---
title: "Learning with Compositional Semantics as Structural Inference"
description: "Choi and Cardie compare two families of methods. 19 references, cited by 71 papers here."
authors: ["Yejin Choi", "Claire Cardie"]
year: 2008
venue: "Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing"
source: "https://aclanthology.org/D08-1083/"
licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)"
licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/choi-2008-learning-compositional-semantics-as-structural/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/choi-2008-learning-compositional-semantics-as-structural/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2008](https://md.research.elamur.ai/free/en/papers/2008/) / Learning with Compositional Semantics as Structural Inference for Subsentential Sentiment Analysis

# Learning with Compositional Semantics as Structural Inference for Subsentential Sentiment Analysis

**Authors:** [Yejin Choi](https://md.research.elamur.ai/free/en/authors/yejin-choi/) and [Claire Cardie](https://md.research.elamur.ai/free/en/authors/claire-cardie/) · **Year:** 2008 · **Venue:** [Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing](https://md.research.elamur.ai/free/en/venues/emnlp/) ([EMNLP 2008](https://md.research.elamur.ai/free/en/venues/emnlp/2008/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/)

**Links:** [Publisher's page](https://aclanthology.org/D08-1083/) · [PDF](https://aclanthology.org/D08-1083.pdf) · [OpenAlex](https://openalex.org/W2160052288)

**Citations:** 332 citations (OpenAlex, 2026-10-14) · [71](https://md.research.elamur.ai/free/en/papers/choi-2008-learning-compositional-semantics-as-structural/cited-by/) in our library, of the 141,840 papers whose bibliographies we read

## Summary
The paper addresses polarity classification of sentiment-bearing expressions below the sentence level. In such expressions, words interact: negators such as "not", or verbs such as "eliminated", can reverse the polarity of their arguments, so simple word voting breaks down. Choi and Cardie compare two families of methods. The first is heuristic: majority voting, voting with polarity flips by negators, and hand-written composition rules inspired by the Principle of Compositionality. The second is learning-based: a plain classifier, and a model that builds compositional inference into training as a structural step. They evaluate on expression-level annotations from the MPQA corpus. Polarity and negator lexicons are built from Wilson et al. (2005), the General Inquirer and WordNet. Compositional heuristics beat both non-compositional heuristics and a simple learner. Combining learning with compositional inference gives the best accuracy (90.7%). The work matters because it links formal compositional semantics with statistical learning. It also shows that content words acting as negators deserve explicit treatment.

## Key points
- Accuracy on MPQA expressions:
  - non-compositional heuristics: 87.7%
  - compositional heuristics (COMPO): 89.7%
  - simple learning without composition: 89.1%
  - learning with compositional inference (CCI-COMPO): 90.7%
  - The cross-validation set has about 4.9k instances.
- Seven composition rules are defined over coarse syntactic patterns detected with POS tags and function words, such as VP+NP or "NP be adj". Two conflict-resolution variants are tested:
  - COMPO MC falls back to the majority class (negative in this data).
  - COMPO PR lets the argument with higher semantic priority win.
- The negator lexicon covers function-word negators and content-word negators. It starts from seed words plus the General Inquirer categories NOTLW and DECREAS, and is expanded through WordNet synonyms using a vote over senses.
- Widening expression boundaries by 1, 5 or unlimited words, up to the sentence boundary, lowers accuracy for every method. NEGEX methods fall below VOTE at sentence scope, while COMPO degrades more gently.
- On SemEval-07 (208 strong-intensity test instances, with 400 MPQA documents used for training), the results are 88.6% for COMPO PR, 90.1% for SCNEGEX and 87.6% for CCICOMPOMC. These are lower than Moilanen and Pulman (2007), which the authors attribute to domain and annotation differences.

## Interesting ideas
- **Learning plus symbolic rules.** Hand-written composition rules can be embedded as a structural inference step inside learning, rather than applied rigidly. The learner then absorbs exceptions, unknown words and missing rules. This hybrid pattern could transfer to other low-resource languages, such as Ukrainian, where some linguistic rules are known but data is limited.
- **Content-word negators as a lexical category.** Verbs and nouns like "eliminate" or "lack" can be treated as polarity flippers. A small seed list grown through a thesaurus is a cheap way to build such a lexicon for a new language.
- **Asymmetric negation scope.** In the composition rules, a negator flips only the constituent that follows it, not the one before it. A negator in second position is instead counted as an ordinary negative item. This is a simple scope heuristic worth reusing in rule-based sentiment systems.
- **Boundary-expansion evaluation.** Gradually adding context around gold spans tests whether extra text helps or hurts. Here it hurt, which argues for detecting opinion spans first and classifying them second. The same probe could reveal context sensitivity in other span-level tasks.
- **Graded baselines.** Comparing plain voting, voting with one flip, voting with n flips, and compositional rules isolates the contribution of each linguistic assumption. This ladder-of-heuristics design is a clean way to justify the complexity of a model.

*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

- [Cornell University](https://md.research.elamur.ai/free/en/universities/cornell-university/) — University, Ithaca, United States · 89 works · 58,422 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 (1): [😊 Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/) · 608 works

## Cite

Cite as: Yejin Choi and Claire Cardie. 2008. Learning with Compositional Semantics as Structural Inference for Subsentential Sentiment Analysis. Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{choi2008learning,
  title = {{Learning with Compositional Semantics as Structural Inference for Subsentential Sentiment Analysis}},
  author = {Yejin Choi and Claire Cardie},
  booktitle = {Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing},
  year = {2008},
  doi = {10.3115/1613715.1613816},
  url = {https://doi.org/10.3115/1613715.1613816}
}
```

- [← previous Lattice-based Minimum Error Rate Training for Statistical Machine Translation](https://md.research.elamur.ai/free/en/papers/macherey-2008-lattice-based-minimum-error-rate/)
- [next → Cheap and Fast – But is it Good? Evaluating Non-Expert Annotations for Natural Language Tasks](https://md.research.elamur.ai/free/en/papers/snow-2008-cheap-fast-but-is-it/)
- [↑ back to papers of 2008](https://md.research.elamur.ai/free/en/papers/2008/)

## References

19 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Steven Abney. 1996. Partial parsing via finite-state cascades. Journal of Natural Language Engineering, 2(4):337344. [find](https://scholar.google.com/scholar?q=Steven+Abney%2E+1996%2E+Partial+parsing+via+finite%2Dstate+cascades%2E+Journal+of+Natural+Language+Engineering%2C+2%284%29%3A337344%2E)
2. Koby Crammer and Yoram Singer. 2003. Ultraconservative online algorithms for multiclass problems. JMLR 3:951. [find](https://scholar.google.com/scholar?q=Koby+Crammer+and+Yoram+Singer%2E+2003%2E+Ultraconservative+online+algorithms+for+multiclass+problems%2E+JMLR+3%3A951%2E)
3. David R. Dowty, Robert E. Wall and Stanley Peters. 1981. Introduction to Montague Semantics. [find](https://scholar.google.com/scholar?q=David+R%2E+Dowty%2C+Robert+E%2E+Wall+and+Stanley+Peters%2E+1981%2E+Introduction+to+Montague+Semantics%2E)
4. Andrea Esuli and Fabrizio Sebastiani. 2006. SentiWord-Net: A Publicly Available Lexical Resource for Opinion Mining. In Proceedings of 5th Conference on Language Resources and Evaluation (LREC),. [find](https://scholar.google.com/scholar?q=Andrea+Esuli+and+Fabrizio+Sebastiani%2E+2006%2E+SentiWord%2DNet%3A+A+Publicly+Available+Lexical+Resource+for+Opinion+Mining%2E+In+Proceedings+of+5th+Conference+on+Language+Resources+and+Evaluation+%28LREC%29%2C%2E)
5. Percy Liang, Hal Daumé III and Dan Klein. 2008. Structure Compilation: Trading Structure for Features. In International Conference on Machine Learning. [find](https://scholar.google.com/scholar?q=Percy+Liang%2C+Hal+Daume%CC%81+III+and+Dan+Klein%2E+2008%2E+Structure+Compilation%3A+Trading+Structure+for+Features%2E+In+International+Conference+on+Machine+Learning%2E)
6. Minqing Hu and Bing Liu. 2004. Mining and summarizing customer reviews. In Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD-2004). [find](https://scholar.google.com/scholar?q=Minqing+Hu+and+Bing+Liu%2E+2004%2E+Mining+and+summarizing+customer+reviews%2E+In+Proceedings+of+the+ACM+SIGKDD+International+Conference+on+Knowledge+Discovery+%26+Data+Mining+%28KDD%2D2004%29%2E)
7. Alistair Kennedy and Diana Inkpen. 2005. Sentiment Classification of Movie and Product Reviews Using Contextual Valence Shifters. In Proceedings of FINEXIN 2005, Workshop on the Analysis of Informal and Formal Information Exchange during Negotiations. [find](https://scholar.google.com/scholar?q=Alistair+Kennedy+and+Diana+Inkpen%2E+2005%2E+Sentiment+Classification+of+Movie+and+Product+Reviews+Using+Contextual+Valence+Shifters%2E+In+Proceedings+of+FINEXIN+2005%2C+Workshop+on+the+Analysis+of+Informal+a)
8. Soo-Min Kim and Eduard Hovy. 2004. [Determining the sentiment of opinions](https://md.research.elamur.ai/free/en/papers/kim-2004-determining-sentiment-opinions/). In Proceedings of COLING.
9. Ryan McDonald, Kerry Hannan, Tyler Neylon, Mike Wells and Jeff Reynar. 2007. Structured Models for Fine-to-Coarse Sentiment Analysis. In Proceedings of Association for Computational Linguistics (ACL) . [find](https://scholar.google.com/scholar?q=Ryan+McDonald%2C+Kerry+Hannan%2C+Tyler+Neylon%2C+Mike+Wells+and+Jeff+Reynar%2E+2007%2E+Structured+Models+for+Fine%2Dto%2DCoarse+Sentiment+Analysis%2E+In+Proceedings+of+Association+for+Computational+Linguistics+%28ACL%29+)
10. George A. Miller. 1995. [WordNet: a lexical database for English](https://md.research.elamur.ai/free/en/papers/miller-1994-wordnet/). In Communications of the ACM, 38(11):3941
11. Richard Montague. 1974. Formal Philosophy; Selected papers of Richard Montague. Yale University Press. [find](https://scholar.google.com/scholar?q=Richard+Montague%2E+1974%2E+Formal+Philosophy%3B+Selected+papers+of+Richard+Montague%2E+Yale+University+Press%2E)
12. Karo Moilanen and Stephen Pulman. 2007. Sentiment Composition. In Proceedings of Recent Advances in Natural Language Processing (RANLP 2007). [find](https://scholar.google.com/scholar?q=Karo+Moilanen+and+Stephen+Pulman%2E+2007%2E+Sentiment+Composition%2E+In+Proceedings+of+Recent+Advances+in+Natural+Language+Processing+%28RANLP+2007%29%2E)
13. Rowan Nairn, Cleo Condoravdi and Lauri Karttunen 2006. [Computing relative polarity for textual inference](https://md.research.elamur.ai/free/en/papers/nairn-2006-computing-relative-polarity-textual-inference/). In Inference in Computational Semantics (ICoS-5).
14. Yun Niu, Xiaodan Zhu, Jianhua Li and Graeme Hirst. 2005. Analysis of polarity information in medical text. In Proceedings of the American Medical Informatics Association 2005 Annual Symposium (AMIA). [find](https://scholar.google.com/scholar?q=Yun+Niu%2C+Xiaodan+Zhu%2C+Jianhua+Li+and+Graeme+Hirst%2E+2005%2E+Analysis+of+polarity+information+in+medical+text%2E+In+Proceedings+of+the+American+Medical+Informatics+Association+2005+Annual+Symposium+%28AMIA%29%2E)
15. Livia Polanyi and Annie Zaenen. 2004. Contextual lexical valence shifters. In Exploring Attitude and Affect in Text: Theories and Applications: Papers from the 2004 Spring Symposium, AAAI. [find](https://scholar.google.com/scholar?q=Livia+Polanyi+and+Annie+Zaenen%2E+2004%2E+Contextual+lexical+valence+shifters%2E+In+Exploring+Attitude+and+Affect+in+Text%3A+Theories+and+Applications%3A+Papers+from+the+2004+Spring+Symposium%2C+AAAI%2E)
16. Mostafa Shaikh, Helmut Prendinger and Mitsuru Ishizuka. 2007. Assessing sentiment of text by semantic dependency and contextual valence analysis. In Proc 2nd Int’l Conf on Affective Computing and Intelligent Interaction (ACII’07). [find](https://scholar.google.com/scholar?q=Mostafa+Shaikh%2C+Helmut+Prendinger+and+Mitsuru+Ishizuka%2E+2007%2E+Assessing+sentiment+of+text+by+semantic+dependency+and+contextual+valence+analysis%2E+In+Proc+2nd+Int%E2%80%99l+Conf+on+Affective+Computing+and+Inte)
17. Carlo Strapparava and Rada Mihalcea. 2007. [Semeval-2007 task 14: Affective text](https://md.research.elamur.ai/free/en/papers/strapparava-2007-semeval-2007-task-14-affective/). In Proceedings of SemEval.
18. Janyce Wiebe, Theresa Wilson and Claire Cardie. 2005. Annotating expressions of opinions and emotions in language. In Language Resources and Evaluation (formerly Computers and the Humanities), 39(2-3):165210. [find](https://scholar.google.com/scholar?q=Janyce+Wiebe%2C+Theresa+Wilson+and+Claire+Cardie%2E+2005%2E+Annotating+expressions+of+opinions+and+emotions+in+language%2E+In+Language+Resources+and+Evaluation+%28formerly+Computers+and+the+Humanities%29%2C+39%282%2D3%29)
19. Theresa Wilson, Janyce Wiebe and Paul Hoffmann. 2005. [Recognizing contextual polarity in phrase-level sentiment analysis](https://md.research.elamur.ai/free/en/papers/wilson-2005-recognizing-contextual-polarity-phrase-level/). In Proceedings of HLT/EMNLP.

*Updated 24 October 2026*