Is there a place for logic in recognizing textual entailment

Is there a place for logic in recognizing textual entailment? Johan Bos1 From a purely theoretical point of view, it makes sense to approach recognizing textual entailment (RTE) with the help of logic. After all, entailment matters are all about logic. In practice, only few RTE sys- tems follow the bumpy road from words to logic. This is probably be- cause it requires a combination of robust, deep semantic analysis and logical inference—and why develop something with this complexity if you perhaps can get away with something simpler? In this article, with the help of an RTE system based on Combinatory Categorial Grammar, Discourse Representation Theory, and first-order theorem proving, we make an empirical assessment of the logic-based approach. High precision paired with low recall is a key characteristic of this sys- tem. The bottleneck in achieving high recall is the lack of a systematic way to produce relevant background knowledge. There is a place for logic in RTE, but it is (still) overshadowed by the knowledge acquisi- tion problem. 1 Introduction Recognizing textual entailment—predicting whether one text entails another—is a task that embraces everything that needs to be accom- plished in natural language understanding. In the past, textual entail- ment was limited to the domain of formal semanticists, who used it as an illustrational device to show that certain natural language inferences hold or not (Gamut 1991; Chierchia and McConnell-Ginet 1991; Kamp and Reyle 1993; Heim and Kratzer 1998). By now, however, recognizing textual entailment (RTE, henceforth) is viewed by many as a key task in the area of natural language processing (Dagan et al. 2006). 1University of Groningen 27 LiLT Volume 9 Perspectives on Semantic Representations for Textual Inference. Copyright c ⃝ 2014, CSLI Publications. 28 / Johan Bos In the early developments of approaches to RTE it soon became clear that RTE is an extremely difficult task: simple baseline systems based on textual surface features are hard to outperform by more sophisti- cated systems. Not only does one need a robust component that gives an accurate analysis of text, the use of external resources to inform the inference process are also essential to achieve a good performance on the standard RTE data sets. Various approaches to RTE have been proposed, ranging from “shal- low” methods working directly on the surface features of texts, to “deep” methods using sophisticated linguistic analyses. The formalism proposed in this article belongs in the latter category, and works by de- termining textual inferences on the basis of deductive logical inference. The idea is simple and rooted in the formal approaches to natural lan- guage semantics mentioned before: we translate the texts into logical formulas, and then use (classical) logical inference to find out whether one text entails the other or the other way around, whether they are consistent or contradictory, and so on. Even though this idea itself sounds simple, its execution is not. In this article we describe a framework for textual inference based on first-order logic and formal theory. It …

Content truncated to 500 words