sarawak language technology (salt) research group salt initiatives: preservation and maintenance of...
Post on 31-Mar-2015
228 Views
Preview:
TRANSCRIPT
Sarawak Language Technology (SaLT)Research Group
SaLT Initiatives:Preservation and Maintenance of Sarawak Languages Faculty of Computer Science and Faculty of Computer Science and
Information TechnologyInformation Technology
Universiti Malaysia SarawakUniversiti Malaysia SarawakAssociate Professor Alvin W. YeoAssociate Professor Alvin W. Yeo
Overview
• Languages in Sarawak• Maintenance and Revitalisation: Holistic Approach
– Sarawak Language Technology (SaLT) Research Group
• SaLT Projects– Borneo Corpus Management System (BCMS)– Iban-English Machine Translation
• TRanslation IBan-English (TRIBE)
– Multimodal-INTegration (MINT) of Sketch and Melanau Daro-Matu Speech in Spatial Queries
– Speech Language Dialog Systems (SLaDS)– Development of Language Tools
• Current findings
Where are we?
East Malaysia> Sarawak> Kuching
Kuching
Introduction (cont’d)
• Sarawak is a state rich in culture. – 27 ethnic groups in Sarawak (STB, 2005), each with
its own culture and language. – Sarawak has 46 living languages
and 1 extinct; according to the
Ethnologue (Gordon, 2005) – Each ethnic group may have
different languages– Sarawak Dewan Bahasa dan Pustaka
• 63 known languages in Sarawak
Rationale
Population No. of languages Cumulative no. of languages
Cumulative (%)
1– 100 4 4 9%
101 – 500 8 12 27%
501 – 1000 4 16 36%
1001 – 5000 18 34 76%
5001 – 10,000 4 38 84%
10,001 – 50,000 6 44 98%
50,001 – 100,000 0 44 98%
100,001 - 1 45 100%
Extinct 1 1
No data available
1 1
Cumulative language and number of speakers (Ethnologue,2005)
Problem
• World’s linguistic and cultural diversity is under threat. – Many minority languages are on
the brink of extinction.
• Minority language communities– Further disadvantaged economically and socially. – Dominant languages– Exogamy
• Revitalizing minority languages can bring economic and social benefits as well as cultural benefits.
Holistic Approach: Framework for Language Revitalization and Maintenance
People
Preservation of Culture
Applications
Internet: Online Presence
Software applications and operating systems
Hardware: Input devices: keyboards, tablets/pen/stylus
Supporting Technologies
StakeholdersStakeholders
Web techno-logies: Java, Flash
Methodologies: engaging communities; development lifecycles
Computing Technologies:Natural Language Processing, Image Processing, Speech Recognition and Generation
Community/civil society
Research institutions
Government agencies
NGOs
Industry
Trainers
Translators
Linguists
IT spec.
Comp. Scientists
Researchers
Social Scientists
Communi-ty readiness: ICT literacy
Ethnic group organisations
…
Sarawak Language Technologies (SaLT) Research Group
SaLT
• Role of technology in language maintenance and revitalisation• On revitalising and maintaining the existing conventional
languages by building corpora, conducting research and developing tools for Sarawak Ethnic Languages.
Sarawak Language Technologies (SaLT) Research Group covers
• Codification of the ethnic languages – Creation of corpora of the various languages in Sarawak
• Research in computational linguistics projects– which involves languages and peoples of Sarawak
• Development of tools: word processors, spell
checkers
Language Technology
• Understanding and explication of language phenomena in a– computationally tractable form, resulting in – techniques for interchanging various linguistic
forms • speech, text, morphology, syntax, semantics/meaning,
discourse, knowledge,
– thus leading to the creation and development of intelligent applications involving language.
Levels of Technology
INPUT (corpus)
APPLICATION (machine translation,
multimodal spatial application)
PROCESSOR (tagger, parser,
multimodal integration)
Lexicographer/Linguist/ comp. scientist
Linguist/ comp. scientist
General and conceptual dictionary
Specialists Needed
• Lexicographers• Computer scientists DBA, SE & N/W (data
maintenance & grid)• Linguists• Information Scientists• Psychologists• Anthropologists• Computational Linguistics Natural Language
Processing
Current Projects
Current Projects (cont’d)
Roadmap for SaLT
Advisors and Organisations Involved
No Name Expertise Organisation
1 Prof. Zaharin Yusoff Computational Linguistics (CL) & Natural Language Proc. (NLP)
MMU
2. Prof. Ahmad Zaki Abu Bakar CL & NLP UTM
3. AP Dr Normaziah Abdul Aziz NLP & Artificial Intelligence UIAM
4. Prof. Dr. Tang Enya Kong CL & NLP MMU
5 Dr. Bali Ranaivo NLP & CL MMU
6. Prof. Dr. Zuraidah Mohd. Don Linguistics UM
7. Dr. Gerry Knowles Phonetics and Phonology MIQUEST Worldwide Sdn
Bhd 8. Professor Dr. Peter Songan Community development UNIMAS
Collaborators
Organisations Involved1. Tun Jugah Foundation2. Dewan Bahasa dan
Pustaka (Sarawak Branch)3. Melanau Association 4. Dayak Bidayuh National
Association5. Sarawak Museum6. Pustaka Negeri Sarawak7. Majlis Adat Istiadat
Universities Involved1. UNIMAS (FCSIT, FCSHD,
FSS, CLS) 2. Multimedia University3. Universiti Teknologi Malaysia4. Universiti Islam Antarabangsa
Malaysia5. Universiti Sains Malaysia6. Universiti Malaya7. Localisation Research Centre,
University of Limerick, Ireland8. University of Waikato, New
Zealand
Team members
a. Staff FCSIT
• AP Dr Alvin Yeo Wee (Head)• AP Dr. Narayanan K.• Dr Edwin Mit• Suhaila Saee• Sarah Flora Samson• Nurfauza Jali• Suriati Khartini Jali• Sy. Fazlin Seyed Fadzir• Lee Jun Choi
FCSHD• Dr. Ng Giap Weng• D’oria Islamiah• Wan Norizan
CLS• Dr. Ting Su Hie• Salbia Hassan• Yvonne Michelle Campbell
Team members (cont’d)
b. Research Assistants1. Beatrice Chin (FCSIT)2. Teh Lee Na (FCSIT)3. Jennifer Wilfred (FCSIT)4. Lai Nyong Fock (FCSIT)5. Mohd. Hanafiah Semuni (FCSHD)6. Loh Chee Wyai (FCSIT) 7. Ang Siaw Tiong (FCSIT)
c. StudentsLevel No. of Students
Post-graduate PhD 2
Master by Research 6
Master by Coursework 5
Undergraduate 22
Total 35
Borneo Corpus Management System (BCMS)
• Problem/Background: – Currently there is no existing corpus management system to manage
corpora available in minority languages of Sarawak
• Solution: – Build a system that is able to manage and maintain the corpora
• Objectives: – To design an easy and usable Corpus Processing Toolkit for
researchers– Integrate the various tools together in one single platform
• Current Status: – Working on the Morphological Analysers and Spell Checkers
Corpus Manager (After processing)
Editable Content
Used to highlight the extracted information
in the content
File tree that display the processed files. The file is
stored in the folder based on category
Original Content Processed Content
Corpus Analyser: Sentence Splitter
The output is each sentence of current document
Iban-Corpus Development
• Problem/Background – Indigenous languages in Sarawak are slowly dying out due to:– One way to stem this “extinction” of languages:
• Provide more local content – but how??
• Solution– Translate English documents to documents in minority languages– MT is needed to facilitates and accelerates the translation process
• Objectives– Identify a methodology that can be used to translate English to minority
languages, by taking Iban as a case study
• Current Status – Built Iban corpus with 23,833 words with 3,831 distinct words– Constructed bilingual lexicon with 1,688 words with 1,192 distinct words
Iban-English Machine Translation
• Problem/Background – Traditional knowledge (TK) is tacit knowledge; generally not
stored and known only by the older generation, who speaks little English
– TK is very important. It needs to be preserved and protected.– Machine Translation (MT) can help to preserve TK– Translate available resources into English so that it is accessible
by all, e.g. researchers (social scientists) and younger generation– However, translation of closely related languages is easier
• Solution– Translate TK documents to English through a closely related
language as pivot language– Case study: Iban as source language, Malay as pivot language
and English as target language
• Objectives– To demonstrate that the performance of translation
through a pivot language is comparable with performance of direct translation
• Realise benefits (efficiency) of translating multiple “similar” languages through a common pivot language
• Current Status – Building of Iban corpus and lexicon– Linguistic comparison on Iban and Malay language
Multimodal Integration: Preamble
User sketching on the Wacom tablet with CogSketch sketch interface describing a place.
Dragon Naturally Speaking software for capturing thespeech with a microphone.
Multimodal Integration of Sketch and Melanau Daro-Matu Speech in Spatial Queries (MINT)
• Problem/Background– English: main communication medium– Language is unique and distinct
• Individual uses different languages may have different approaches in conceptualizing, communicating, reasoning, expressing their thoughts
– Translation is not sufficient enough – Building the entire system for certain targeted speakers is
time consuming
• Solution– Internationalisation (i18n)– Localisation (l10n)
• Objectives– Integrate Melanau Daro-Matu speech and sketch
(image) modalities– Identify the interaction patterns of Melanau users.– Identify the similarities and difference of English,
Malay and Melanau (extending to Iban as well)– Localise architecture and representation of multimodal
integration in Melanau Daro-Matu, and other languages
Input Capturing
Input Interpretation
Modalities Representation Speech Representation
Sketch Interpretation
Sketch
Speech Interpretation
Part-Of-Speech Tagging
Language-Dependent ComponentsLanguage-Dependent Components
Tokenization
Tagging using trained corpusTagging using trained corpus
Tagging corrections acquired from templates
Tagging corrections acquired from templates
Lexicon required
Grammar rules required
Annotated Text
Spatial information retrieval
Speech
Sketch Representation
Modalities Integration Sketch and Speech Integration
Database Searching
Sentence Splitter
Transcription
Spoken Language Dialogue System (SLaDS)
• Problem/Background– Spoken language system (SLS) has become an ever-increasing human-
system interface. – Many studies have been conducted by foreign researchers to unravel
the challenge in the design of spoken language system. – This study focuses on the design and development of spoken language
dialogue system within the context of Malaysian user.
• Solution– The project is performed by conducting a simulation test of the real SLS
system with local user. – The system is then evaluated by adopting the Wizard of Oz method with
the objectives to determine its efficacy. – The result of this testing will be useful for the future development of
Malaysian SLS.
• Objectives– To investigate the spoken language and interaction design,
and its employment in the development of Spoken Language Dialog Systems
– To determine the efficacy of imported usability evaluation techniques applied in the Spoken Language Dialogue Systems
– Identify speech patterns to develop a predictive model for speech recognition
• Current Status– To date, the study is already in its testing stage to capture the
dialogue content. – Respondent is prompt to interact with the system. – The dialogue from the interaction will be taped, transcribed and
analysed.
Wizard’s Control Panel
User’s view
SCREENSHOTS VIDEO
Video showing interaction sample;
Research Projects: Fundamental Research Grant
• Minority Languages Online (MiLO): Preserving Cultures by Mobilising Minority Languages (of Sarawak) Online. (completed 30 June 2007) – Continued with CLS, Univ. of Waikato– Wikipedia approach to development of Bidayuh lexicon
• Bario Lakuh Digital Library (completed) – Recordings of Kelabit songs– Transcibed, translated – With audio and video
e-Vocabulary for Sarawak Malay
• Problems: Language endangerment• Vocabulary of Sarawak Malay (Original source) • Main source: Vocabulary book written by W.S.B.BUCK
from Bau, which was published by Sarawak Civil Service on 11th May, 1932.
• Total of word entries: 1026 words
Background• One of the most widely used computer application
nowadays is the word processor.• Open Source Software (OSS): can used, studied, and
redistributed in modified or unmodified form without restriction
Solution/Objectives• AbiWord (comprehensive word processor) to be
localised• To identify the processes of translation of computing
terminology
AbiWord in Local Languages
Current Status:
Task Progress
Data collection:Template Ongoing
Interface:ToolbarMenuSubmenuIconTooltipsOperation
Completed
Completed
Ongoing
Ongoing
Ongoing
Running
Screen shots
Interface
Example of Menu Panel
Current Findings: Challenges
• Resources of some languages available– Generally lacking; data collection very challenging
• Writing systems and grammar rules do not exist
• Lack of human resources– Fluent in the (untainted) form (translating, POS
tagging)
Current Findings: Bright future
• Community Awareness– Associations of ethnic groups aware of need– Advanced in age interested, younger generation not so
• Protocol followed– Upper management support required to “open doors”
• Local researchers are interested– Colleagues & students
• Machine translation, speech to text, text to speech
• Development of speech corpus
Multi-ethnic Group
Concluding Remarks
• Decreasing number of speakers of languages in Sarawak• Maintenance and Revitalisation: Holistic Approach
– Sarawak Language Technology (SaLT) Research Group
• SaLT Projects– Machine translation, multimodal integration, speech language dialog
system, corpus management systems, online dictionaries/repositories, digital libraries
• Challenges: community involvement and data collection and analysis
• Silver lining: committed NGOs and researchers• Internationalisation and localisation approach
Acknowledgements
• Institutional support from – Universiti Malaysia Sarawak– Jugah Foundation, Melanau Association, Dewan
Bahasa dan Pustaka (Sarawak Branch), Majlis Adat Istiadat, Dayak Bidayuh National Association
• Financial Support grants – UNIMAS Fundamental Research Grant Scheme– Federal Ministry of Science, Technology and Innovation
Science Fund Grant Scheme (01-09-SF0028, SF0029, SF0030)
Fifth International Cyberspace Conference on Ergonomics (CybErg 2008)– Theme: Local knowledge, Global Applications– Special Discussion on Maintenance and Preservation of
Languages
– On-going 15 Sept – 15 Oct 2008– Free Registration– http://www.cyberg08.org/forum
Sixth International
Conference on IT In Asia (CITA’09)•Theme: “Enabling technologies for Knowledge-driven Society: People-Powered Systems”•Tracks on Computational Linguistics, Human Computer Interaction, Software Engineering•Kuching, Malaysia, 6- 9 July 2009; Rainforest Music Festival
Thank YouTerima Kasih
Jian Kenin
top related