Multilingual Indirect Answer Interpretation
Multilingual NLP for understanding indirect answers to yes-no questions.
Interpreting Indirect Answers to Yes-No Questions in Multiple Languages
People often answer yes-no questions indirectly rather than explicitly saying “yes” or “no.” While humans can usually infer the intended meaning, this remains challenging for language understanding systems, particularly across languages.
In this work, we study the interpretation of indirect answers to yes-no questions across multiple languages.
We develop multilingual benchmarks covering eight languages and investigate how models trained in one or more languages generalize across languages, including lower-resource settings.
Key Findings
- Multilingual fine-tuning improves the interpretation of indirect answers across languages.
- Cross-lingual transfer is particularly useful when limited training data is available in the target language.
- The results demonstrate both the potential and limitations of multilingual models for pragmatic language understanding.
This work was published in Findings of EMNLP 2023.
Paper · PDF · Code & Data