Elamur Research⌕
☰
home / papers / 2017 / Temporal Information Extraction for Question Answering Using Syntactic Dependencies in an LSTM-based Architecture

Temporal Information Extraction for Question Answering Using Syntactic Dependencies in an LSTM-based Architecture

Paper · 2017 · Yuanliang Meng, Anna Rumshisky and Alexey Romanov

Summary

The paper addresses temporal information extraction: detecting events and time expressions in text and working out their chronological order. Systems are judged mainly by how useful their output is for answering temporal questions. The authors build a pipeline called TEA. HeidelTime tags TIMEX expressions, an LSTM classifier marks single-word events, and NewsReader supplies tokenization, POS tags and dependency parses. Temporal links (TLINKs) come from three LSTM models with an identical design, covering pairs inside one sentence, pairs in neighbouring sentences, and links to the document creation time. Each model reads only the word embeddings along the shortest dependency path between two entities, split into two branches that meet at their lowest common ancestor. Rules handle TIMEX–TIMEX pairs. Every pair is also classified in reversed order ("double-checking"), and a global pruning step removes contradictory links. On QA-TempEval (SemEval 2015 Task 5), TEA clearly beats the 2015 winner on news and Wikipedia texts, though not on blogs. It also matches or exceeds the best prior scores on TimeBank-Dense. This suggests that a light, uniform neural design can rival heavily engineered temporal systems.

Key points

  • Data: the QA-TempEval training set has 276 TimeML files, mostly news. A subset of 25 files with 79 questions served for validation. The test set contains 28 unannotated documents (10 news, 10 Wikipedia, 8 blogs) with 294 yes/no questions.
  • Across all test genres, TEA reaches F1 0.47 (precision 0.59, recall 0.39). The top 2015 system, hlt-fbk-ev2-trel2, scores 0.37, or 0.40 with the TREFL reasoner. Applying TREFL to TEA slightly lowers its score to 0.46.
  • On TimeBank-Dense (22/5/9 document split), a single uniform configuration gets F1 0.505. Tuning each component separately gets 0.519, versus the previous best of 0.511.
  • The event detector looks at a 9-token window of embeddings plus 4 binary features of the target token (main verb, predicate, verb, noun). It does not predict event attributes.
  • The TLINK models use no POS tags, grammatical relation labels or WordNet features, only word embeddings. Training data is augmented by adding each pair in flipped order with the inverse label.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of Massachusetts LowellLowell · United States15 works · 3,942 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Question answering · 441 worksInformation extraction · 167 works

Cite

Cite as Yuanliang Meng, Anna Rumshisky and Alexey Romanov. 2017. Temporal Information Extraction for Question Answering Using Syntactic Dependencies in an LSTM-based Architecture. Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/d17-1092 · arXiv 1703.05851

@inproceedings{meng2017temporal,
  title = {{Temporal Information Extraction for Question Answering Using Syntactic Dependencies in an LSTM-based Architecture}},
  author = {Yuanliang Meng, Anna Rumshisky and Alexey Romanov},
  booktitle = {Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing},
  year = {2017},
  doi = {10.18653/v1/d17-1092},
  eprint = {1703.05851},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.18653/v1/d17-1092}
}

References

23 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Paramita Mirza and Anne-Lyse Minard. 2015. Hlt-fbk: a complete temporal processing system for qa tempeval. In Proc. of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 801– 805. Association for Computational Linguistics. find
  2. Rodrigo Agerri, Josu Bermudez, and German Rigau. 2014. Ixa pipeline: Efficient and ready to use multilingual nlp tools. In Proc. of the 9th Language Resources and Evaluation Conference (LREC2014), pages 26–31. find
  3. James Pustejovsky, José Castaño, Robert Ingria, Roser Saurı̀, Robert Gaizauskas, Andrea Setzer, and Graham Katz. 2003. TimeML: Robust specification of event and temporal expressions in text. In in Fifth International Workshop on Computational Semantics (IWCS-5). find
  4. Steven Bethard, Leon Derczynski, James Pustejovsky, and Marc Verhagen. 2015. Semeval-2015 task 6: Clinical tempeval. In Proc. of the 9th International Workshop on Semantic Evaluation (SemEval 2015). Association for Computational Linguistics. find
  5. Jannik Strötgen and Michael Gertz. 2013. Multilingual and cross-domain temporal tagging. Language Resources and Evaluation, 47(2):269–298. find
  6. Nathanael Chambers, Taylor Cassidy, Bill McDowell, and Steven Bethard. 2014. Dense event ordering with a multi-pass architecture. Transactions of the Association for Computational Linguistics, 2:273– 284. find
  7. Weiyi Sun. 2014. Time Well Tell: Temporal Reasoning in Clinical Narratives. PhD dissertation. Department of Informatics, University at Albany, SUNY. find
  8. Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural Computation, 9(8):1735–1780. find
  9. Weiyi Sun, Anna Rumshisky, and Ozlem Uzuner. 2013. Evaluating temporal relations in clinical text: 2012 i2b2 challenge. Journal of the American Medical Informatics Association, 20(5):806–813. find
  10. Richard Karp. 1972. Reducibility among combinatorial problems. In Complexity of Computer Computations, Proc. Sympos., pages 85–103. find
  11. Naushad UzZaman, Hector Llorens, James Allen, Leon Derczynski, Marc Verhagen, and James Pustejovsky. 2012. Tempeval-3: Evaluating events, time expressions, and temporal relations. arXiv preprint arXiv:1206.5333. find
  12. Hector Llorens, Nathanael Chambers, Naushad UzZaman, Nasrin Mostafazadeh, James Allen, and James Pustejovsky. 2015a. Semeval-2015 task 5: Qa tempeval - evaluating temporal information understanding with question answering. In Proc. of the 9th International Workshop on Semantic Evaluation, pages 46–5… find
  13. Marc Verhagen, Robert Gaizauskas, Frank Schilder, Mark Hepple, Graham Katz, and James Pustejovsky. 2007. Semeval-2007 task 15: Tempeval temporal relation identification. In Proc. of the 4th International Workshop on Semantic Evaluations, pages 75–80. Association for Computational Linguistics. find
  14. Hector Llorens, Nathanael Chambers, Naushad Uz-Zaman, Nasrin Mostafazadeh, James Allen, and James Pustejovsky. 2015b. Semeval-2015 task 5: Qa tempeval-evaluating temporal information understanding with question answering. In Proc. of the International Workshop on Semantic Evaluation (SemEval-2015). find
  15. Marc Verhagen, Roser Sauri, Tommaso Caselli, and James Pustejovsky. 2010. Semeval-2010 task 13: Tempeval-2. In Proc. of the 5th international workshop on semantic evaluation, pages 57–62. Association for Computational Linguistics. find
  16. Inderjeet Mani, Ben Wellner, Marc Verhagen, and James Pustejovsky. 2007. Three approaches to learning tlinks in timeml. Technical Report CS-07– 268, Computer Science Department. find
  17. Ngoc Thang Vu, Heike Adel, Pankaj Gupta, and Hinrich Schütze. 2016. Combining recurrent and convolutional neural networks for relation classification. CoRR, abs/1605.07333. find
  18. Anne-Lyse Minard, Manuela Speranza, Eneko Agirre, Itziar Aldabe, Marieke van Erp, Bernardo Magnini, German Rigau, Rubén Urizar, and Fondazione Bruno Kessler. 2015. Semeval-2015 task 4: Timeline: Cross-document event ordering. In Proc. of the International Workshop on Semantic Evaluation (SemEval-20… find
  19. Yan Xu, Lili Mou, Ge Li, Yunchuan Chen, Hao Peng, and Zhi Jin. 2015. Classifying relations via long short term memory networks along shortest dependency paths. In Proc. of EMNLP 2015, pages 1785– 1794. Association for Computational Linguistics.
  20. Daojian Zeng, Kang Liu, Siwei Lai, Guangyou Zhou, and Jun Zhao. 2014. Relation classification via convolutional deep neural network. In Proc. of COL-ING 2014, pages 2335–2344.
  21. P Mirza and S Tonelli. 2016. Catena: Causal and temporal relation extraction from natural language texts. In The 26th International Conference on Computational Linguistics, pages 64–75. Association for Computational Linguistics. find
  22. Dongxu Zhang and Dong Wang. 2015. Relation classification via recurrent neural network. CoRR, abs/1508.01006. find
  23. Paramita Mirza. 2016. Extracting temporal and causal relations between events. CoRR, abs/1604.08120. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/meng-2017-temporal-information-extraction-question-answering/