--- title: "SoNLP-DP System for ConLL-2016 English Shallow Discourse" description: "The authors build a mostly pipelined parser with four stages: a connective identifier, a… 31.18 F1. 11 references, cited by 1 paper here." authors: ["Fang Kong", "Sheng Li", "Junhui Li", "Muhua Zhu", "Guodong Zhou"] year: 2016 venue: "Proceedings of the CoNLL-16 shared task" source: "https://aclanthology.org/K16-2009/" doi: "10.18653/v1/k16-2009" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/kong-2016-sonlp-dp-system-conll-2016/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/kong-2016-sonlp-dp-system-conll-2016/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2016](https://md.research.elamur.ai/free/en/papers/2016/page/4/) / SoNLP-DP System for ConLL-2016 English Shallow Discourse Parsing # SoNLP-DP System for ConLL-2016 English Shallow Discourse Parsing **Authors:** [Fang Kong](https://md.research.elamur.ai/free/en/authors/fang-kong/), [Sheng Li](https://md.research.elamur.ai/free/en/authors/sheng-li/), Junhui Li, [Muhua Zhu](https://md.research.elamur.ai/free/en/authors/muhua-zhu/) and [Guodong Zhou](https://md.research.elamur.ai/free/en/authors/guodong-zhou/) · **Year:** 2016 · **Venue:** [Proceedings of the CoNLL-16 shared task](https://md.research.elamur.ai/free/en/venues/conll/) ([CoNLL 2016](https://md.research.elamur.ai/free/en/venues/conll/2016/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/K16-2009/) · [PDF](https://aclanthology.org/K16-2009.pdf) · [DOI 10.18653/v1/k16-2009](https://doi.org/10.18653/v1/k16-2009) · [OpenAlex](https://openalex.org/W2512354421) **Citations:** 2 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/kong-2016-sonlp-dp-system-conll-2016/cited-by/) in our library, of the 141,840 papers whose bibliographies we read ## Summary SoNLP-DP is the Soochow University entry to the CoNLL-2016 shared task on end-to-end English shallow discourse parsing in the closed track. The task is to find discourse relations in PDTB-style text. Each relation is a connective, either explicit or implicit, with two argument spans and a sense label. The authors build a mostly pipelined parser with four stages: - a connective identifier, - a constituent-based argument labeler, - an explicit sense classifier, - a non-explicit sense classifier for adjacent sentence pairs inside a paragraph. The pipeline is not strictly sequential, because the authors account for interactions between argument labeling and sense classification. Every component is a maximum entropy classifier built on hand-designed lexical and syntactic features taken from constituent and dependency parses. The system reaches an overall F1 of 24.31% on the blind test set and 28.78% on the test set. On the blind set this slightly beats the best system from 2015. A split evaluation shows that implicit and other non-explicit relations remain the main bottleneck. ## Key points - Connective detection first collects candidates from the 100 connective types listed in the PDTB. Each candidate is then disambiguated with two kinds of features: - lexical: the word, its POS, neighbouring words and its position in the sentence; - syntactic: the dominating node and its context, whether the right sibling contains a VP, and the path to the root. - Argument labeling follows Kong et al. (2014). Parse-tree constituents are labeled Arg1, Arg2 or NULL and then merged into spans, with no ILP joint inference. - The non-explicit classifier uses five feature groups: - production rules, with rules seen fewer than 5 times dropped; - dependency rules; - first, last and first-three words; - the top 500 word pairs selected by mutual information; - pairs of the 3200 Brown clusters supplied by the organizers. - All classifiers use the OpenNLP maxent package with default settings: no smoothing and 100 iterations. - Results on dev / test / blind: - overall parser F1: 35.56 / 28.78 / 24.31, against 24.00 for Wang and Lan (2015) on blind; - connective F1: about 94 / 94.7 / 91.6. - On blind data, explicit relations reach 31.18 F1 and non-explicit relations only 16.10. ## Interesting ideas - **Arguments from tree nodes.** Argument spans are found by labeling each tree node as Arg1, Arg2 or nothing and then joining the labeled nodes, so every span boundary comes from the syntax tree rather than from free token windows. - **Closed list, then disambiguation.** Connectives are first matched against a fixed list, and a classifier then decides whether each match really works as a connective, using cues like the dominating node and whether a VP follows. - **An explicit "no relation" class.** Adjacent sentences with no connection are labeled NoRel instead of being forced into some relation. - **Separate scores for easy and hard parts.** Reporting explicit and non-explicit results separately shows exactly which subtask drags the total score down. - **Dropping rare features.** Syntactic production rules seen fewer than 5 times are discarded before training, which keeps the feature set small and less noisy. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (frames): Links between clauses such as cause or contrast could be attached to UD subtrees in the same way, which keeps each discourse link traceable to tree nodes. - **I2** (tree): Ambiguous words like "since" or "and" could be checked with a few explainable structural cues over the UD tree before they are treated as discourse markers. - **I3** (world): When ordering events, the world model should be allowed to record "no link" between neighbouring sentences rather than inventing a causal or temporal edge. - **I4** (learn): Evaluations could be split by phenomenon, for example marked versus unmarked relations, so that training effort goes to the weakest stage. - **I5** (tree): A similar frequency threshold on structural feature templates could shrink the perceptron parser's model without hurting how interpretable it is. *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 - [Alibaba Group](https://md.research.elamur.ai/free/en/companies/alibaba-group/) — Company, Hangzhou, China · 64 works · 14,113 citations - [Soochow University](https://md.research.elamur.ai/free/en/universities/soochow-university/) — University, Suzhou, China · 37 works · 4,291 citations 📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,086 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,889 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 9,296 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 9,206 works 🧭 Topics (2): [English](https://md.research.elamur.ai/free/en/topics/english/) · 1,373 works · [🧵 Discourse analysis](https://md.research.elamur.ai/free/en/topics/discourse-analysis/) · 394 works ## Cite Cite as: Fang Kong, Sheng Li, Junhui Li, Muhua Zhu and Guodong Zhou. 2016. SoNLP-DP System for ConLL-2016 English Shallow Discourse Parsing. Proceedings of the CoNLL-16 shared task. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{kong2016sonlp, title = {{SoNLP-DP System for ConLL-2016 English Shallow Discourse Parsing}}, author = {Fang Kong, Sheng Li, Junhui Li, Muhua Zhu and Guodong Zhou}, booktitle = {Proceedings of the CoNLL-16 shared task}, year = {2016}, doi = {10.18653/v1/k16-2009}, url = {https://doi.org/10.18653/v1/k16-2009} } ``` - [← previous Corpora for Learning the Mutual Relationship between Semantic Relatedness and Textual Entailment](https://md.research.elamur.ai/free/en/papers/vo-2016-corpora-learning-mutual-relationship-between/) - [next → The IBM 2016 English Conversational Telephone Speech Recognition System](https://md.research.elamur.ai/free/en/papers/saon-2016-ibm-2016-english-conversational-telephone/) - [↑ back to papers of 2016](https://md.research.elamur.ai/free/en/papers/2016/page/4/) ## References 11 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Rashmi Prasad, Nikhil Dinesh, Alan Lee, Eleni Milt-sakaki, Livio Robaldo, Aravind Joshi, and Bonnie Webber. 2008. [The Penn Discourse TreeBank 2.0](https://md.research.elamur.ai/free/en/papers/prasad-2008-penn-discourse-treebank-2-0/). In Proceedings of the LREC 2008 Conference. 2. Ziheng Lin, Min-Yen Kan, and Hwee Tou Ng. 2009. [Recognizing implicit discourse relations in the Penn Discourse Treebank](https://md.research.elamur.ai/free/en/papers/lin-2009-recognizing-implicit-discourse-relations-penn/). In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing. 3. Rashmi Prasad, Aravind Joshi, and Bonnie Webber. 2010. [Exploiting scope for shallow discourse parsing](https://md.research.elamur.ai/free/en/papers/prasad-2010-exploiting-scope-shallow-discourse-parsing/). In Proceedings of the Seventh International Conference on Language Resources and Evaluation. 4. Jianxiang Wang and Man Lan. 2015. [A refined end-to-end discourse parser](https://md.research.elamur.ai/free/en/papers/wang-2015-refined-end-end-discourse-parser/). In Proceedings of the Nineteenth Conference on Computational Natural Language Learning - Shared Task, pages 17–24, Beijing, China, July. Association for Computational Linguistics. 5. Ziheng Lin, Hwee Tou Ng, and Min-Yen Kan. 2011. [Automatically evaluating text coherence using discourse relations](https://md.research.elamur.ai/free/en/papers/lin-2011-automatically-evaluating-text-coherence-discourse/). In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics. 6. Ziheng Lin, Chang Liu, Hwee Tou Ng, and Min-Yen Kan. 2012. Combining coherence models and machine translation evaluation metrics for summarization evaluation. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Ziheng+Lin%2C+Chang+Liu%2C+Hwee+Tou+Ng%2C+and+Min%2DYen+Kan%2E+2012%2E+Combining+coherence+models+and+machine+translation+evaluation+metrics+for+summarization+evaluation%2E+In+Proceedings+of+the+50th+Annual+Meeting) 7. Bonnie Webber, Marcus Egg, and Valia Kordoni. 2012. Discourse structure and language technology. Natural Language Engineering, 18(4):437–490, 10. [↗](https://doi.org/10.1017/s1351324911000337) 8. Ziheng Lin, Hwee Tou Ng, and Min-Yen Kan. 2014. A PDTB-styled end-to-end discourse parser. Natural Language Engineering, 20(2):151–184. [find](https://scholar.google.com/scholar?q=Ziheng+Lin%2C+Hwee+Tou+Ng%2C+and+Min%2DYen+Kan%2E+2014%2E+A+PDTB%2Dstyled+end%2Dto%2Dend+discourse+parser%2E+Natural+Language+Engineering%2C+20%282%29%3A151%E2%80%93184%2E) 9. Mitchell P. Marcus, Beatrice Santorini, and Mary Ann Marcinkiewicz. 1993. [Building a large annotated corpus of English: The Penn Treebank](https://md.research.elamur.ai/free/en/papers/marcus-1993-building-large-annotated-corpus-english/). Computational Linguistics, 19(2):313–330. [↗](https://doi.org/10.21236/ada273556) 10. Thomas Meyer and Bonnie Webber. 2013. Implicita-tion of discourse connectives in (machine) translation. In Proceedings of the Workshop on Discourse in Machine Translation. [find](https://scholar.google.com/scholar?q=Thomas+Meyer+and+Bonnie+Webber%2E+2013%2E+Implicita%2Dtion+of+discourse+connectives+in+%28machine%29+translation%2E+In+Proceedings+of+the+Workshop+on+Discourse+in+Machine+Translation%2E) 11. Emily Pitler and Ani Nenkova. 2009. [Using syntax to disambiguate explicit discourse connectives in text](https://md.research.elamur.ai/free/en/papers/pitler-2009-syntax-disambiguate-explicit-discourse-connectives/). In Proceedings of the ACL-IJCNLP 2009 Conference Short Papers. *Updated 24 October 2026*