A Cross-lingual Comparison of Human and Model Relative Word Importance
Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Documents
- Fulltext
Final published version, 600 KB, PDF document
Relative word importance is a key metric for natural language processing. In this work, we compare human and model relative word importance to investigate if pretrained neural language models focus on the same words as humans cross-lingually. We perform an extensive study using several importance metrics (gradient-based saliency and attention-based) in monolingual and multilingual models, including eye-tracking corpora from four languages (German, Dutch, English, and Russian). We find that gradient-based saliency, first-layer attention, and attention flow correlate strongly with human eye-tracking data across all four languages. We further analyze the role of word length and word frequency in determining relative importance and find that it strongly correlates with length and frequency, however, the mechanisms behind these non-linear relations remain elusive. We obtain a cross-lingual approximation of the similarity between human and computational language processing and insights into the usability of several importance metrics.
Original language | English |
---|---|
Title of host publication | Proceedings of the 2022 CLASP Conference on (Dis)embodiment |
Number of pages | 13 |
Publisher | Association for Computational Linguistics (ACL) |
Publication date | 2022 |
Pages | 11-23 |
Publication status | Published - 2022 |
Event | 2022 CLASP Conference on (Dis)embodiment - Gothenburg / Online, Sweden Duration: 15 Sep 2022 → 16 Sep 2022 |
Conference
Conference | 2022 CLASP Conference on (Dis)embodiment |
---|---|
Land | Sweden |
By | Gothenburg / Online |
Periode | 15/09/2022 → 16/09/2022 |
Links
- https://aclanthology.org/2022.clasp-1.2
Final published version
ID: 331507496