상으로하기로결정했다. 흥미로운점은이정의 ( 이후언어제공자모집절차에적용된기준 ) 에서청각장애를언급하지않는다는것이다. 말뭉치구축에참여한한언어제공자는앞서언급한기준을충족하는건청 ( 健聽 ) 인이다. 지역별및방언별변이에유의하면서우리가정한언어제공자선별기준중최우선기준은바로 현지성 (rootedness) 이었다. 해당지역에서최소 10년이상거주한사람을선별했으며같은지역에서태어나자란사람들을우선시했다. 추가기준으로는사회적지위 ( 주로 화이트칼라 사무직종사자및 블루칼라 생산 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 115

114 3 부발표 직 서비스직구분 ) 와수어습득연령이있다. 수어습득연령의경우, 특정나이를기준으로적용시고령 참여자에서문제가발생하는것을발견했다. 언어제공자모집 각지역별로현지농인 ( 연락담당자, 지역별 1-2명 ) 에게자신의개인네트워크및농인모임등을활용하여 1차적선별기준을충족하는언어제공자후보모집을요청했다. 실제그지역에서필요한언어제공자수보다최소 30% 더많은후보를모집했다. 그다음함부르크팀이후보명단에서참여자를선별했다. 후보자가연락담당자와함께작성한설문지, 연락담당자추천을고려하여선별했으며동시에 2차적선별기준과연령그룹, 성별및소지역간균형을고려하여결정했다. 한지역내분포를고르게하기어려운경우지역간에걸쳐균형을이루도록보완했다. 사전동의및메타데이터 프로젝트종료후에도데이터를보존하는것이필수목적이자동시에현재및미래저작권, 데이터보호권리등법적요구사항및윤리기준을충족하는말뭉치구축프로젝트를위한연구참여동의서마련이어렵다는점은이미널리알려진사실이다. 특히익명성이불가능한수어말뭉치연구에서는더욱어렵다. 연구초기단계부터동료연구자들과법률전문가와함께논의했다. 이논의를바탕으로데이터수집에서미성년자는참여자에서제외하기로결정했다. 저작권법법률전문가의도움으로사전동의서를마련했으며동의서는독일어및 DGS로작성됐다. 선별자들에게는촬영일자보다훨씬이른시기에사전동의서를서면및 DVD 형태로제공하여가정에서논의할수있도록했으며또한현장에서연락담당자와동의서에대해의논할수있는상담시간을진행한후동의서내용을충분히이해했음을확인하는서면서명및영상서명을요청했다. 또한데이터수집에참여요청을받은대상자는각자집에서또는촬영시작전연락담당자와함께 2차설문지를작성했다. 유도과제 (ELICITATION TASKS) 언어제공자들에게 20가지의과제를요청했다. 과제구성방식은스토리텔링 ( 수어, 사진또는영상형태의프롬프트 ), 주어진주제에관한토론및자유로운대화등의형태를취했다. 꼼꼼하게계획한과제로다양한형태의조합을재미있게구성하여총 5시간의촬영시간에도불구하고대부분의참여자들은즐겁게촬영했다. 이전의구어및수어말뭉치프로젝트에서사용했던과제도포함하여비교연구를진행할수있는근간을마련했으며새로운형태의과제도포함했다. 향후데이터수집에도활용하기위해새로개발한유도자료 국립국어원국제학술대회 보이는언어의기록, 수어사전

115 미래의수어사전 토마스항케 전부를공개하지는않았으나요청시연구자들에게제공가능하다. 영상촬영 이동식스튜디오에는 2명의언어제공자및사회자 ( 대부분의경우연락담당자 ) 와의데이터수집세션을 HD로촬영할수있는장비를구비했으며장소를이전및유지하기쉽고임대한장소에맞춰준비했다 ( 스튜디오를촬영장소로이송하여해당장소에서서비스제공업체가설치함 ). 대부분의촬영장소에광대역인터넷연결이어려워촬영데이터는특수서류가방에하드디스크형태로함부르크로발송됐다. 각지역별프로젝트연락담당자는언어제공자모집뿐만아니라사회자로서촬영세션을진행하는방법에관해심도있는교육을받았다. 지역별언어제공자가선별된후연락담당자들의첫업무는각촬영일별로언어제공자그룹을정하는것이었다. 쌍으로정해진두언어제공자는되도록동일한소지역출신으로구성했다. 이렇게선별된언어제공자그룹은대부분같은연령그룹및성별로구성되었다. 사전에언어제공자들에게단색의상을입도록요청했다. 이를잊고온참여자들에게는검은색티셔츠를제공했다. 촬영당시제공된티셔츠는사용여부와관계없이가져갈수있도록선물로제공했다. 언어제공자들에게촬영이총 7시간소요될것으로안내했다. 두번의휴식시간과넉넉한점심시간을제외하고약 시간의촬영이진행됐다. 촬영프로그램을변경해야하는경우도발생했다 ( 한차례단전발생, 폭설또는더위로인한대중교통운행지연 ). 언어제공자한명의건강상의이유로촬영일정을이틀에걸쳐진행한경우는한건발생했다. 검토절차 촬영후언어제공자들에게각자본인의촬영영상을 DVD 형태로 TV 또는컴퓨터로확인할수있도록제공했다. 촬영영상에타입코드를입혀프로젝트또는출판본에서제외하고싶은부분을적을수있도록했다. 우리가예상했던것보다언어제공자들이콘텐츠를이유로제외를요청했던부분은적었다.( 오히려언어제공자들보다연구진에서더많은부분을출판본에서삭제했다.) 몇몇참여자들은상당부분에대한삭제를요청했으나주로코를긁는행위등과같은행위가삭제요청사유였다. 총 330명중 3명은검토서류를제출하지않아동의를철회한것으로처리될예정이다. 말뭉치주석 (CORPUS ANNOTATION) 수어에관한기계판독이가능한참조말뭉치를구축하기위해서는기본주석으로분할 (segmentation) 및표제화 (lemmatisation) 뿐만아니라문어로번역 ( 현실적인이유로 ) 하는작업이필요하다 NIKL Conference Records of Visible Language: Sign Language Dictionary 117

116 3 부발표 번역 다른말뭉치프로젝트경험에따라번역을먼저진행하고다음에토큰 (token) 분할및표제어추출작업을진행한다. 호주수어말뭉치프로젝트 (Auslan Corpus Project) 와같은여타프로젝트와는달리의역을추구하지않았다. 대신번역은가능한한 DGS 발화 ( 원천언어 ) 에가깝게진행했다. 이러한수어텍스트말모듬 (chunk) 은결속성을갖고이해할수있는발화단위를구성해야한다. 발화단위구분 ( chunking 또는경계표지 boundary marker ) 을위해사용되는신호는수어에서사용하는 머리끄덕이기, 몸움직이기, 응시또는리듬을사용하는 등이있다. 이렇게작업된짧은독일어번역문은 DGS 영상에맞춰배치되며이는몇가지역할을한다. DGS를모르는이들도수어텍스트내용을이해할수있으며 DGS 이해수준이다양한주석자역할을하는학생 ( 청인 ) 들이표제어추출하는데도움을준다. 또한검색이가능하며특정수어토큰이사용된문맥을검색할때의발화단위를결정한다. DGS 및독일어를모르는연구자들이본말뭉치데이터를활용할수있도록독일어번역문은다시영어로번역되었다. 분할및표제어추출 수어자동인식프로그램또는도구가없는실정이기에토큰화및표제어작업은여전히주석자들에의해수동으로진행되는시간소모가큰과정이다. 표제어작업전수어텍스트를분할해야한다. 즉, 각수어의발화시간의범위 (time span) 를시작점과종결점이있는토큰표식 (tag) 으로구분해야한다. 이작업은광의적또는협의적방법으로할수있다. 광의적방법이란하나의수어토큰이시작되는지점이바로직전의수어의종결점이라는것을의미한다. 이방법은기호간중간동작이있다는사실을무시한다. 즉, 한가지기호유형의토큰을확인할경우시각적노이즈가존재한다. 수어자동인식을위해머신러닝모델을학습시킬때매우중요한요소이다. 따라서우리는각기호토큰별로시작점과종결점을표시하는협의적의미의표식을적용했다. 분할작업은구어말뭉치작업에서의토큰화에해당된다. 어휘화 (lexicalisation) 정도와관련된번역차용과합성어간구분등에대한몇가지아직풀리지않은문제들이있다. 본프로젝트에서의주석작업환경에서토큰이란단순히시간에맞춰배치된일련의텍스트나열이아닌 (HamNoSys를사용한 ) 유형 (type) 의형태에대한정보를담고있거나대체하는유형과연결된데이터베이스개체이다. 수어어휘주석 (gloss) 은유형표지및일부텍스트의토큰과연결하는표식역할을하나구조적연관성은없다. Johnston의확인용어휘주석 ID-gloss 개념은기저의데이터베이스모델에서이미그기능을하기때문에이맥락에서는필요치않다. 유형-토큰매칭결정은확인하고자하는토큰영상과동일한유형의다른토큰을보여주는비디오를비교함으로써쉽게확인할수있다. 번역및표제어작업외에도주석작성첫번째단계에서입모양 (mouthing) 주석작업과관찰가능한마 국립국어원국제학술대회 보이는언어의기록, 수어사전

117 미래의수어사전 토마스항케 우스제스처 (mouth gesture) 를표시하기로결정했다. 이런결정의배경에는입모양이 DGS 기호토큰의의미에중요한단서이자기호의형태와함께토큰을매칭할기호하위유형을검색할때사용할수있기때문이다. 또한입모양은관습화된수어 (conventional sign) 와생산적인수어 (productive sign) 를구분할때사용된다. 입모양은소문자로주석을단다. 실제발화는무시하고입술모양에서읽을수있는의도했던어휘 ( 어간 ) 를주석처리한다. 현재는마우스제스처의경우, 마우스티어 (tier) 에 [MG] 를더하는단순한방법으로처리하고있다. 마우스제스처를구분할수있는 머신러닝시스템 을개발하기위해박사과정프로젝트를진행중이다. 공개말뭉치 50시간의영상자료와함께공개말뭉치를제공하는것은본프로젝트의주요목표중하나이다. 두개의대상그룹인언어학및기타학문연구자들과 농공동체 는각각 니즈 (Needs) 가다르기에이를충족하기위해우리는두개의별도포털을제공한다. 출판된자료는 농공동체 의문화유산에중요한의의를지닌다. 사람들은데이터를검색하면서 2차세계대전및베를린장벽붕괴당시목격자들의증언, 농인들의학교생활을포함한개인적인삶에대한이야기등재미있는이야기혹은나라전역에걸쳐다양한방식의수어를확인할수있다. DGS를유창하게구사하지못하더라도농인들의삶에대해알고싶어하는이용자또는수어를배우는사람들을위해독일어자막도함께제공한다. 언어학적관점혹은다른목적으로말뭉치를살피고자하는연구자들을위해서는기본주석과함께영상을제공하고독일어-영어번역문도함께제공한다. 총 550시간중공개말뭉치에담을 50시간을선별할때몇가지기준을사용했다. 참여했던모든언어제공자들이공개말뭉치에포함될수있도록했으나주로스토리텔링과같은과제보다자유로운대화형태의영상을담고자했다. 스토리텔링영상의경우, 일부만을공개하여이과제에관심있는사람들이전체영상을확인하기위해연구진측과계약을맺기전샘플로활용할수있도록했다. 자신을소개하고수어이름을설명하는과제는언어제공자들의개인정보를공개하지않는원칙에상충되기에공개말뭉치에서는제외되었다. 공개말뭉치를선택하는과정에있어언어제공자들이제시한출판제한사항을존중할뿐만아니라동시에데이터작업시전사자들이한표시도반영했다. 이렇게우리측에서제외한자료의주된이유는언어제공자들이스스로데이터를검토하면서문제삼지않았더라도 수어공동체 와같이작은공동체에서쉽게누구인지파악할수있는제3자에대한내용과미래에부정적으로영향을미칠수있는언어제공자자신에대한매우개인적인정보등을담았기때문이다. 이는주로법에저촉되는행위및직장에서의부적절한행위등을포함한다. 단, 정치적으로부적합한대화를제외하고자한목적은없었다. 곧공개할말뭉치는농공동체포털의경우거의 50시간분량, 언어학적연구목적의데이터포털의경우 30시간이상의영상과함께공개될예정이다. 두포털은영상에적용되는라이선스와출판된주석데이터에대한설명자료를함께제공한다. 물론라이 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 119

118 3 부발표 선스내용이언어제공자들의연구참여동의서상선택한사항을반영하나언어제공자들이영상에드러날수밖에없다는사실을감안하고있다. 독일법에따라다른말뭉치프로젝트에비해본영상물에대한라이선스가더제약적이다. 라이선스에서허용하는공개말뭉치의데이터이상으로데이터를사용하고자하는연구자들은함부르크학회측에연락하여계약체결후희망하는목적에따라데이터를사용할수있다. 농공동체의참여 우리의주요고려사항중하나는농공동체와협력하고바로자신의언어에대한연구가진행됨에있어참여하고영향을미칠수있도록하는것이다. 수어공동체사람들은우리프로젝트의여러단계에걸쳐참여했다. 데이터수집시언어제공자뿐만아니라프로젝트팀구성원 ( 직원혹은학생보조 ), 연락담당자, 사전사용에대한이용자설문조사참여자, 웹기반 DGS 피드백제공자및포커스그룹 (Focus Group) 구성원등의형태로참여했다. 포커스그룹은농공동체에소속된 11명의농인으로구성되어있다. 농공동체내에서프로젝트, 피드백및결과물을홍보하며프로젝트에기여한다. 또한개별기호및기호의사용에대한정보를제공하여프로젝트에도움을준다. 포커스그룹은우리에게농공동체의관점을알수있는기회가되어생각을공유할수있는중요한역할을한다. 매년열리는포커스그룹총회에서말뭉치에서직접뽑은예시, 후보표제어의형태및내용, 레이아웃기능등사전의여러측면에대해논의해왔다. 일례로, 포커스그룹에서논의한내용을토대로사전표제어에기술하고있는각의미별로용례를한가지이상수록하여이용자들이서면정보에의존하는대신자신의수어능력을사용하여의미를구별할수있도록하자는의견에따라이를반영했다. 말뭉치의미래활용 DGS-Korpus가함부르크와그외지역에서다수의프로젝트가진행될수있도록근간이되어이장기사업을통해개발한말뭉치및어휘적자원이충분히연구되기를바란다. 프로젝트형태는학사, 석사학위연구에서부터다년간의제3자후원프로젝트등다양할수있다. DGS-Korpus가 DGS에대한연구에있어중요한자원이되기를바란다. 말뭉치기반의사전편찬 상당한규모의표제화된말뭉치없이는사전편찬자들은모어사용자들의직관력과수어설명에관한참여자관찰에만의존할수밖에없을것이다. 상당규모의맥락이있는유창한수어사용을검토및분석하지못하는상황에서수어의의미및용법에대한구체적이고신뢰할수있는설명을하기란거의불가능에가깝다. 바로이것이말뭉치의힘이다 국립국어원국제학술대회 보이는언어의기록, 수어사전

119 미래의수어사전 토마스항케 이제활용가능한표제화 DGS 말뭉치데이터의풍부한규모및범위를토대로우리의사전편찬작업을다음단계로발전시킬수있다. 적어도사용빈도가높은수어의경우, 모든중요한기호정보를실제맥락에서사용되는기호의용법및의미를분석한내용을토대로작성할수있게되었다. 또한구어말뭉치기반사전편찬과정에서적용되는방법을수어사전에도적용할수있을것이다. 발생빈도를기준으로표제어선정시말뭉치데이터를활용한다. 우리의사전편찬을위한표제화원칙 ( 표제어구축establishment of lemmas ) 에따라표제어의범위가결정된다. 그다음, 표제어를작성하기위해필요한수어에관한모든정보를준비하기위해관련말뭉치데이터를검토및분석한다. 사전표제어는수어형태, 변이형, 문법정보및관련있는경우어휘의분포정보에대한전반적인정보를담고있다. 동형이의어상호참조, 연관수어및유사한형태의수어도주어진다. 추가설명부분에서는해당수어를포함하여자주발생하는합성어유사한구성및관용표현을기술한다. 사전표제어의핵심은의미나열형태로수어의뜻과용법을설명하는것이다. 뜻과용법을분석하기위해등재후보수어의상당히많은양의토큰및맥락을검토하여맥락적의미및기타관련사실을분석한다. 그다음유사한용법사례를묶어하나의의미로설명한다. 각의미별로그의미를파악, 설명및묘사하기위해여러종류의정보를제공하고그외관련유용한정보를제공한다. 독일어로기술한사전학적정의는해당수어의의미를설명하고 공동의미 (co-sense) 와구별한다. 말뭉치에서인용한예문은의미를더욱설명한다. 어휘변이, 동의어, 유사하거나연관된의미를갖고있는수어및반의어를기술하고상호참조한다. 이는독일어로작성한정의를읽고이해할필요없이특정의미를이해하는데추가적으로도움이된다. 일반적인입모양및연어패턴등기타정보또한제공한다. 활용할수있는말뭉치가있다는것은우리에게수어에대한새롭고구체적인통찰력을제공하고기호사용에대한깊은분석및정의가가능해진다. 그결과우리의사전표제어에는수어및기술된의미에대한높은수준의정보와다양한범위의정보를담고있다. 따라서한항목에서한수어에관한모든가용한정보를포괄적으로제공함에따라표제어구성이복잡한편이다. 현재단계에서는정보를글로스사용없이제공하는방법을모색하고있다. 대신기호형태는마이콘 (micon, 움직이는아이콘moving icon ) 형태와마우스를올려놓으면미니어처영상을재생하는섬네일 (thumbnail) 형태로나타내고있다. 컴퓨터비전 : 이용자가무엇을검색하는가? 독일어전통적사전에서표제어를찾기위해서는이용자는단어 ( 적어도앞부분이라도 ) 표기의철자를알고알파벳순서를알아야했다. 이렇듯표제어를찾기위해상당한지식이필요한데알파벳외문자체계를사용하는언어혹은문자체계가없는언어의경우더욱어려워진다. 수어의경우정해진문어체계가없다. 사전편찬자들은다른 ( 구어 ) 언어에의존하기를원치않아수어의음성학적변수를사용하여사전표제어순서를정했었다. 예를들어, 수위및수형의순서로배치했던것이다. 하지만이용자들은사전편찬자들만큼수형또는수위에따른분류라는개념에익숙하지않아사전을사용하는데또한가지장벽이생기는 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 121

120 3 부발표 것이다. 이관점에서전자사전도놀랄만큼의, 약간의개선안을더했다. 사용된문자체계의알파벳순서를알필요는없지만그외모든문제점은그대로존재했다. 전반적으로사전을사용하는방법을아는사람들이적어지면서사전편찬자들은전통적인접근방법으로수어사전을구축하기에는이용자들이이러한 접근 을위한기술을배워야하는등어려움이존재한다. 바로여기서 예시검색 (Search by example) 이상당히간단한대안이될수있다. 컴퓨터카메라에찾고자하는수어를표현하는것이다 ( 사전은비싸고사용하기어려운기술없이언제어디서든사용가능해야하기에사이버슈트 (cyber suit) 또는데이터글러브 (data glove) 는고려하지않는다.). 사실 2011년에 EU 프로젝트 DictaSign 를통해이러한인터페이스시제품을개발한적이있다. 카메라를향해수어를표현하면시스템은수어동작을하는아바타형태로가장근접한기호들의명단을보여준다. 이명단 ( 되도록짧았으면하는 ) 중에서찾고자했던기호를클릭할수있다. 그당시시제품이 220 기호기준으로잘작동되었으나규모를늘리기는어려웠다. 또한키넥트 (Kinect) 센서가다른유형의웹캠또는휴대폰카메라로대체된다면더개선될것이다. 아직개선의여지가있다는것을의미한다. 최근에는미국카네기멜런대학교 (Carnegie Mellon University) 에서개발한머신러닝프레임워크인 오픈포즈 (OpenPose) 의활용가능성을검토하기시작했다. 이름에서시사하듯영상이미지로사람수와무관하게포즈 ( 동작 ) 를인식하는것이다. 영상속인물을추적하지는않으나프레임별로재인식한다. 간단한신체골격과관절의위치를 2D 형태로보여주며각포즈가얼마나정확한지를알려주는신뢰수치를제공한다. 이라이브러리의최근버전은얼굴및손을더욱구체적으로인식할수있다 ( 처리시간은더욱길어졌다 ). 우선오픈포즈를활용하여적은양의데이터를실험해본결과가능성이있어보인다. 수어영상처리에관한지식을활용한다면오픈포즈데이터를토대로 3D정보도재현할수있을것이다. 오픈포즈에추가지도학습 (supervised learning) 층위를더하고학습에사용하는수단으로주석말뭉치데이터를활용하는것이다. 과거의접근방법에비해이방법이규모를확장하기수월할것이라판단되지만데이터처리가완료될때까지기다린후에공식평가를할수있을것이다. 이러한진전이있음에도불구하고이용자가달성하고자하는것에따라 예시검색 의대안방법도모색해야할필요가있다는점을잊지말아야할것이다. 한가지시나리오를들어보자. 모르는수어를본이용자가있다. 불행히도이이용자는수어를구체적으로보지못했다 ( 혹은보았지만일부잊었다.). 이를해결하기위한접근방법은두가지가있다. 첫번째는이용자가디폴트 데이터 를추가하여수어를완성하고이를카메라에표현한후보기중제외하고자하는파라미터를선택하는것이다. 또다른방법으로는찾고자하는수형, 수위, 움직임을시스템에전달하는유저인터페이스가있어이용자는확실한정보만을입력하고더세부적인유저인터페이스를사용하여시스템이총체적접근을채택할때보다더많은디테일을인식할수있도록하는것이다. 예를들어, 웹캠을통해상체를캡처하는방법보다더구체적으로수형을인식하는립모션 (Leap Motion) 카메라를활용하는시스템몇가지가개발중이다. DictaSign에도수위를구체화하는시제품이있었다 국립국어원국제학술대회 보이는언어의기록, 수어사전

121 미래의수어사전 토마스항케 아바타기술 : 사전이당신에게수어하게하라! 독자모두가수어아바타를본적이있을것이다. 사실전세계적으로진행되고있는수어아바타프로젝트가굉장히많아엄청난양의수동작업을통해조율하지않는이상이해할수있고믿을수있는수어동작을할수있는아바타가많지않다는사실이놀라울정도이다. 모든수어표제어를수동으로조율하는것이가능할수있겠지만우리는아바타가말뭉치의예시를수어동작으로보여주는것에관심을갖고있다. 물론이경우수동조율은불가능하다. 자동적수어표현은모션캡처데이터또는 HamNoSys의음성학적데이터등을토대로한합성생성 (synthetic generation) 으로재현한다. 첫번째방법은특수장비및상당한후처리시간이소요된다. 또한한명의혹은여러명의수어모션캡처데이터를어떻게결합할것인지도불명확하다. 합성생성방법은어휘데이터베이스에있는음성학적데이터를토대로이루어져큰노력없이가능하다. 단, 표기법에따른수어는로봇이수어하는느낌이있다. 필자의의견은하이브리드방식이가장잠재력이있다. 사용가능한경우표준모션캡처데이터를사용하나필요한경우합성생성으로구현하는것이다. 일반적인지속시간 (duration) 또는실제지속시간등과같은말뭉치데이터를활용할수있으며 ( 앞으로는오픈포즈데이터도활용하여 ) 조금더사람이하는수어동작에가까워질수있을것이다. 시카고드폴대학교 (DePaul University) 의폴라프로젝트 (Paula Project) 에폴라아바타가하이브리드데이터를기반으로 DGS 수어동작을표현할수있도록참여하고있다. 이시점에서높은품질의영상이있는데말뭉치의예문을아바타가보여주는방식을채택하고자하는이유에대해궁금해질수있다. 사전에예시로사용할수있는공개말뭉치상의좋은수어데이터를보유하고있다는점을자랑스럽게생각한다. 하지만말뭉치규모가제한적이라는점을감안했을때 ( 최근문어말뭉치에비해제한적인규모 ), 타당한이유로공개말뭉치에서는제외됐지만제외된예시문도제공하고싶은것이다.( 개인의삶이야기를많이포함하고있어본연구말뭉치에서는종종발생한다.) 제3자에대한내용이라면수어모델을활용하여언어제공자가스튜디오에서한수어를따라하고제3자언급에대한내용은대체하는방식을채택할수도있다. 하지만자신의수어가아닌수어영상이인터넷상에공유되기때문에타인의수어를그대로따라하는것을불편해하는수어모델들도있다. 바로이러한경우에아바타가완벽한익명성을보장하는전략이된다. 연어패턴도또하나의잠재응용분야이다. 모든패턴요소가단한명의수어를기준으로할가능성은거의없다. 여러사람이연어패턴을표현하는것은이용자를불편하게할수있다. 따라서마찬가지로이분야에서수어모델또는아바타에의존해야한다. 말뭉치예문대신만들어진예문을사용하는사전의경우, 사전편찬자는아바타로하여금이용자에게맞춘방언변이형으로수어동작을하도록하여사전을맞춤화할수도있다. 15년전디지털영상대신아바타를사용하면서절감할수있는광대역을이유로들어아바타를옹호하는근거가되었을수있다. 이제는대부분의공동체에서이주장은더이상근거가없지만일부의경우에는아직유효할수있다. 사전편찬자들에게선택의여지가있다 NIKL Conference Records of Visible Language: Sign Language Dictionary 123

122 3 부발표 사전의시각적탐색 (VISUAL EXPLORATION) 기호형태를매우정확하게검색하는것이가능해진다하더라도이용자들이단일의표제어를찾는목적이상으로사전의콘텐츠를탐색하고활용하는것이사전편찬자들의꿈일것이다. 과거종이사전시대에는찾고자하는표제어를찾기위해종이를뒤적이다가자연스럽게여러항목을확인하게되고그중일부가시선을끌었을것이다. 오늘날사전편찬자들은이용자들이다른표제어도찾아보도록유도하기위해서는다른표제어에상호참조를제공하거나다수의표제어를제공해야한다. ASL-LEX 프로젝트는잘작성된그래프와함께굉장히매력적인콘텐츠시각화를구현했다. 하지만이방법이모든사전에적합한것은아니다. 우리가목표했던크기의사전에맞춰적용했을때는 ASL-LEX 프로젝트에비해시선을끌지못했다. 따라서오늘논의하는사전프로젝트별로각각다른해결책이필요할것이다. 우리가구축한 DGS 말뭉치에는 DGS에서발견되는변이형 ( 지역별및연령별 ) 의상당수를반영한다. 피드백에따르면지역별변이형은 DGS 공동체에게큰관심사이며우리가맵 (map) 으로써제공할수있는데이터에이용자들의만족도가높은것으로나타났다. 이때잘구현된맵을작성하기위해 d3js 표준웹기술을적용했다. 결론 본발표를마무리하기전에미래의수어사전의모습은어떨지에대해전망해보도록하자. 잭인더박스 (Jack-in-the-box) 장난감을기억하는가? 그아이디어를수어에적용해보자. 다시말해서, 아바타를통해수어를확인하고싶을때박스를열어바로확인하는것이다. 대신박스속잭 (Jack) 이아닌가상의 ( 혹은실존의 ) 수어사용자의홀로그램디스플레이인것이다. 그리고박스일필요가없다. 삼성 20h 또는 iphone 20xh를책상에엎어놓고홀로그램수신기를켠다고상상해보라. 수어통역이필요한사람옆에프로젝터를활용하여수어통역사를띄우는모습을상상해보라. 미술관에서안내하고있는작품바로옆에가상의안내원이있는모습을상상해보라. 만약홀로그램이너무 스타워즈 와같은것으로들린다면증강현실 (AR) 을떠올리면될것이다. 특수안경을통해볼수있도록컴퓨터가생성한콘텐츠를현실세계에이미지로투사할수있도록하는새로운기술이다. 물론우리가원하는증강현실시나리오속가상의미술관안내원은전시회안내동안발화만이아닌수어로안내하는시나리오일것이다! 안내원이사용한수어중에이해하지못한기호가있다면간단하게눈을깜빡여바로사전의관련페이지로연결되는모습을상상해보라. 마치오늘날킨들 (Kindle) 로외국도서를읽을때처럼말이다. 추신. 사전페이지 라고했던가? 사전은종이사전에서하이퍼텍스트 (hypertext) 형태로변했다. 미래사전의모습을상상할때는페이지개념을무시해야할지도모른다. 가상의 3D 세상에서는사전표제어가마치방이여러개있고이웃과연결되는문이많은아파트와도같은구성일지도모른다. 경청해주셔서고맙습니다 국립국어원국제학술대회 보이는언어의기록, 수어사전

123 Sign Language Dictionaries of Tomorrow Thomas Hanke SIGN LANGUAGE DICTIONARIES OF TOMORROW Thomas Hanke Hamburg University, Germany INTRODUCTION If you expect this talk to be on cybertechnology or any kind of science fiction applied to sign language dictionaries, I have to disappoint you. Instead, this talk focuses on technologies already available or becoming available in the very near future that would help us to evolve the current state of the art in sign language dictionary design. This focus is motivated by the project I am currently involved in: I am leading a long-term research project of the Academy of Sciences in Hamburg which started in 2009 with collecting a corpus of DGS (German Sign Language) and shall end in 2023 with a corpus-oriented dictionary of DGS, i.e. in slightly more than six years. So the future core technologies I will cover here are Computer Vision and Avatar Technology. Any other technologies that I see relevant for future sign language dictionaries are already available today and only need to be applied. After giving a bit of background on our project, I will outline why I consider Computer Vision and Avatar Technology essential for sign language dictionaries and how they fit. Finally, I will give some ideas how these technologies combine with standard web technologies to provide visually explorable dictionaries. DGS-KORPUS: THE LONG JOURNEY FROM CORPUS TO DICTIONARY DGS-Korpus is a long-term project of the Academy of Sciences and Humanities in Hamburg and Hamburg University. It started in 2009 and is scheduled to end in The project has two major objectives: To build a reference corpus of German Sign Language (DGS) language data and to publish a dictionary of contemporary DGS based on the corpus data. Beyond the output, it is a major concern of the project to involve the language community in all phases of the project NIKL Conference Records of Visible Language: Sign Language Dictionary 125

124 Session 3 We believe that when building a linguistic corpus for a sign language, there currently is no way around collecting data from informants and to annotate these. While there is a huge number of DGS video data available on the internet and some from public broadcasters, it is practically impossible and both legally and ethically very difficult to sample data from these sources to be representative for the language with respect to the users of the language and contexts in which the language is used. DATA COLLECTION REGIONS From experiences in earlier projects, it was one of the key decisions to have a mobile studio to be set up in different places across Germany. The idea was to have quite a local spirit with the recording location in the region and all persons involved coming from that region while still ensuring high-quality recordings needed for transcription. Obviously, the number of locations selected for recordings needs to be a compromise between localness in the above sense, but also relevant for the informants travel times, and the logistics. Our solution was the definition of thirteen data collection regions, trying to respect the catchment areas of current and former Schools for the Deaf, state (Bundesland) borders determining a. o. educational settings, especially the former border between West and East Germany, suspected dialectal borders, but also practical considerations such as travel times to the recording locations. The regions were further subdivided into up to five sub-regions relevant for informant selection. Large metropolitan areas form their own sub-regions, in contrast to others with mixed or more rural structures. Due to the lack of census data on the Deaf population, the target number of informants per region (and a suggestion for the distribution onto the sub-regions) was determined from the population figures of the community at large, with a weight of 2 for larger cities to reflect (unproven) experience that Deaf people often prefer to live in larger cities. Together with a set minimum of 16 informants per region (to cover four age groups times two sexes with at least two informants each), this resulted in a target number of 328 participants. We actually filmed 330 participants. INFORMANT SELECTION CRITERIA It was and still is an open question who would be the ideal informants for a corpus of a sign language this is especially true for the question whether one should exclusively target Deaf of Deaf (2nd or higher generation deaf) or at least prioritise this group. After intensive discussions within the research community, our decision was to consider people who (subjectively) use sign language as their primary means of communication. Interestingly, this definition (that was then implemented by the recruiting procedures) does not even mention deafness, and in fact we have one hearing informant in the corpus who completely satisfies the criteria defined. With an eye on regional/dialectal variation, rootedness in the region was set a high-priority criterion: People 국립국어원국제학술대회 보이는언어의기록, 수어사전

125 Sign Language Dictionaries of Tomorrow Thomas Hanke needed to be resident in a region since at least 10 years, with a preference for people born and raised at the same place. Social background (primarily just white vs. blue collar) and sign language acquisition age were additional criteria. In the latter case, it turned out that any fixed cut-off age would have turned out problematic especially for elderly people. INFORMANT RECRUITING For each region, we contracted local deaf persons ( contact persons, one or two per region) who used their personal networks as well as events in Deaf clubs etc. to recruit informant candidates fulfilling the primary criteria, at least 30% more people than actually needed for each region. It was then the team in Hamburg to decide whom to invite from the list of candidates. These decisions were based on a questionnaire filled in by the candidate together with the contact person, recommendations by the contact persons, also considering the secondary criteria and the balance between age groups, sexes and sub-regions. What could not be balanced within a region was then compensated across regions. INFORMED CONSENT & METADATA It is well known by now how difficult it is to set up an informed consent for participation in a linguistic corpus project where it is an integral aim to preserve the data beyond the end of the project, satisfying legal requirements such as current and upcoming copyright, data protection right etc. as well as ethical standards. This is even more true for sign language corpora where anonymisation is close to impossible. We had started discussions with colleagues and legal experts early on. This led us to the decision not to include minors in the data collection. Our informed consent was developed in cooperation with a legal expert on copyright law und formulated in both German and DGS. The invitees received the informed consent on paper and on DVD for consultation at home well before their recording day and had time on-site to discuss the informed consent with the contact person before being asked to sign the paper and sign into the camera that they had signed the contract fully understanding what it meant. People invited to participate in the data collection also received a second questionnaire to be filled out at home or together with the contact person right before the recording sessions started. ELICITATION TASKS We used a set of 20 different tasks for the informant. The formats ranged from story-retelling (with prompts in sign, picture, or movie) to discussions on a given topic as well as free conversations. With careful planning, it 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 127

126 Session 3 was possible to make the mix of formats diverting enough that most participants enjoyed the recording session despite a net length of 5 hours. The set contains a number of tasks previously used in other corpus projects, both on spoken and sign languages, to lay a basis for cross-linguistic research, as well as new formats. Not all details of the newly developed elicitation materials are available in the publications in order to keep the material suitable for future data collections. The materials are, however, available to other researchers upon request. RECORDING The mobile studio was equipped to record the data collection sessions with two informants and a moderator (in most cases the contact person) in HD quality while remaining easy to relocate and maintain, and adaptable to the rooms rented. (The studio was transported to the recording locations and built up there by a contracted service provider.) As most recording locations did not offer broadband internet connections, data was shipped to Hamburg on hard disk sets in special suitcases. The project s contact persons in the region had not only received extensive training on informant recruitment, but also on how to conduct the recording sessions as moderators. Their first task after the informants for the region had been selected was to arrange pairs of informants for each recording day. Preferably the two persons should be from the same sub-region when possible. The majority of pairs filmed are same-age group and samesex. Informants had been asked beforehand to wear single-colour clothing. Where people had forgotten this request, we offered them black T-shirts. (This was a gadget to take home whether they had used it in the recording session or not.) Informants were asked to plan in 7 hours on-site. This included two coffee breaks and a longer lunch break, leaving 5 to 5.5 hours for the recording. In some cases, the recording programme had to be adapted (one power outage, public transport delays due to snowfall and heat). There was just one case where the recording session had to be split into two days due to the health condition of one of the informants. REVIEW PROCEDURE After the recordings, informants received their recordings as a set of video DVDs to be viewed on a TV set or a computer. The recordings were overlaid with timecodes, allowing the informants to write down time spans from the recordings they wanted to be excluded from any usage in the project or just from publication. Informants marked fewer spans to be excluded for contents reasons than we had expected (and in fact we exclude more passages from publication than the informants themselves). Some people did mark quite a number of time spans, but they were more concerned about things like scratching their noses. Three out of 330 people did not return their review sheets, now to be treated as if they had retracted their 국립국어원국제학술대회 보이는언어의기록, 수어사전

127 Sign Language Dictionaries of Tomorrow Thomas Hanke informed consent. CORPUS ANNOTATION In order to build a machine-readable reference corpus of sign language, segmentation and lemmatisation as well as, for pragmatic reasons, translation into a written language are minimally required as the core of any basic annotation. TRANSLATION Following experiences from other corpus projects, we start with the translation first, then segment and lemmatise tokens. Unlike other projects e.g. the Auslan Corpus Project, we do not aim for a free translation. Instead, the translation should be as close to the DGS utterance (source language) as possible. These chunks of signed text should form a coherent and comprehensible meaning or utterance unit. Further cues for chunking (or boundary markers ) are signals inherent to sign languages like head nod, body shift, gaze or rhythm. The more or less short written German sentences are aligned with the DGS video and serve several functions: They give access to the content of the signed texts even to those not knowing DGS and guide our (hearing) student annotators with a varying competence of DGS when lemmatising. In addition, they are searchable and define preliminary utterance units when looking for the context in which a sign token is used. In order to make the corpus data accessible to researchers knowing neither DGS nor German, the German translations are also translated into English. SEGMENTATION AND LEMMATISATION With no automatic sign recognition software or tools at hand, tokenisation and lemmatisation still remain a time consuming step done by human annotators. Before lemmatising, the signed text has to be segmented, i.e. the time span of each articulation of a sign must be determined by creating token tags with start and end points. This can be done in a broad or narrow way. The broad way means that one defines the starting point of a sign token being the ending point of the previous sign or vice versa. This approach ignores the fact that there are transitional movements between signs, i.e. when checking sign tokens of one sign type there is more visual noise. This is crucial when trying to train machine-learning models for automatic sign recognition. We therefore opted for a narrow segmentation in defining start and end for each sign token. Segmentation roughly corresponds to tokenisation in vocal language corpus processing. There are a number of open questions such as the distinction between loan translation and compounds, closely related to the degree of lexicalisation. In our annotation environment, tokens are not mere time-aligned text strings, but database entities linked to a 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 129

128 Session 3 type, inheriting or superseding the type s form information (using HamNoSys). Glosses are labels for types and linked tokens convenient in some contexts, but they have no structural relevance. Johnston s notion of ID-glosses is superfluous in this context as its function is already guaranteed by the underlying database model. Decisions in type-token matching can be easily verified by comparing the token video in question with videos showing other tokens of the same type. Besides translation and lemmatisation, we decided to also annotate mouthings and give an indication of observable mouth gestures in the first phase of annotation. The reason is that mouthings are an important clue to the meaning of a DGS sign token which, in combination with the sign form, can be used to search for the appropriate sign sub-type the token should be matched to. Furthermore, mouthings are used to distinguish between conventional and productive uses of sign types. Mouthings are annotated in lower case. Disregarding the actual articulation, at least the intended word (word stem) to be lip-read is annotated. For the time being, mouth gestures are annotated in a simplistic way by just adding [MG] in the mouth tier, as we have a PhD project running to develop a machine learning system to tell mouth gestures apart. PUBLIC CORPUS A public corpus with 50 hours of video is one of the key deliverables of the project. In order to best serve two target groups with rather different needs, i.e. researchers from linguistics and other disciplines on the one hand and the language community on the other, we provide two separate portals: For the language community, the published material is an important contribution to cultural heritage. People are invited to browse through the data to find interesting stories like eye-witness reports from WW II to the fall of the Berlin wall, personal stories from Deaf life, especially school experience, or just different signing styles from across the country. For people interested in learning more about Deaf life without being fluent in DGS or for people learning sign language, German subtitles are provided. For researchers interested in exploring the corpus linguistically or otherwise, the videos are provided together with basic annotation as well as translation into German and English. There are a number of criteria we used to identify the 50 out of 550 hours to become part of the public corpus. While we wanted everyone who participated to be visible in the public corpus, we strongly preferred relatively free conversations over tasks such as story-retelling. Of the latter category, we only included a couple of examples so that people interested in these tasks can view an example before entering an agreement with us to have access to the full set. The tasks on presenting oneself and explaining one s sign name was completely excluded from the public corpus as it conflicts with the rule not to make informants contact data public. The public corpus selection obviously respects the publication restrictions as set up by the informants, but also marks set by our transcribers when working on the data. Primary reasons for such exclusions from our side were conversations about third persons who could be easily identified from the content within a community as small as the sign language community, very private detail as well as facts about the informants themselves that 국립국어원국제학술대회 보이는언어의기록, 수어사전

129 Sign Language Dictionaries of Tomorrow Thomas Hanke might shed a bad light on them in the future, even if the informants themselves obviously did not consider these passages problematic when reviewing the data. This includes conflicts with law and inappropriate behaviour at the working place. It was not our aim, however, to exclude politically incorrect talk. The upcoming new release of the public corpus will almost reach the 50 hours for the language community portal and more than 30 hours for the linguistic research data portal. The portals also explain the licences applicable to the video and annotation data published. Of course, the licenses respect the options the informants chose in their informed consent, but they also take into account that the informants are inevitably visible in the videos. Following the constraints of German legislation, these licenses are more restrictive than those found with other corpus projects. Researchers interested to use data beyond what is possible with the licenses on the public corpus are of course invited to contact us and arrange contractual agreements hopefully allowing their intended uses. INVOLVEMENT OF THE LANGUAGE COMMUNITY One of our key concerns is to cooperate with the language community and to make sure that it has some influence on and part in the research on their language. Members of the signing community are involved at many levels of our project: They are not only informants in the data collection, but also members of the project team (as staff and student assistants), contact persons, participants of the user survey on dictionary use, participants of the web-based DGS-Feedback and members of the Focus Group. The Focus Group currently consists of eleven Deaf members who are well rooted in the Deaf Community. They support the project within the Deaf Community by advertising the project, the feedback and our products. Also, they support our work in answering questions on individual signs and their uses. The group is an important place for us to discuss some ideas because it makes the Language Community s perspectives on relevant issues available to us. In yearly meetings with the focus group we have discussed many aspects of the dictionary such as examples taken directly from the corpus, form and content of preliminary entries as well as layout functions. One decision following a discussion in the focus group was to generally include more than one example for each sense displayed in a dictionary entry to enable users to differentiate senses using their sign language competence rather than having to rely on written information. FUTURE USE OF THE CORPUS DGS-Korpus hopefully lays the ground for a multitude of satellite projects, in Hamburg and elsewhere, fully exploring the corpus and lexical resources developed in this long-term enterprise. These projects may range from BA and MA theses projects up to multi-year third-party funded projects. We expect the DGS-Korpus to be an important language resource for generations of research on DGS NIKL Conference Records of Visible Language: Sign Language Dictionary 131

130 Session 3 CORPUS-BASED LEXICOGRAPHY Without a lemmatised corpus of a considerable size, dictionary makers have to rely solely on native speakers intuition and participant observation for the description of signs. It is at least very difficult if not almost impossible to achieve a detailed and reliable description of the signs meanings and uses without the possibility of reviewing and analysing a considerable amount of natural fluent signing in context that is corpus evidence. With the substantial size and scope of the lemmatized DGS Corpus data now available, we are finally able to take our lexicographic work to the next level. At least for signs of higher frequency we can now base all crucial sign entry information on analyses of actual uses and meanings of the signs in context and we can start to employ methods well established in corpus-based lexicography of spoken languages. The corpus data is used for lemma selection on the basis of frequency. Following our lemmatisation rules for the dictionary ( establishment of lemmas ), the scope of an entry is defined. Then relevant corpus data are reviewed and analysed to prepare all information on the signs relevant for writing an entry. Dictionary entries contain general information on the sign form and its variants and some grammatical and, if relevant, distributional information. Cross-references to homonyms, related sign and signs of similar form are given. In the run-on section, frequent compound-like constructions and idiomatic phrases that contain the sign are listed. The core of a dictionary entry is a description of the attested meanings and uses of a sign in the form of a list of dictionary senses. For the analysis of the meanings and uses, we review a substantial number of tokens of each sign candidate and their contexts and take notes on their contextual meanings and other relevant facts. Similar instances of use are then grouped together and described as individual senses. For each sense several kinds of information are given to identify, describe and illustrate this particular sense and to provide other useful information relating to it. A lexicographic definition written in German explains the given sense and distinguishes it from its co-senses. Example sentences taken directly from the corpus illustrate the sense. We list and cross-reference lexical variants, synonyms and signs of similar and related meanings as well as antonyms. This provides an additional approach to grasp a particular sense without reading and understanding a definition written in German. Other information types like typical mouthings and (collocation) patterns are also provided. Having a corpus to work with provides us with new and more detailed insights on signs and allows for more detailed analyses and description of their uses. Consequently, our dictionary entries contain a high degree of detail and a large variety of different kinds of information on the signs and their listed senses. Therefore, the entry structure is rather complex as it provides a comprehensive view of all the available information on one sign in one place. At the present stage we are experimenting with ways to present the information structure without using any glosses. Instead, forms are represented by micons ( moving icons ), thumbnails that play a miniature movie showing the sign as soon as the user hovers over the thumbnail 국립국어원국제학술대회 보이는언어의기록, 수어사전

131 Sign Language Dictionaries of Tomorrow Thomas Hanke COMPUTER VISION: WHAT IS THE USER SEARCHING FOR? In order to find an entry in a traditional dictionary of German for example, the user has to know how to spell the word s citation form (at least the beginning), and has to know the sorting order of the Latin alphabet. This is quite a lot of knowledge just to find an entry, but it gets worse with languages for a non-alphabetic writing system or even no writing system at all. For sign languages, there are no established writing systems. Dictionary makers not wanting to rely on another (spoken) language often used selected sign language phonetic parameters to order the entries in a dictionary, such as by location height and handshape. But as users not used to the level of abstraction used by the dictionary maker in grouping handshapes or locations, this means another hurdle to access the dictionary. Electronic dictionaries are only a surprisingly small step forward in this respect: While they eliminate the need to know the alphabetic ordering of the writing system used, all other issues remain valid. As lexicographers in general are faced with fewer and fewer people having the knowledge how to use a dictionary, a traditional approach is not likely in the case of sign language dictionaries where people interested would need to learn these access techniques. This is where the Search by example provides a seemingly straightforward alternative: You just sign the sign you are interested in into the computer s camera. (I am not considering cybersuits or datagloves as a dictionary needs to be accessible anywhere anytime without expensive or difficult-to-use technology.) In fact, we developed a prototype of such an interface in the EU project DictaSign back in You sign into a camera, the system replies with a list of the closest matches represented by an avatar signing them. This allows you to select the sign you are looking for by simply clicking onto it in a (hopefully short) list. While that prototype worked quite well with a set of 220 signs, it did not scale well. Furthermore, it would be nice if the Kinect could be replaced by any kind of webcam or phone camera. So there is quite some work to be done! We have recently started to explore the possibilities of OpenPose, a Machine Learning framework from Carnegie Mellon University (CMU, USA). As the name implies, it is meant to recognize the poses of any number of persons in video images. It does not track persons, but re-recognizes them frame by frame. The output is 2D positions of joints in a rather simplified skeleton, together with confidence values telling you how much you can rely on each of the positions. The latest version of this library allows more detailed recognizers for the face and the hands to be added (at the cost of even longer processing times). Our experiments with a small set of data already processed with OpenPose are rather promising. Using our knowledge about the sign language videos being processed, we can even reconstruct 3D information from the OpenPose data. The idea is to build a secondlayer supervised learning system on top of OpenPose, with the training set being the annotated corpus data. We are confident that this system will scale a lot better than previous approaches but of course have to wait for the data processing to be completed until we can start formal evaluation. Despite this progress we should not forget that it is also necessary to consider alternatives to Search by example, depending on what the user wants to achieve. Let us assume the user has seen a sign s/he has does not know. Unfortunately, the user could not see the sign in all detail (or has forgotten some detail). There are two 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 133

132 Session 3 approaches to cover this scenario: In the first one, the user completes the sign by adding default default data and signs that to the camera and then selects which parameters of the exemplar are to be ignored. Alternatively, there are user interfaces to tell the system which handshape you are looking for, what would be the location, what would be the movement trajectory. So the user would only input the detail s/he is sure about, using more specific user interfaces potentially allowing the system to recognize more detail than in a holistic approach. E.g. there are several systems being developed using Leap Motion cameras to recognize the handshape in more detail than probably possible with current webcams capturing the upper half of the body. In DictaSign there was also a prototype to specify the location. AVATAR TECHNOLOGY: HAVE THE DICTIONARY SIGN TO YOU! I am sure everyone in the audience has seen signing avatars before. In fact, there have been so many signing avatar projects around the world that is surprising that most are miles away from being able to sign in an understandable and believable way except when fine-tuned by massive amounts of human resources. While it might be possible to hand-tune every headword signed, we are also interested in having the avatar sign examples from the corpus. Clearly, in that case hand-tuning is no longer an option. Automatic sign generation either relies on motion capture data or on synthetic generation, such as from HamNoSys phonetic data. The first approach requires special equipment and lots of post-processing time, and it remains unclear how to best glue signing mocap data from one single or even several people together. Synthetic generation works from phonetic data available in the lexical database anyway, so it kind of comes for free. However, signing from notation often gives the impression of a robot signing. In my eyes, hybrid approaches are most promising: They use normalized mocap data where available, but fall back to synthetic generation where necessary. Corpus data such as typical or actual durations can be used (and hopefully OpenPose data in the future) allow to come closer to human signing. We are cooperating with the Paula Project at DePaul University in Chicago to enable the Paula avatar to sign DGS from hybrid data. At this point, you might ask why we want to have an avatar sign example sentences from the corpus if we have good-quality videos available. In fact, we are proud to have lots of nice signing data from the public corpus that can be used as examples in the dictionary. However, given the limited size of the corpus (compared at least to recent written language corpora), we would like to also show examples not part of the public corpus for good reasons. (This happens quite often in our corpus due to the high percentage of life stories included.) If it is just reference to a third person, we might also use sign models copying the informant s signing in the studio and replacing that third-person reference. But there are cases where the sign models feel uneasy about exactly repeating someone else s signing as they would become visible in the internet with signing that they consider not to be their own. In that case, the avatar is a perfect anonymisation strategy. Collocation patterns is another potential area of application: It is highly improbable that you find exemplars of 국립국어원국제학술대회 보이는언어의기록, 수어사전

133 Sign Language Dictionaries of Tomorrow Thomas Hanke all pattern elements signed by one single person. Having the pattern signed by different persons might irritate the user. So once again you have to rely on either sign models or avatars. For dictionaries using invented examples instead of corpus examples, avatars might allow the dictionary maker to personalize the dictionary by signing dialectal variants fitting to where the user is from. Fifteen years ago, bandwidth saving from using an avatar instead of digital video would have been the major argument in favour of avatars. While this no longer is a point for many communities, even in the mobile world, it might still be a valid point for others. Definitely, dictionary makers should have the choice. VISUAL EXPLORATION OF A DICTIONARY Even if it will become feasible to implement very precise search for sign form, it remains the dictionary maker s dream that the users explore the dictionary s content beyond looking up single entries. In the days of printed dictionaries, when you had to browse through the pages to find the entry you were looking for, you automatically scanned several articles and might be caught by some. Nowadays, the dictionary maker has to explicitly offer cross references to other entries or provide a universe of entries inviting the user to explore it. The ASL-LEX project has a particularly attractive visualisation of its contents as a nicely laid-out graph. Unfortunately, their approach is not the right one for every size of a dictionary. It looks much less attractive when scaled to the dictionary size we want to end up with, so the problem probably needs to be addressed separately for each dictionary project discussed today. Our DGS corpus reflects the high degree of variation found in DGS, both in the regional and age dimension. As it turns out through the feedback we get, regional variation is of high interest to the DGS community, and people are very satisfied with the data we can provide as maps. Here we use standard web technologies, in this case d3js, to generate nicely drawn maps. CONCLUDING REMARKS Now, before closing, let us speculate about what sign language dictionaries of the day after tomorrow might look like: You remember Jack-in-the-box? Apply that idea to sign language: Whenever you need an avatar to show signs, you get it right where you need it by opening a box. But instead of Jack out of the box, it is a holographic display of a virtual (or even real) signer out of the box. And it need not be a box: Just put your Samsung 20h or iphone 20xh face-down on the table to start the holographic transmitter Think about interpreters projected into the room where you need them, next to the person you want to interact with. Think about the virtual museum guide side by side with the objects it is presenting. If holographic displays sound too much Star Wars to you, just replace them with Augmented Reality (AR). This is upcoming technology 2018 NIKL Conference Records of Visible Language: Sign Language Dictionary 135

134 Session 3 allowing apps to project computer-generated contents onto the real world images you see through special glasses. Of course we want the virtual museum guide in such an AR scenario not only being able to talk you through the exposition in speech, but also to sign you through the exposition! And then, as you do not understand a sign being used by the guide, you simply eyeblink to it in order to have the corresponding dictionary page opened, very much as you may be used to when reading a Kindle book in a foreign language today. By the way: Did I say dictionary page? Dictionary pages somehow survived the transition from paper volumes to hypertext. Maybe we have to get rid of the concept when thinking about dictionaries for the day after tomorrow. In a virtual 3D world, a dictionary entry might actually look more like an apartment consisting of several rooms and lots of doors to the neighbours. Thank you very much for your attention! 국립국어원국제학술대회 보이는언어의기록, 수어사전

135

136

137

138

139

140 2018 국제학술대회보이는언어의기록, 수어사전 Records of Visible Language: Sign Language Dictionary 발행인 기획 편집 소강춘 정호성 서민경 발행일 발행처 디자인 국립국어원서울시강서구금낭화로 유니원커뮤니케이션즈 인쇄 상상그라픽스 저작권법보호를받는간행물이므로어떠한형태로든무단전재및복제를금합니다.