Generalizing over Lexical Features: Selectional Preferences for Semantic Role Classification
Zapirain, Beñat and Agirre, Eneko and Màrquez, Llu'is

Article Structure

Abstract

This paper explores methods to alleviate the effect of lexical sparseness in the classification of verbal arguments.

Introduction

Semantic Role Labeling (SRL) systems usually approach the problem as a sequence of two subtasks: argument identification and classification.

Related Work

Automatic acquisition of selectional preferences is a relatively old topic, and will mention the most relevant references.

Selectional Preference Models

In this section we present all the variants for acquiring selectional preferences used in our study, and how we apply them to the SR classification.

Experimental Setting

The data used in this work is the benchmark corpus provided by the CoNLL-2005 shared task on SRL (Carreras and Marquez, 2005).

Results and Discussion

The results are presented in Table 1.

Conclusions and Future Work

We have empirically shown how automatically generated selectional preferences, using WordNet and distributional similarity measures, are able to effectively generalize lexical features and, thus, improve classification performance in a large-scale argument classification task on the CoNLL-2005 dataset.

Topics

distributional similarity

Appears in 13 sentences as: distributional similarities (3) Distributional similarity (1) distributional similarity (9)
In Generalizing over Lexical Features: Selectional Preferences for Semantic Role Classification
  1. The best results are obtained with a novel second-order distributional similarity measure, and the positive effect is specially relevant for out-of-domain data.
    Page 1, “Abstract”
  2. Distributional similarity has also been used to tackle syntactic ambiguity.
    Page 2, “Related Work”
  3. Pantel and Lin (2000) obtained very good results using the distributional similarity measure defined by Lin (1998).
    Page 2, “Related Work”
  4. The results over 100 frame-specific roles showed that distributional similarities get smaller error rates than Resnik and EM, with Lin’s formula having the smallest error rate.
    Page 2, “Related Work”
  5. Moreover, coverage of distributional similarities and Resnik are rather low.
    Page 2, “Related Work”
  6. Currently, there are several models of distributional similarity that could be used for selectional preferences.
    Page 2, “Related Work”
  7. More recently, Pado and Lapata (2007) presented a study of several parameters that define a broad family of distributional similarity models, including publicly available software.
    Page 2, “Related Work”
  8. Distributional SP models: Given the availability of publicly available resources for distributional similarity , we used 1) a ready-made thesaurus (Lin, 1998), and 2) software (Pado and Lapata, 2007) which we run on the British National Corpus (BNC).
    Page 2, “Selectional Preference Models”
  9. Regarding the selectional preference variants, WordNet—based and first-order distributional similarity models attain similar levels of precision, but the former are clearly worse on recall and F1.
    Page 3, “Results and Discussion”
  10. The second-order distributional similarity measures perform best overall, both in precision and recall.
    Page 4, “Results and Discussion”
  11. Regarding the similarity metrics, the cosine seems to perform consistently better for first-order distributional similarity , while J accard provided slightly better results for second-order similarity.
    Page 4, “Results and Discussion”

See all papers in Proc. ACL 2009 that mention distributional similarity.

See all papers in Proc. ACL that mention distributional similarity.

Back to top.

similarity measure

Appears in 6 sentences as: similarity measure (3) similarity measures (3)
In Generalizing over Lexical Features: Selectional Preferences for Semantic Role Classification
  1. The best results are obtained with a novel second-order distributional similarity measure , and the positive effect is specially relevant for out-of-domain data.
    Page 1, “Abstract”
  2. Pantel and Lin (2000) obtained very good results using the distributional similarity measure defined by Lin (1998).
    Page 2, “Related Work”
  3. We will refer to this similarity measure as simg‘n.
    Page 2, “Selectional Preference Models”
  4. We will refer to these similarity measures as simE-ZE and simi’gg hereinafter.
    Page 2, “Selectional Preference Models”
  5. The second-order distributional similarity measures perform best overall, both in precision and recall.
    Page 4, “Results and Discussion”
  6. We have empirically shown how automatically generated selectional preferences, using WordNet and distributional similarity measures , are able to effectively generalize lexical features and, thus, improve classification performance in a large-scale argument classification task on the CoNLL-2005 dataset.
    Page 4, “Conclusions and Future Work”

See all papers in Proc. ACL 2009 that mention similarity measure.

See all papers in Proc. ACL that mention similarity measure.

Back to top.

Role Labeling

Appears in 5 sentences as: role label (2) Role Labeling (2) role labels (1)
In Generalizing over Lexical Features: Selectional Preferences for Semantic Role Classification
  1. Our findings suggest that selectional preferences have potential for improving a full system for Semantic Role Labeling .
    Page 1, “Abstract”
  2. Semantic Role Labeling (SRL) systems usually approach the problem as a sequence of two subtasks: argument identification and classification.
    Page 1, “Introduction”
  3. This first step allows us to analyze the potential of selectional preferences as a source of semantic knowledge for discriminating among different role labels .
    Page 1, “Introduction”
  4. Given a target sentence where a predicate and several potential argument and adjunct head words occur, the goal is to assign a role label to each of the head words.
    Page 2, “Selectional Preference Models”
  5. The test set contains 4,134 pairs (covering 505 different predicates) to be classified into the appropriate role label .
    Page 3, “Experimental Setting”

See all papers in Proc. ACL 2009 that mention Role Labeling.

See all papers in Proc. ACL that mention Role Labeling.

Back to top.

Semantic Role

Appears in 5 sentences as: Semantic Role (2) semantic role (1) semantic roles (2)
In Generalizing over Lexical Features: Selectional Preferences for Semantic Role Classification
  1. We show how automatically generated selectional preferences are able to generalize and perform better than lexical features in a large dataset for semantic role classification.
    Page 1, “Abstract”
  2. Our findings suggest that selectional preferences have potential for improving a full system for Semantic Role Labeling.
    Page 1, “Abstract”
  3. Semantic Role Labeling (SRL) systems usually approach the problem as a sequence of two subtasks: argument identification and classification.
    Page 1, “Introduction”
  4. The application of selectional preferences to semantic roles (as opposed to syntactic functions) is more recent.
    Page 2, “Related Work”
  5. Other papers applying semantic preferences in the context of semantic roles , rely on the evaluation on pseudo tasks or human plausibility judgments.
    Page 2, “Related Work”

See all papers in Proc. ACL 2009 that mention Semantic Role.

See all papers in Proc. ACL that mention Semantic Role.

Back to top.

WordNet

Appears in 5 sentences as: WordNet (5)
In Generalizing over Lexical Features: Selectional Preferences for Semantic Role Classification
  1. Resnik (1993) proposed to model selectional preferences using semantic classes from WordNet in order to tackle ambiguity issues in syntax (noun-compounds, coordination, PP-attachment).
    Page 1, “Related Work”
  2. This happens more often with WordNet based models, which have a lower word coverage compared to distributional similarity—based models.
    Page 3, “Experimental Setting”
  3. The performance loss on recall can be explained by the worse lexical coverage of WordNet when compared to automatically generated thesauri.
    Page 4, “Results and Discussion”
  4. Examples of words missing in WordNet include abbreviations (e.g., Inc., Corp.) and brand names (e.g., Texaco, Sony).
    Page 4, “Results and Discussion”
  5. We have empirically shown how automatically generated selectional preferences, using WordNet and distributional similarity measures, are able to effectively generalize lexical features and, thus, improve classification performance in a large-scale argument classification task on the CoNLL-2005 dataset.
    Page 4, “Conclusions and Future Work”

See all papers in Proc. ACL 2009 that mention WordNet.

See all papers in Proc. ACL that mention WordNet.

Back to top.

PDF available from www.aclweb.org.
Send us an email