Semantic Kernels for Semantic Parsing

Abstract

We present an empirical study on the use of semantic information for Concept Segmentation and Labeling (CSL), which is an important step for semantic parsing. We represent the alternative analyses output by a state-of-the-art CSL parser with tree structures, which we rerank with a classifier trained on two types of semantic tree kernels: one processing structures built with words, concepts and Brown clusters, and another one using semantic similarity among the words composing the structure. The results on a corpus from the restaurant domain show that our semantic kernels exploiting similarity measures outperform state-of-the-art rerankers.

Learning to Differentiate Better from Worse Translations

“…This would also include more semantic information, e.g., in the form of Brown clusters or using semantic similarity between the words composing the structure calculated with latent semantic analysis (Saleh et al, 2014b).…”

SMARTies: Sentiment Models for Arabic Target entities

“…Entity Clusters It has been shown consistently that semantic word clusters improve the performance of named entity recognition (Täckström et al, 2012;Zirikly and Hagiwara, 2015;Turian et al, 2010) and semantic parsing (Saleh et al, 2014); we are not aware of such work for identifying entity targets of sentiment.…”

Syntactic and Semantic Structures for Relation Extraction

“…In order to propose a framework that can address the above three challenges, we have identified the Tree kernel representation to be a solid foundation for our work as it allows us to capture a variety of information including semantic concepts, words, POS tags, shallow and full syntax, dependency parsing, and discourse trees (Xu et al, 2013;Saleh et al, 2014;Zhou et al, 2010;Nguyen et al, 2009;Bunescu et al, 2005). In this study, we will deal with the above three challenges by exploiting the tree kernel structure as follows:…”

“…named entities concepts and syntactic, e.g. explicit relation nodes (Moschitti, 2006;Zhou et al, 2010;Saleh et al, 2014) for relation extraction as outlined in the following section.…”

“…Also, RE methods that have employed bootstrapping (Xu et al, 2007;Xu et al, 2010) are limited in their application to new domains due to their focus on relations that are domain specific. We believe that the tree kernel can be a rich syntactic and semantic structure that includes semantic concepts, words, POS tags, shallow and full syntax, dependency parsing and discourse tree (Xu et al, 2013;Saleh et al, 2014;Zhou et al, 2010;Nguyen et al, 2009), which can help to improve the performance when identifying pieces of relation information in a sentence. We suggest that the tree kernel has potential for improving the performance of ER techniques.…”