ThamizhiFST: A Morphological Analyser and Generator for Tamil Verbs
| Autor: | Gihan Dias, Miriam Butt, K. Sarveswaran |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Agglutinative language
Grammar Lexical functional grammar Computer science business.industry media_common.quotation_subject 020206 networking & telecommunications Tamil grammar Verb 02 engineering and technology computer.software_genre language.human_language Morpheme Noun Tamil 0202 electrical engineering electronic engineering information engineering language 020201 artificial intelligence & image processing Artificial intelligence ddc:400 business computer Natural language processing Generator (mathematics) media_common |
| Zdroj: | 2018 3rd International Conference on Information Technology Research (ICITR). |
| DOI: | 10.1109/icitr.2018.8736139 |
| Popis: | ThamizhiFST is a Morphological Analyser and Generator (MAG) for Tamil. It was developed to extend the coverage of the computational Tamil grammar being developed using Lexical Functional Grammar (LFG). ThamizhiFST covers the simple verbs in Tamil as an initial step. A Finite State Transducer (FST) approach was used to develop the MAG and it was implemented using the FOMA Open Source Software. Since morphological rules are of a finite nature and represent a known quantity, a rule-based approach like FST is more appropriate than possible machine learning alternatives, especially with respect to achieving reliably good accuracy that is required for computational grammar development. A set of 3250 Tamil verb lemmas from 13 paradigms together with their 260 conjugation forms were used in the construction of ThamizhiFST. Further, a set of 27 labels were used to mark the morphosyntactic information of the verbs. The whole system was developed as a three-layer web-based system to tackle the issues arising when processing an agglutinative language like Tamil and to ensure its extendability. Unlike other existing MAGs, ThamizhiFST also provides the morpheme corresponding to each morphosyntactic label and marks morpheme boundaries. An evaluation shows that ThamizhiFST has an f-measure of 0.97 for simple verbs. Future and current work include work on extending the system to cover more verbs and nouns and make it generally available. published |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|