Starting from Scratch in Semantic Role Labeling

Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 989–998, Uppsala, Sweden, 11-16 July 2010. c⃝2010 Association for Computational Linguistics Starting From Scratch in Semantic Role Labeling Michael Connor University of Illinois connor2@uiuc.edu Yael Gertner University of Illinois ygertner@cyrus.psych.uiuc.edu Cynthia Fisher University of Illinois cfisher@cyrus.psych.uiuc.edu Dan Roth University of Illinois danr@illinois.edu Abstract A fundamental step in sentence compre- hension involves assigning semantic roles to sentence constituents. To accomplish this, the listener must parse the sentence, find constituents that are candidate argu- ments, and assign semantic roles to those constituents. Each step depends on prior lexical and syntactic knowledge. Where do children learning their first languages begin in solving this problem? In this pa- per we focus on the parsing and argument- identification steps that precede Seman- tic Role Labeling (SRL) training. We combine a simplified SRL with an un- supervised HMM part of speech tagger, and experiment with psycholinguistically- motivated ways to label clusters resulting from the HMM so that they can be used to parse input for the SRL system. The results show that proposed shallow rep- resentations of sentence structure are ro- bust to reductions in parsing accuracy, and that the contribution of alternative repre- sentations of sentence structure to suc- cessful semantic role labeling varies with the integrity of the parsing and argument- identification stages. 1 Introduction In this paper we present experiments with an au- tomatic system for semantic role labeling (SRL) that is designed to model aspects of human lan- guage acquisition. This simplified SRL system is inspired by the syntactic bootstrapping theory, and by an account of syntactic bootstrapping known as ’structure-mapping’ (Fisher, 1996; Gillette et al., 1999; Lidz et al., 2003). Syntactic bootstrap- ping theory proposes that young children use their very partial knowledge of syntax to guide sen- tence comprehension. The structure-mapping ac- count makes three key assumptions: First, sen- tence comprehension is grounded by the acquisi- tion of an initial set of concrete nouns. Nouns are arguably less dependent on prior linguistic knowl- edge for their acquisition than are verbs; thus chil- dren are assumed to be able to identify the refer- ents of some nouns via cross-situational observa- tion (Gillette et al., 1999). Second, these nouns, once identified, yield a skeletal sentence structure. Children treat each noun as a candidate argument, and thus interpret the number of nouns in the sen- tence as a cue to its semantic predicate-argument structure (Fisher, 1996). Third, children represent sentences in an abstract format that permits gener- alization to new verbs (Gertner et al., 2006). The structure-mapping account of early syn- tactic bootstrapping makes strong predictions, in- cluding predictions of tell-tale errors. In the sen- tence “Ellen and John laughed”, an intransitive verb appears with two nouns. If young chil- dren rely on representations of sentences as sim- ple as an ordered set of nouns, then they should have trouble distinguishing such sentences from transitive sentences. Experimental evidence sug- gests that they do: 21-month-olds mistakenly in- terpreted word order in …

Content truncated to 500 words