Coarse Lexical Frame Acquisition at the Syntax–Semantics Interface Using a Latent-Variable PCFG Model

Proceedings of the 7th Joint Conference on Lexical and Computational Semantics (*SEM), pages 130–141 New Orleans, June 5-6, 2018. c⃝2018 Association for Computational Linguistics Coarse Lexical Frame Acquisition at the Syntax–Semantics Interface Using a Latent-Variable PCFG Model Laura Kallmeyer* & Behrang Q. Zadeh* SFB991, Heinrich-Heine-Universit¨at D¨usseldorf kallmeyer@hhu.de zadeh@phil.hhu.de Jackie Chi Kit Cheung McGill University jcheung@cs.mcgill.ca Abstract We present a method for unsupervised lexi- cal frame acquisition at the syntax–semantics interface. Given a set of input strings de- rived from dependency parses, our method generates a set of clusters that resemble lex- ical frame structures. Our work is motivated not only by its practical applications (e.g., to build, or expand the coverage of lexical frame databases), but also to gain linguistic insight into frame structures with respect to lexical distributions in relation to grammatical struc- tures. We model our task using a hierarchical Bayesian network and employ tools and meth- ods from latent variable probabilistic context free grammars (L-PCFGs) for statistical infer- ence and parameter fitting, for which we pro- pose a new split and merge procedure. We show that our model outperforms several base- lines on a portion of the Wall Street Journal sentences that we have newly annotated for evaluation purposes. 1 Introduction We propose a method for building coarse lexi- cal frames automatically from dependency parsed sentences; i.e., without using any explicit seman- tic information as training data. The task involves grouping verbs that evoke the same frame (i.e., are considered to be the head of this frame) and fur- ther clustering their syntactic arguments into latent semantic roles. Hence, our target structures stand between FrameNet (Ruppenhofer et al., 2016) and PropBank (Palmer et al., 2005) frames. Similar to FrameNet and in contrast to PropBank, we as- sume a many-to-many relationship between verb types and frame types. But similar to PropBank, we aim to cluster syntactic arguments into gen- eral semantic roles instead of frame-specific slot *Both authors contributed equally to this work. types in FrameNet. This allows us to generalize across frames concerning semantic roles. As part of this, we study possible ways to automatically generate more abstract lexical-semantic represen- tations from lexicalized dependency structures. In our task, grouping verb tokens into frames requires not only distinguishing between differ- ent senses of verbs, but also identifying a range of lexical relationships (e.g., synonymy, opposite verbs, troponymy, etc.) among them. Hence (as Modi et al., 2012; Green et al., 2004), our prob- lem definition differs from most work on unsu- pervised fine-grained frame induction using verb sense disambiguation (e.g., Kawahara et al., 2014; Peng et al., 2017). Similarly, forming role clusters yields generalization from several alternate link- ings between semantic roles and their syntactic re- alization. Given, for instance, an occurrence of the verb pack and its syntactic arguments, not only do we aim to distinguish different senses of the verb pack (e.g., as used to evoke the FILLING frame, or the PLACING frame), but also to group these in- stances of ‘pack’ with other verbs that evoke …

Content truncated to 500 words