| _version_ | 1866902182765264896 |
|---|---|
| author | Nedunchezhian, Arun |
| author_facet | Nedunchezhian, Arun |
| contents | <p>A novel evaluation framework for machine translation quality assessment specifically tailored to short-chat messages addresses critical gaps in current practice. Traditional metrics like BLEU and METEOR demonstrate significantly reduced reliability when applied to brief conversational exchanges that dominate contemporary digital communication platforms. The framework integrates both semantic and syntactic similarity measures, leveraging comprehensive multilingual dictionaries and sophisticated part-of-speech tagging techniques to evaluate translation quality beyond surface-level lexical matching. Semantic evaluation extracts representations for each word in source and translated messages, computes similarity scores between corresponding terms, and employs word sense disambiguation to address polysemy challenges. Syntactic evaluation acknowledges that different languages follow distinct grammatical patterns while potentially conveying identical meaning, focusing on the preservation of core grammatical components regardless of sequential order. When combined through adaptive weighting mechanisms that adjust based on specific language pairs and conversational contexts, this integrated approach demonstrates a substantially higher correlation with human judgments than traditional metrics applied in isolation, particularly for typologically distant languages and culturally embedded expressions. The methodology provides a more linguistically informed foundation for evaluating machine translation performance in increasingly prevalent short-message communication environments.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_17200012 |
| institution | Zenodo |
| language | eng |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | A novel evaluation methodology for machine translation of short chat messages Nedunchezhian, Arun Machine Translation Chat Message Translation Semantic Similarity Syntactic Evaluation Cross-Linguistic Communication Translation Quality Assessment <p>A novel evaluation framework for machine translation quality assessment specifically tailored to short-chat messages addresses critical gaps in current practice. Traditional metrics like BLEU and METEOR demonstrate significantly reduced reliability when applied to brief conversational exchanges that dominate contemporary digital communication platforms. The framework integrates both semantic and syntactic similarity measures, leveraging comprehensive multilingual dictionaries and sophisticated part-of-speech tagging techniques to evaluate translation quality beyond surface-level lexical matching. Semantic evaluation extracts representations for each word in source and translated messages, computes similarity scores between corresponding terms, and employs word sense disambiguation to address polysemy challenges. Syntactic evaluation acknowledges that different languages follow distinct grammatical patterns while potentially conveying identical meaning, focusing on the preservation of core grammatical components regardless of sequential order. When combined through adaptive weighting mechanisms that adjust based on specific language pairs and conversational contexts, this integrated approach demonstrates a substantially higher correlation with human judgments than traditional metrics applied in isolation, particularly for typologically distant languages and culturally embedded expressions. The methodology provides a more linguistically informed foundation for evaluating machine translation performance in increasingly prevalent short-message communication environments.</p> |
| title | A novel evaluation methodology for machine translation of short chat messages |
| topic | Machine Translation Chat Message Translation Semantic Similarity Syntactic Evaluation Cross-Linguistic Communication Translation Quality Assessment |
| url | https://doi.org/10.5281/zenodo.17200012 |