Semantic Proto-Roles

A substantial amount of both computational and linguistic research on event representations has focused on categorical, coarse thematic roles such as “Agent” and “Patient”.  Despite this, there is evidence going back to work by Dowty from 1991 that this is the wrong grain size to look for thematic information: the appearance of roles like Agent represents a conglomeration of finer-grained thematic properties such as causation, volition, change-of-state, etc.  Decomposing thematic roles into fine-grained proto-properties provides both theoretical and pratical gains: not only does it lead to a better account of the mapping of semantic representations to syntax, it also leads to more tractable semantic annotation tasks.  In this line of work we have developed a semantic proto-role annotation task, and shown that it readily leads to quality annotations with minimal training in crowd-sourcing situations.

Papers

1. Drew Reisinger, Rachel Rudinger, Francis Ferraro, Craig Harman, Kyle Rawlins, Benjamin Van Durme.  Semantic proto-roles.  Transactions of the ACL 3, 2015.

We present the first large-scale, corpus based verification of the seminal work of Dowty on the notion of thematic proto-roles. Our results demonstrate both the need for and the feasibility of a property-based annotation scheme of semantic relationships, as opposed to the currently dominant notion of categorical roles.

Paper link: TACL site, PDF download

Slides from EMNLP 2015: PDF download

Data: PropBank annotated with Semantic Proto-Role property judgments (English)

2. Aaron Steven White, Drew Reisinger, Rachel Rudinger, Kyle Rawlins, Benjamin Van Durme.  Computational linking theory. Manuscript, July 2016.

A linking theory explains how verbs’ semantic arguments are mapped to their syntactic arguments—the inverse of the semantic role labeling task from the shallow semantic parsing literature. In this paper, we introduce the computational linking theory (CLT) framework and use it to implement two classic linking theories. We then propose the semantic proto-role linking model (SPRoLiM), which implements a generalization of Dowty’s seminal proto-role theory. We use SPRoLiM to induce natural semantic roles that we evaluate using the CLT framework. We show that SPRoLiM provides a viable theoretically motivated alternative to previous work in semantic role induction.

Paper link: arXiv, LingBuzz

Slides from a talk by Kyle Rawlins at UPenn Linguistics in Feb 2016, covering both the TACL paper and material from this manuscript: UPenn Proto-roles talk

3. Aaron Steven White, Drew Reisinger, Keisuke Sakaguchi, Tim Vieira, Sheng Zhang, Rachel Rudinger, Kyle Rawlins, Benjamin Van Durme. Universal decompositional semantics on universal dependencies. Proceedings of EMNLP, 2016.

We present a framework for augmenting data sets from the Universal Dependencies project with Universal Decompositional Semantics. Where the Universal Dependencies project aims to provide a syntactic annotation standard that can be used consistently across many languages as well as a collection of corpora that use that standard, our extension has similar aims for semantic annotation. We describe results from annotating the English Universal Dependencies treebank, dealing with word senses, semantic roles, and event properties.

Paper link: PDF download

Data: Universal Dependencies (v1.2) annotated with Semantic Proto-Role property judgments, WordNet senses, and event realis (English)

Code: PredPatt

4. Adam Teichert, Adam Poliak, Benjamin Van Durme and Matthew Gormley. Semantic Proto-Role Labeling. AAAI. 2017.

The semantic function tags of Bonial, Stowe, and Palmer (2013) and the ordinal, multi-property annotations of Reisinger et al. (2015) draw inspiration from Dowty’s semantic proto-role theory. We approach proto-role labeling as a multi-label classification problem and establish strong results for the task by adapting a successful model of traditional semantic role labeling.

5. Aaron Steven White, Kyle Rawlins, and Benjamin Van Durme. The Semantic Proto-Role Linking Model. EACL. 2017.

We propose the semantic proto-role linking model, which jointly induces both predicate-specific semantic roles and predicate-general semantic proto-roles based on semantic proto-role property likelihood judgments. We use this model to empirically evaluate Dowty’s thematic proto-role linking theory.

Presentations

Past

  • Prague, 2014, Fred Jelinek Memorial Workshop.
  • EMNLP, 2015.  Semantic proto-roles. (Ben Van Durme).
  • MACSIM, 2015.  Semantic proto-roles. (Drew Reisinger)
  • UPenn Linguistics, 2016.  Assessing (proto-)roles. (Kyle Rawlins)
  • EMNLP, 2016. Universal decompositional semantics on universal dependencies. (Aaron Steven White)
  • AAAI, 2017. Semantic Proto-Role Labeling. (Adam Teichert)

Upcoming

  • EACL, 2017. The Semantic Proto-Role Linking Model. (Aaron Steven White)