Robust textual inference via graph matching

Abstract

We present a system for deciding whether a given sentence can be inferred from text. Each sentence is represented as a directed graph (extracted from a dependency parser) in which the nodes represent words or phrases, and the links represent syntactic and semantic relationships. We develop a learned graph matching model to approximate entailment by the amount of the sentence's semantic content which is contained in the text. We present results on the Recognizing Textual Entailment dataset (Dagan et al., 2005), and show that our approach outperforms BagOf-Words and TF-IDF models. In addition, we explore common sources of errors in our approach and how to remedy them.

Exploring Robustness Enhancements for Logic-Based Passage Filtering

“…5 From the full set of 200 questions, all follow-up questions were eliminated since discourse processing is not of relevance here. Definition questions were omitted as well since knowing the logical correctness of an answer is not sufficient for deciding if it is suitable as a definition.…”

“…In this paper, we are interested in robustness-enhancing techniques which can be added to existing theorem provers with minor internal changes. Therefore we do not try to build an approximate inference engine by adopting an approximate graph matcher like [5] and adding support for logical rules -this would almost amount to building a prover from scratch. More suitable solutions are the extraction of useful information from failed proofs [6], combining logic-based and shallow features using machine learning [7,8], and finally relaxation techniques [4,6] which reduce the query by subsequently dropping literals until a proof of the simplified query succeeds.…”

Operations on Conversational Mind-Graphs

“…In a similar approach, [Haghighi et al, 2005] has developed a graph matching model for sentence inference from texts. Many related approaches regarding the graph representation for texts and documents have been proposed since last few years by [Mani, 1997], [Montes-y Gómez et al, 2000], [Schenker et al, 2003] and [Hensman, 2004] .…”

Author verification using a Graph-based Representation

“…On the other hand, in [5,6] different methods are described for extracting semantic graphs through the text, where each vertex corresponds to a word and the edges represent the semantic dependency between two words. In this representation, these papers propose the generation of knowledge through relationships established between the words and the semantic dependencies (or categories).…”