---
title: "A Combined Memory-Based Semantic Role Labeler of English"
description: "The paper describes a system entered in the closed track of the CoNLL-2008 shared task. 86.07 LAS. 13 references, cited by 5 papers here."
authors: ["Roser Morante", "Walter Daelemans", "Vincent Van Asch"]
year: 2008
venue: "CoNLL 2008: Proceedings of the Twelfth Conference on Computational Natural Language Learning"
source: "https://aclanthology.org/W08-2128/"
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/morante-2008-combined-memory-based-semantic-role/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/morante-2008-combined-memory-based-semantic-role/)

[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/page/2/) / A Combined Memory-Based Semantic Role Labeler of English

# A Combined Memory-Based Semantic Role Labeler of English

**Authors:** [Roser Morante](https://md.research.elamur.ai/free/en/authors/roser-morante/), [Walter Daelemans](https://md.research.elamur.ai/free/en/authors/walter-daelemans/) and [Vincent Van Asch](https://md.research.elamur.ai/free/en/authors/vincent-van-asch/) · **Year:** 2008 · **Venue:** [CoNLL 2008: Proceedings of the Twelfth Conference on Computational Natural Language Learning](https://md.research.elamur.ai/free/en/venues/conll-proceedings-of-the-conference-on-computational-natural-language-learning/) ([CoNLL: Proceedings of the Conference on Computational Natural Language Learning 2008](https://md.research.elamur.ai/free/en/venues/conll-proceedings-of-the-conference-on-computational-natural-language-learning/2008/)) · **Publisher:** [International Committee on Computational Linguistics (ICCL)](https://md.research.elamur.ai/free/en/publishers/international-committee-on-computational-linguistics-iccl/) · **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/W08-2128/) · [PDF](https://aclanthology.org/W08-2128.pdf) · [OpenAlex](https://openalex.org/W2100347937)

**Citations:** 15 citations (OpenAlex, 2026-10-14) · 5 in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper describes a system entered in the closed track of the CoNLL-2008 shared task. The task asked for both syntactic dependencies and semantic role dependencies for verbal and nominal predicates. The authors handle the two levels separately. Syntax comes from MaltParser 0.4, trained with SVMs and an existing English configuration. Semantics comes from a cascade of memory-based (k-nearest-neighbour) classifiers built with TiMBL's IB1. The cascade first decides which nouns and verbs are predicates, then labels the role of each word for each predicate. A simple agreement rule merges three groups of role classifiers. No word sense disambiguation was built, so every predicate receives sense ".01".

The system reaches 78.43 labeled macro F1 overall, 86.07 LAS and 70.51 semantic labeled F1. An error analysis with oracle inputs locates the weak points:
- Errors in predicate detection carry through the whole pipeline.
- Nouns are harder than verbs.
- Recall lags well behind precision.

The work shows how a transparent, example-based learner performs in a dependency-based SRL setting.

## Key points
- Results by test set:
  - In-domain WSJ: 71.88 semantic labeled F1 and 86.88 LAS.
  - Out-of-domain Brown: 59.23 semantic labeled F1 and 79.58 LAS.
- Predicate identification on the development set reaches 0.9599 accuracy for verbs and 0.8981 for nouns. Only nouns and verbs are considered as candidates.
- The role-labeling stage has three classifier groups:
  - G1 makes a joint decision on attachment and role.
  - G2 and G3 use different feature sets and are applied only to pairs that G1 marked.
  - When G2 and G3 agree, their label is used; otherwise G1's label stands.
  - On the development set this raised labeled F1 from 66.07 (G1 alone) to 69.75, a 10.8% error reduction.
- With gold predicates, labeled F1 rises to 80.89, which is 9.01 points above the submitted system.
- Labeled precision exceeds recall by 12.40 points. The gap stays at about 10 points even with gold syntax and gold predicates (89.20 P / 79.47 R), so it is inherent to the labeler.

## Interesting ideas
- Two weaker classifiers with different feature views act as a joint veto over a stronger primary classifier; they override it only when they agree, which curbs overgeneration of roles.
- Candidates are pruned from the training instances when their category never carries a role in the training data (e.g. determiners), which shrinks the instance space at almost no cost.
- Oracle ablations (gold predicates, then gold syntax plus predicates) measure how much each upstream stage costs a cascaded pipeline.
- Memory-based k-NN keeps literal training examples, so exceptions and subregularities are not smoothed away, and each decision can be traced back to its nearest stored neighbours.
- Features include POS chains between the word and the predicate, and POS chains linking the word to the neighbouring predicates. These give the classifier context about competing predicates in the same sentence.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (learn): The same rule could merge silver data from witness parsers: keep the main parser's arc unless two independent witnesses agree on a different label.
- **I2** (frames): Before assigning VerbNet roles, the role assigner could skip dependents whose UD category or relation never fills a role in its data, making the search cheaper and easier to explain.
- **I3** (learn): Running each stage on gold vs. predicted input from the previous stage would show whether errors in tree or frame detection drive downstream failures.
- **I4** (frames): A nearest-example role labeler would match the explainability requirement, since each assigned role could cite the training frames that justified it.
- **I5** (frames): When a sentence has several verbs, such inter-predicate path features could help decide which frame claims a given argument.

*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 Antwerp](https://md.research.elamur.ai/free/en/universities/university-of-antwerp/) — University, Antwerp, Belgium · 17 works · 3,490 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works

🧭 Topics (2): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,379 works · [🎭 Semantic role labeling](https://md.research.elamur.ai/free/en/topics/semantic-role-labeling/) · 146 works

## Cite

Cite as: Roser Morante, Walter Daelemans and Vincent Van Asch. 2008. A Combined Memory-Based Semantic Role Labeler of English. CoNLL 2008: Proceedings of the Twelfth Conference on Computational Natural Language Learning.

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

```bibtex
@inproceedings{morante2008combined,
  title = {{A Combined Memory-Based Semantic Role Labeler of English}},
  author = {Roser Morante, Walter Daelemans and Vincent {Van Asch}},
  booktitle = {CoNLL 2008: Proceedings of the Twelfth Conference on Computational Natural Language Learning},
  year = {2008},
  doi = {10.3115/1596324.1596361},
  url = {https://doi.org/10.3115/1596324.1596361}
}
```

- [← previous Investigating the Portability of Corpus-Derived Cue Phrases for Dialogue Act Classification](https://md.research.elamur.ai/free/en/papers/webb-2008-investigating-portability-corpus-derived-cue/)
- [next → Parser Evaluation and the BNC: Evaluating 4 constituency parsers with 3 metrics](https://md.research.elamur.ai/free/en/papers/foster-2008-parser-evaluation-bnc-evaluating-4/)
- [↑ back to papers of 2008](https://md.research.elamur.ai/free/en/papers/2008/page/2/)

## References

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

1. Cover, T. M. and P. E. Hart. 1967. Nearest neighbor pattern classification. Institute of Electrical and Electronics Engineers Transactions on Information Theory, 13:21–27. [↗](https://doi.org/10.1109/tit.1967.1053964)
2. Daelemans, W. and A. van den Bosch. 2005. Memory-based language processing. Cambridge University Press, Cambridge, UK. [↗](https://doi.org/10.1017/cbo9780511486579)
3. Daelemans, W., A. Van den Bosch, and J. Zavrel. 1999. Forgetting exceptions is harmful in language learning. Machine Learning, Special issue on Natural Language Learning, 34:11–41. [↗](https://doi.org/10.1023/a:1007585615670)
4. Daelemans, W., J. Zavrel, K. Van der Sloot, and A. Van den Bosch. 2007. TiMBL: Tilburg memory based learner, version 6.1, reference guide. Technical Report Series 07-07, ILK, Tilburg, The Netherlands. [find](https://scholar.google.com/scholar?q=Daelemans%2C+W%2E%2C+J%2E+Zavrel%2C+K%2E+Van+der+Sloot%2C+and+A%2E+Van+den+Bosch%2E+2007%2E+TiMBL%3A+Tilburg+memory+based+learner%2C+version+6%2E1%2C+reference+guide%2E+Technical+Report+Series+07%2D07%2C+ILK%2C+Tilburg%2C+The+Netherlands%2E)
5. Hacioglu, K. 2004. Semantic role labeling using dependency trees. In COLING ’04: Proceedings of the 20th international conference on Computational Linguistics, Morristown, NJ, USA. ACL. [find](https://scholar.google.com/scholar?q=Hacioglu%2C+K%2E+2004%2E+Semantic+role+labeling+using+dependency+trees%2E+In+COLING+%E2%80%9904%3A+Proceedings+of+the+20th+international+conference+on+Computational+Linguistics%2C+Morristown%2C+NJ%2C+USA%2E+ACL%2E)
6. Morante, R. and B. Busser. 2007. ILK2: Semantic role labelling for Catalan and Spanish using TiMBL. In Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 183– 186. [find](https://scholar.google.com/scholar?q=Morante%2C+R%2E+and+B%2E+Busser%2E+2007%2E+ILK2%3A+Semantic+role+labelling+for+Catalan+and+Spanish+using+TiMBL%2E+In+Proceedings+of+the+4th+International+Workshop+on+Semantic+Evaluations+%28SemEval%2D2007%29%2C+pages+183%E2%80%93+)
7. Morante, R. 2008. Semantic role labeling tools trained on the Cast3LB-CoNLL-SemRol corpus. In Proceedings of the LREC 2008, Marrakech, Morocco. [find](https://scholar.google.com/scholar?q=Morante%2C+R%2E+2008%2E+Semantic+role+labeling+tools+trained+on+the+Cast3LB%2DCoNLL%2DSemRol+corpus%2E+In+Proceedings+of+the+LREC+2008%2C+Marrakech%2C+Morocco%2E)
8. Nivre, J., J. Hall, J. Nilsson, G. Eryigit, and S. Marinov. 2006. [Labeled pseudo–projective dependency parsing with support vector machines](https://md.research.elamur.ai/free/en/papers/nivre-2006-labeled-pseudo-projective-dependency-parsing/). In Proceedings of the Tenth Conference on Computational Natural Language Learning, CoNLL-X, New York City, NY, June.
9. Nivre, J., J. Hall, J. Nilsson, A. Chanev, G. Eryigit, S. Kübler, S. Marinov, and E. Marsi. 2007. Malt-Parser: a language-independent system for data-driven dependency parsing. Natural Language Engineering, 13(2):95–135. [find](https://scholar.google.com/scholar?q=Nivre%2C+J%2E%2C+J%2E+Hall%2C+J%2E+Nilsson%2C+A%2E+Chanev%2C+G%2E+Eryigit%2C+S%2E+Ku%CC%88bler%2C+S%2E+Marinov%2C+and+E%2E+Marsi%2E+2007%2E+Malt%2DParser%3A+a+language%2Dindependent+system+for+data%2Ddriven+dependency+parsing%2E+Natural+Language+Engin)
10. Nivre, J. 2006. Springer. Inductive Dependency Parsing. [find](https://scholar.google.com/scholar?q=Nivre%2C+J%2E+2006%2E+Springer%2E+Inductive+Dependency+Parsing%2E)
11. Surdeanu, Mihai, Richard Johansson, Adam Meyers, Lluı́s Màrquez, and Joakim Nivre. 2008. [The CoNLL-2008 shared task on joint parsing of syntactic and semantic dependencies](https://md.research.elamur.ai/free/en/papers/surdeanu-2008-conll-2008-shared-task-joint/). In Proceedings of the 12th Conference on Computational Natural Language Learning (CoNLL-2008).
12. Tjong Kim Sang, E., S. Canisius, A. van den Bosch, and T. Bogers. 2005. Applying spelling error correction techniques for improving semantic role labelling. In Proceedings of the Ninth Conference on Natural Language Learning (CoNLL-2005), Ann Arbor, MI. [find](https://scholar.google.com/scholar?q=Tjong+Kim+Sang%2C+E%2E%2C+S%2E+Canisius%2C+A%2E+van+den+Bosch%2C+and+T%2E+Bogers%2E+2005%2E+Applying+spelling+error+correction+techniques+for+improving+semantic+role+labelling%2E+In+Proceedings+of+the+Ninth+Conference+on+N)
13. van den Bosch, A., S. Canisius, W. Daelemans, I. Hendrickx, and E. Tjong Kim Sang. 2004. Memory-based semantic role labeling: Optimizing features, algorithm, and output. In Ng, H.T. and E. Riloff, editors, Proceedings of the Eighth Conference on Computational Natural Language Learning (CoNLL-2004), Boston, MA, USA. [find](https://scholar.google.com/scholar?q=van+den+Bosch%2C+A%2E%2C+S%2E+Canisius%2C+W%2E+Daelemans%2C+I%2E+Hendrickx%2C+and+E%2E+Tjong+Kim+Sang%2E+2004%2E+Memory%2Dbased+semantic+role+labeling%3A+Optimizing+features%2C+algorithm%2C+and+output%2E+In+Ng%2C+H%2ET%2E+and+E%2E+Riloff%2C+edi)

*Updated 1 November 2026*