Aggregated Word Pair Features for Implicit Discourse Relation Disambiguation
Summary
The paper tackles the task of classifying discourse relations that are left implicit in the Penn Discourse Treebank (PDTB), where no connective such as but or because signals the relation. Earlier systems used one sparse feature per word pair. PDTB is too small to train so many features well, and the most informative pairs turned out to be frequent function words. Biran and McKeown instead group the pairs by discourse marker. They collect pairs that occur around each of the 102 PDTB markers in the Gigaword corpus and treat each marker as a "document." Each relation instance then gets a score per marker: the cosine similarity between its own pairs and that document, weighted with tf-idf or pmi-idf. Logistic regression models built on these 102 dense features beat the earlier sparse word-pair features for three of the four top-level classes. Results stay largely the same when a stop list is applied, which suggests the useful pairs carry real content. A full system with extra lexical features approaches state-of-the-art results without gold parses or surrounding context, so it can be applied more broadly.
Key points
- Setup: Four binary classifiers, one each for Comparison, Contingency, Expansion and Temporal. EntRel is counted as Expansion, training uses sections 2–20, testing uses sections 21–22, and explicit relations are excluded.
- Word-pair results (F-measure), compared with Pitler et al. (2009):
- tf-idf: Comparison 23 vs. 21.96, Expansion 66.48 vs. 63.84, Temporal 19.54 vs. 16.21, Contingency 44.03 vs. 45.6.
- pmi-idf: best for Comparison at 24.38, but tf-based weighting is generally stronger.
- Stop list: Removing 50 function words changes little except for Temporal, which drops from 19.54 to 16.98. The authors suggest that words like was and will carry temporal meaning.
- New feature groups: The full system adds WordNet, Affect (Dictionary of Affect in Language), Length and Negation features. It also uses verb class (Levin/LCS), MPQA polarity, modality, money/percent/number counts and content similarity.
- Final system: It outperforms Pitler et al. (2009) on Comparison and Temporal. It is competitive with Zhou et al. (2010) and Park and Cardie (2012) on Contingency and Expansion while using no syntactic or context features.
Cited by
- PDTB Discourse Parsing as a Tagging Task: The Two Taggers Approach
- Comparing Word Representations for Implicit Discourse Relation Classification
- Learning Connective-based Word Representations for Implicit Discourse Relation Identification
- Implicit Discourse Relation Detection via a Deep Architecture with Gated Relevance Network
- A Regularization Approach for Incorporating Event Knowledge and Coreference Relations into Neural Discourse Parsing
- Closing the Gap: Domain Adaptation from Explicit to Implicit Discourse Relations
- One Vector is Not Enough: Entity-Augmented Distributed Semantics for Discourse Relations
- Weakly-Supervised Modeling of Contextualized Event Embedding for Discourse Relations
- Reducing Sparsity Improves the Recognition of Implicit Discourse Relations
- Automatic Prediction of Discourse Connectives
- Implicit Discourse Relation Recognition with Context-aware Character-enhanced Embeddings
- A Stacking Gated Neural Architecture for Implicit Discourse Relation Classification
- Adversarial Connective-exploiting Networks for Implicit Discourse Relation Classification
- Discovering Implicit Discourse Relations Through Brown Cluster Pair Representation and Coreference Patterns
- Improving the Inference of Implicit Discourse Relations via Classifying Explicit Discourse Connectives
- Using Explicit Discourse Connectives in Translation for Implicit Discourse Relation Classification
- On the Need of Cross Validation for Discourse Relation Classification
- Next Sentence Prediction helps Implicit Discourse Relation Classification within and across Domains
- Discourse Relation Prediction: Revisiting Word Pairs with Convolutional Networks
- Bilingually-constrained Synthetic Data for Implicit Discourse Relation Recognition
- Improving Implicit Discourse Relation Recognition with Discourse-specific Word Embeddings
- Encoding and Fusing Semantic Connection and Linguistic Evidence for Implicit Discourse Relation Recognition
- The CoNLL-2015 Shared Task on Shallow Discourse Parsing
- CoNLL 2016 Shared Task on Multilingual Shallow Discourse Parsing
- Shallow Convolutional Neural Network for Implicit Discourse Relation Recognition
Cite
@inproceedings{biran2013aggregated,
title = {{Aggregated Word Pair Features for Implicit Discourse Relation Disambiguation}},
author = {Biran, Or and McKeown, Kathleen},
booktitle = {Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)},
year = {2013},
doi = {10.7916/d8pn9fz4},
url = {https://doi.org/10.7916/d8pn9fz4}
}References
- Nicholas Asher and Alex Lascarides. 2003. Logics of Conversation. Studies in Natural Language Processing Series. Cambridge University Press. find
- Or Biran and Owen Rambow. 2011. Identifying justifications in written dialog by classifying text as argumentative. International Journal of Semantic Computing, 5(4):363–381, December. find
- Sasha Blair-Goldensohn, Kathleen McKeown, and Owen Rambow. 2007. Building and refining rhetorical-semantic relation models. In HLT-NAACL, pages 428–435. The Association for Computational Linguistics. find
- Bonnie J. Dorr. 2001. LCS Verb Database, Online Software Database of Lexical Conceptual Structures and Documentation. University Of Maryland College Park. find
- Christiane Fellbaum, editor. 1998. WordNet An Electronic Lexical Database. The MIT Press. find
- Beth Levin. 1993. English Verb Classes and Alternations: A Preliminary Investigation. University Of Chicago Press. find
- Ziheng Lin, Min-Yen Kan, and Hwee Tou Ng. 2009. Recognizing implicit discourse relations in the penn discourse treebank. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 343–351. find our card
- William C. Mann and Sandra A. Thompson. 1987. Rhetorical Structure Theory: A theory of text organization. Technical Report ISI/RS-87-190, ISI. find
- Daniel Marcu and Abdessamad Echihabi. 2002. An unsupervised approach to recognizing discourse relations. In ACL, pages 368–375. ACL. find our card
- Joonsuk Park and Claire Cardie. 2012. Improving implicit discourse relation recognition through feature set optimization. In Proceedings of the 13th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 108–112. find our card
- Emily Pitler and Ani Nenkova. 2009. Using syntax to disambiguate explicit discourse connectives in text. In ACL/IJCNLP (Short Papers), pages 13–16. The Association for Computer Linguistics. find our card
- Emily Pitler, Annie Louis, and Ani Nenkova. 2009. Automatic sense prediction for implicit discourse relations in text. In ACL/IJCNLP, pages 683–691. The Association for Computer Linguistics. find our card
- Rashmi Prasad, Nikhil Dinesh, Alan Lee, Eleni Miltsakaki, Livio Robaldo, Aravind Joshi, and Bonnie Webber. 2008. The penn discourse treebank 2.0. In In Proceedings of LREC. find our card
- Cynthia M. Whissell. 1989. The dictionary of affect in language. find
- Theresa Wilson, Janyce Wiebe, and Paul Hoffmann. 2005. Recognizing contextual polarity in phrase-level sentiment analysis. In Proceedings of the conference on Human Language Technology and Empirical Methods in Natural Language Processing, pages 347–354. find
- Zhi-Min Zhou, Yu Xu, Zheng-Yu Niu, Man Lan, Jian Su, and Chew Lim Tan. 2010. Predicting discourse connectives for implicit discourse relation recognition. In Proceedings of the 23rd International Conference on Computational Linguistics. find our card