Semantic Frame Identification with Distributed Word Representations

Semantic Frame Identification with Distributed Word Representations Karl Moritz Hermann‡∗ Dipanjan Das† Jason Weston† Kuzman Ganchev† ‡Department of Computer Science, University of Oxford, Oxford OX1 3QD, United Kingdom † Google Inc., 76 9th Avenue, New York, NY 10011, United States karl.moritz.hermann@cs.ox.ac.uk {dipanjand,kuzman}@google.com jaseweston@gmail.com Abstract We present a novel technique for semantic frame identification using distributed rep- resentations of predicates and their syntac- tic context; this technique leverages auto- matic syntactic parses and a generic set of word embeddings. Given labeled data annotated with frame-semantic parses, we learn a model that projects the set of word representations for the syntactic context around a predicate to a low dimensional representation. The latter is used for se- mantic frame identification; with a stan- dard argument identification method in- spired by prior work, we achieve state-of- the-art results on FrameNet-style frame- semantic analysis. Additionally, we report strong results on PropBank-style semantic role labeling in comparison to prior work. 1 Introduction Distributed representations of words have proved useful for a number of tasks. By providing richer representations of meaning than what can be en- compassed in a discrete representation, such ap- proaches have successfully been applied to tasks such as sentiment analysis (Socher et al., 2011), topic classification (Klementiev et al., 2012) or word-word similarity (Mitchell and Lapata, 2008). We present a new technique for semantic frame identification that leverages distributed word rep- resentations. According to the theory of frame se- mantics (Fillmore, 1982), a semantic frame rep- resents an event or scenario, and possesses frame elements (or semantic roles) that participate in the ∗The majority of this research was carried out during an internship at Google. event. Most work on frame-semantic parsing has usually divided the task into two major subtasks: frame identification, namely the disambiguation of a given predicate to a frame, and argument iden- tification (or semantic role labeling), the analysis of words and phrases in the sentential context that satisfy the frame’s semantic roles (Das et al., 2010; Das et al., 2014).1 Here, we focus on the first sub- task of frame identification for given predicates; we use our novel method (§3) in conjunction with a standard argument identification model (§4) to perform full frame-semantic parsing. We present experiments on two tasks. First, we show that for frame identification on the FrameNet corpus (Baker et al., 1998; Fillmore et al., 2003), we outperform the prior state of the art (Das et al., 2014). Moreover, for full frame-semantic parsing, with the presented frame identification technique followed by our argument identification method, we report the best results on this task to date. Sec- ond, we present results on PropBank-style seman- tic role labeling (Palmer et al., 2005; Meyers et al., 2004; M`arquez et al., 2008), that approach strong baselines, and are on par with prior state of the art (Punyakanok et al., 2008). 2 Overview Early work in frame-semantic analysis was pio- neered by Gildea and Jurafsky (2002). Subsequent work in this area focused on either the FrameNet or PropBank frameworks, and research …

Content truncated to 500 words