Detection and Resolution of Verb Phrase Ellipsis

Linguistic Issues in Language Technology – LiLT Detection and Resolution of Verb Phrase Ellipsis Marjorie McShane Petr Babkin Published by CSLI Publications LiLT volume 13, issue 1 2016 Detection and Resolution of Verb Phrase Ellipsis Marjorie McShane, Rensselaer Polytechnic Institute, Petr Babkin, Rensselaer Polytechnic Institute Abstract Verb phrase (VP) ellipsis is the omission of a verb phrase whose mean- ing can be reconstructed from the linguistic or real-world context. It is licensed in English by auxiliary verbs, often modal auxiliaries: She can go to Hawaii but he can’t [e]. This paper describes a system called ViPER (VP Ellipsis Resolver) that detects and resolves VP ellipsis, relying on linguistic principles such as syntactic parallelism, modality correlations, and the delineation of core vs. peripheral sentence con- stituents. The key insight guiding the work is that not all cases of el- lipsis are equally difficult: some can be detected and resolved with high confidence even before we are able to build systems with human-level semantic and pragmatic understanding of text. 1 LiLT Volume 13, Issue 1, 2016. Detection and Resolution of Verb Phrase Ellipsis. Copyright c ⃝ 2016, CSLI Publications. 2 / LiLT volume 13, issue 1 2016 1 Introduction Verb phrase (VP) ellipsis is the omission of a VP whose meaning can be reconstructed from the linguistic or real-world context. For example, the clause he could in (1) cannot be fully interpreted without recon- structing the elided VP as teach illiterate women to read and write.1 (1) Aid workers in war-ravaged Kabul were stunned when a toddler from a poor family offered to teach illiterate women to read and write – and then promptly proved he could [e]. Ellipsis should be resolved by language processing systems because un- resolved ellipses mask information that is available to a human reader. Ellipsis has not been treated much by NLP systems. Bos and Spe- nader 2011 assess why: “First, from a purely practical perspective, automatically locating el- lipsis and their antecedents is a hard task, not subsumed by ordi- nary natural language processing components. Recent empirical work (Hardt 1997; Nielsen 2005) indeed confirms that VPE identification is difficult. Second, most theoretical work begins at the point at which the ellipsis example and the rough location of its antecedent are already identified, focussing on the resolution task” (pp. 464-465). Certainly, both of the subtasks of ellipsis processing – detection and resolution – can be very difficult, so it is unrealistic to expect a full- coverage, near-term solution for open text. However, although some instances of VP ellipsis require human-level reasoning, not all are so dif- ficult. In fact, it is possible to automatically detect which cases are sim- pler and resolve them with high precision, which is exactly what ViPER (VP Ellipsis Resolver) does. So, like many NLP systems, ViPER treats only a subset of instances; but unlike many systems, ViPER automat- ically selects the subset it knows how to treat and processes those in- stances fully automatically, without the need for manual corpus anno- tation. …

Content truncated to 500 words