SherLIiC: A Typed Event-Focused Lexical Inference Benchmark for Evaluating Natural Language Inference

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 902–914 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 902 SherLIiC: A Typed Event-Focused Lexical Inference Benchmark for Evaluating Natural Language Inference Martin Schmitt and Hinrich Sch¨utze Center for Information and Language Processing (CIS) LMU Munich, Germany martin@cis.lmu.de Abstract We present SherLIiC,1 a testbed for lexical in- ference in context (LIiC), consisting of 3985 manually annotated inference rule candidates (InfCands), accompanied by (i) ~960k un- labeled InfCands, and (ii) ~190k typed tex- tual relations between Freebase entities ex- tracted from the large entity-linked corpus ClueWeb09. Each InfCand consists of one of these relations, expressed as a lemmatized de- pendency path, and two argument placehold- ers, each linked to one or more Freebase types. Due to our candidate selection process based on strong distributional evidence, SherLIiC is much harder than existing testbeds because distributional evidence is of little utility in the classification of InfCands. We also show that, due to its construction, many of SherLIiC’s correct InfCands are novel and missing from existing rule bases. We evaluate a number of strong baselines on SherLIiC, ranging from se- mantic vector space models to state of the art neural models of natural language inference (NLI). We show that SherLIiC poses a tough challenge to existing NLI systems. 1 Introduction Lexical inference (LI) can be seen as a focused variant of natural language inference (NLI), also called recognizing textual entailment (Dagan et al., 2013). Recently, Gururangan et al. (2018) showed that annotation artifacts in current NLI testbeds distort our impression of the performance of state of the art systems, giving rise to the need for new evaluation methods for NLI. Glockner et al. (2018) investigated LI as a way of evaluating NLI systems and found that even simple cases are challenging to current systems. In this paper, we release SherLIiC, a testbed specifically designed for evaluating a sys- tem’s ability to solve the hard problem of modeling lexical entailment in context. 1https://github.com/mnschmit/SherLIiC (1) troponymy ORGF[A] is granting to EMPL[B] ⇒ ORGF[A] is giving to EMPL[B] (2) synonymy + ORGF[A] is supporter of ORGF[B] derivation ⇒ ORGF[A] is backing ORGF[B] (3) typical AUTH[A] is president of LOC[B] actions ⇒ AUTH[A] is representing LOC[B] (4) script PER[A] is interviewing AUTH[B] knowledge ⇒ PER[A] is asking AUTH[B] (5) common sense ORGF[A] claims LOC[B] knowledge ⇒ ORGF[A] is wanting LOC[B] Table 1: Examples of SherLIiC InfCands and NLI challenges they cover. ORGF=organization founder, EMPL=employer, AUTH=book author, LOC=location, POL=politician, PER=person. Levy and Dagan (2016) identified context- sensitive – as opposed to “context-free” – entail- ment as an important evaluation criterion and cre- ated a dataset for LI in context (LIiC). In their data, WordNet (Miller, 1995; Fellbaum, 2005) synsets serve as context for one side of a binary relation, but the other side is still instantiated with a sin- gle concrete expression. We aim to improve this setting in two ways. First, we type our relations on both sides, thus making …

Content truncated to 500 words