Semantic Frame Labeling with Target-based Neural Model

Proceedings of the 6th Joint Conference on Lexical and Computational Semantics (*SEM 2017), pages 91–96, Vancouver, Canada, August 3-4, 2017. c⃝2017 Association for Computational Linguistics Semantic Frame Labeling with Target-based Neural Model Yukun Feng1, Dong Yu1∗, Jian Xu2, ChunHua Liu1 1 Beijing Language and Culture University, 2 University of Science and Technology of China yukunfg@gmail.com,yudong@blcu.edu.cn, jianxu1@mail.ustc.edu.cn,chunhualiu596@gmail.com Abstract This paper explores the automatic learning of distributed representations of the tar- get’s context for semantic frame labeling with target-based neural model. We con- strain the whole sentence as the model’s input without feature extraction from the sentence. This is different from many pre- vious works in which local feature extrac- tion of the targets is widely used. This constraint makes the task harder, especial- ly with long sentences, but also makes our model easily applicable to a range of re- sources and other similar tasks. We evalu- ate our model on several resources and get the state-of-the-art result on subtask 2 of SemEval 2015 task 15. Finally, we extend the task to word-sense disambiguation task and we also achieve a strong result in com- parison to state-of-the-art work. 1 Introduction and Related Work Semantic frame labeling is the task of selecting the correct frame for a given target based on it- s semantic scene. A target is often called lexi- cal unit which evokes the corresponding seman- tic frame. The lexical unit can be a verb, ad- jective or noun. Generally, a semantic frame de- scribes how the lexical unit is used and specifies its characteristic interactions. There are many se- mantic frame resources, such as FrameNet (Bak- er et al., 1998), VerbNet (Schuler, 2006), Prop- Bank (Palmer et al., 2005) and Corpus Pattern Analysis (CPA) frames (Hanks, 2012). However, most existing frame resources are manually creat- ed, which is time-consuming and expensive. Au- tomatic semantic frame labeling can lead to the de- velopment of a broader range of resources. ∗The corresponding author Early works for semantic frame labeling main- ly focus on FrameNet, PropBank and VerbNet re- sources. But most of them focus only one re- source and rely heavily on feature engineering (e.g., Honnibal and Hawker 2005; Abend et al. 2008). Recently, there are some works on learn- ing CPA frames based on a new semantic frame resource, the Pattern Dictionary of English Verbs (PDEV) (El Maarouf and Baisa, 2013; El Maarouf et al., 2014). The above two works also rely on features and both are only tested on 25 verbs. Most works aim at constructing the context represen- tations of the target with explicit rules based on some basic features, e.g., Parts Of Speech (POS), Named Entities (NE) and dependency relations re- lated to the target. Currently, some deep learning models have been applied with dependency fea- tures. Hermann et al. (2014) used the direct de- pendents and dependency path to extract the con- text representation based on distributed word em- beddings on English FrameNet. Inspired by the work, Zhao et al. (2016) used a deep feed forward neural …

Content truncated to 500 words