The Role of Implicit Argumentation in Nominal SRL

Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the ACL, pages 146–154, Boulder, Colorado, June 2009. c⃝2009 Association for Computational Linguistics The Role of Implicit Argumentation in Nominal SRL Matt Gerber Dept. of Computer Science Michigan State University gerberm2@msu.edu Joyce Y. Chai Dept. of Computer Science Michigan State University jchai@cse.msu.edu Adam Meyers Dept. of Computer Science New York University meyers@cs.nyu.edu Abstract Nominals frequently surface without overtly expressed arguments. In order to measure the potential benefit of nominal SRL for down- stream processes, such nominals must be ac- counted for. In this paper, we show that a state-of-the-art nominal SRL system with an overall argument F1 of 0.76 suffers a perfor- mance loss of more than 9% when nominals with implicit arguments are included in the evaluation. We then develop a system that takes implicit argumentation into account, im- proving overall performance by nearly 5%. Our results indicate that the degree of implicit argumentation varies widely across nominals, making automated detection of implicit argu- mentation an important step for nominal SRL. 1 Introduction In the past few years, a number of studies have focused on verbal semantic role labeling (SRL). Driven by annotation resources such as FrameNet (Baker et al., 1998) and PropBank (Palmer et al., 2005), many systems developed in these studies have achieved argument F1 scores near 80% in large-scale evaluations such as the one reported by Carreras and M`arquez (2005). More recently, the automatic identification of nominal argument structure has received increased attention due to the release of the NomBank cor- pus (Meyers, 2007a). NomBank annotates predicat- ing nouns in the same way that PropBank annotates predicating verbs. Consider the following example of the verbal predicate distribute from the PropBank corpus: (1) Freeport-McMoRan Energy Partners will be liquidated and [Arg1 shares of the new company] [Predicate distributed] [Arg2 to the partnership’s unitholders]. The NomBank corpus contains a similar instance of the deverbal nominalization distribution: (2) Searle will give [Arg0 pharmacists] [Arg1 brochures] [Arg1 on the use of prescription drugs] for [Predicate distribution] [Location in their stores]. This instance demonstrates the annotation of split ar- guments (Arg1) and modifying adjuncts (Location), which are also annotated in PropBank. In cases where a nominal has a verbal counterpart, the inter- pretation of argument positions Arg0-Arg5 is con- sistent between the two corpora. In addition to deverbal (i.e., event-based) nomi- nalizations, NomBank annotates a wide variety of nouns that are not derived from verbs and do not de- note events. An example is given below of the parti- tive noun percent: (3) Hallwood owns about 11 [Predicate %] [Arg1 of Integra]. In this case, the noun phrase headed by the predicate % (i.e., “about 11% of Integra”) denotes a fractional part of the argument in position Arg1. Since NomBank’s release, a number of studies have applied verbal SRL techniques to the task of nominal SRL. For example, Liu and Ng (2007) re- ported an argument F1 of 0.7283. Although this result is encouraging, it does not take …

Content truncated to 500 words