Feature Detection for Minority Languages

1
Feature Detection for Minority Languages

• Katharina Probst,
• Ralf Brown, Jaime Carbonell, Alon Lavie, Lori
  Levin, Erik Peterson
• Language Technologies Institute
• Carnegie Mellon University
• Pittsburgh, PA, U.S.A.

2
Problem Statement

• Scarce resources for minority languages
• Bilingual text
• Monolingual text
• Target language grammar
• Due to scarce resources statistical and
  example-based methods will likely not perform
  well
• Our approach a system that elicits necessary
  information about the target language from a
  bilingual informant who is not trained in
  linguistics

3
Related Work

• Project Boas (Sherematyeva, Nirenburg 2000)
• Diplomat project (Frederking, Rudnicky, Hogan,
  Lenzo 2001)
• Twisted-pair grammar (Jones, Douglas R. Havrilla
  1998)
• Translation Templates from Examples (Güvenir,
  Cicekli 1998)
• Inversion transduction grammar (Wu 1997)

4
Approach

• Feature detection
• Prediction of transfer rules
• Seeded version space learning to refine transfer
  rules (and their coverage)
• Run-time module

5
Architecture Diagram
SL Input
Run-Time Module
Learning Module
SL Parser
EBMT Engine
Elicitation Process
SVS Learning Process
Transfer Rules
Transfer Engine
TL Generator
User
Unifier Module
TL Output
6
Feature Detection

• Corpus of well-chosen sentences, aimed at
  covering major linguistic phenomena
• Organized in minimal pairs
• Pruning of corpus is based on Implicational
  Universals
• Minimize training time
• Minimize informant frustration

7
Example Elicitation Corpus

• You (John) are falling. 2nd person masc.
  subject, present tense
• Tu (Juan) estás cayendo
• Eimi(Kuan) tranmekeymi
• You (Mary) are falling. 2nd person fem.
  subject, present tense
• Tu (María) estás cayendo
• Eimi tranmekeymi (Maria)
• You (Mary) fell. 2nd person fem. subject, past
  tense
• Tu (María) caiste
• Eymi tranimi (Maria)
•  

8
Feature Detection

• Corpus of well-chosen sentences, aimed at
  covering major linguistic phenomena
• Organized in minimal pairs
• Pruning of corpus is based on Implicational
  Universals
• Minimize training time
• Minimize informant frustration

9
Elicitation of DataSpanish Mapudungun Example
10
Organization of Tests
Dual
Plural
Paucal
Diagnostic Tests
Obj-V Agr
Subj-V Agr


11
Conclusions

• Automated learning of transfer rules
• Largely language-independent framework for
  building a rule-based MT system
• Based on the study of a diverse set of languages
• Allows applying MT to low-density languages, as
  it works with little (albeit well-chosen) data
• New paradigm of MT research collaboration with
  indigenous communities

12
References

• Frederking, Robert A.Rudnicky C. Hogan K.
  Lenzo. (2001). Interactive Speech Translation in
  the DIPLOMAT Project. In Machine Translation
  Journal, Special issue on Spoken Language
  Translation.
• Guvenir, H.A. H. Altay I. Cicekli. (1998).
  Learning Translation Templates from Examples.
  Information Systems, 23(6)353--363.
• Jones, Douglas R. Havrilla. (1998). Twisted Pair
  Grammar Support for Rapid Development of Machine
  Translation for Low Density Languages. In
  Proceedings of AMTA.
• Sherematyeva, Svetlana S. Nirenburg.
  (2000).Towards a Universal Tool For NLP Resource
  Acquisition. In Proceedings of The Second
  International Conference on Language Resources
  and Evaluation (Greece, Athens, May 31 -June 3,
  2000).
• Wu, D. (1997) Stochastic inversion transduction
  grammars and bilingual parsing of parallel
  corpora. Computational Linguistics, 23 (3),
  377-403.