| Popis: |
We discuss approaches for improving bilingual lexicon coverage by automatically suggesting translations for Out-Of-Vocabulary (OOV) terms, employing existing validated bilingual lexicon entries. Resource poor languages such as Hindi, Konkani and Sanskrit characterized by highly inflectional morphology were employed in our experiments. Known surface translations are mined for morphological similarities and bilingual morphemes thus learnt are used in suggesting word-word and phrase translations. Also, word-word translations are generated for the language pair Hindi-Sanskrit by pivoting bilingual stems and suffixes, with Konkani and English as bridge language, former a morphologically rich language while latter morphologically poor. |
