Abstract
Bidirectional long short-term memory (bi-LSTM) networks have recently proven successful for various NLP sequence modeling tasks, but little is known about their reliance to input representations, target languages, data set size, and label noise. We address these issues and evaluate bi-LSTMs with word, character, and unicode byte embeddings for POS tagging. We compare bi-LSTMs to traditional POS taggers across languages and data sizes. We also present a novel bi-LSTM model, which combines the POS tagging loss function with an auxiliary loss function that accounts for rare words. The model obtains state-of-the-art performance across 22 languages, and works especially well for morphologically complex languages. Our analysis suggests that bi-LSTMs are less sensitive to training data size and label corruptions (at small noise levels) than previously assumed.
Original language | English |
---|---|
Title of host publication | 54th Annual Meeting of the Association for Computational Linguistics, ACL 2016 - Short Papers |
Publisher | Association for Computational Linguistics (ACL) |
Pages | 412-418 |
Number of pages | 7 |
ISBN (Electronic) | 9781510827592 |
DOIs | |
State | Published - 2016 |
Event | 54th Annual Meeting of the Association for Computational Linguistics, ACL 2016 - Berlin, Germany Duration: 7 Aug 2016 → 12 Aug 2016 |
Publication series
Name | 54th Annual Meeting of the Association for Computational Linguistics, ACL 2016 - Short Papers |
---|
Conference
Conference | 54th Annual Meeting of the Association for Computational Linguistics, ACL 2016 |
---|---|
Country/Territory | Germany |
City | Berlin |
Period | 7/08/16 → 12/08/16 |
Bibliographical note
Publisher Copyright:© 2016 Association for Computational Linguistics.