A Corpus of Encyclopedia Articles with Logical Forms

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1051–1060 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1051 A Corpus of Encyclopedia Articles with Logical Forms Nathan Ellis Rasmussen, William Schuler Department of Linguistics The Ohio State University {rasmussen.63,schuler}@ling.osu.edu Abstract People can extract precise, complex logical meanings from text in documents such as tax forms and game rules, but language processing systems lack adequate training and evaluation resources to do these kinds of tasks reliably. This paper describes a corpus of annotated typed lambda calculus translations for approximately 2,000 sentences in Simple English Wikipedia, which is assumed to constitute a broad-coverage domain for precise, complex descriptions. The corpus described in this paper contains a large number of quantifiers and interesting scoping configurations, and is presented specifically as a resource for quantifier scope disambiguation systems, but also more generally as an object of linguistic study. Keywords: quantifier scope, coreference, logical form, explanatory text, Wikipedia 1. Introduction People can extract precise, complex logical meanings from explanatory text in documents such as tax forms and game rules, but language processing systems lack adequate train- ing and evaluation resources to do these kinds of tasks re- liably. This paper describes a corpus of annotated typed lambda calculus translations (Church, 1940) for approx- imately 2,000 sentences in Simple English Wikipedia, which is assumed to constitute a broad-coverage domain for precise, complex descriptions. These typed lambda cal- culus expressions are intended to serve as a theory-neutral formal semantic representation, to which other representa- tions (Kamp, 1981; Heim, 1982) can be translated. More- over, output from systems trained on typed lambda calcu- lus representations like these may be run with only minor pre-processing as programs in a functional programming language such as Haskell or Ocaml as a way to evaluate in- put statements as goals in some world model. The corpus described in this paper contains a large number of quan- tifiers and interesting scoping configurations, and is pre- sented specifically as a resource for quantifier scope disam- biguation systems, and also more generally as an object of linguistic study. 2. Related Work The semantic task of determining all possible scopal read- ings of a sentence can be addressed with compositional rules, and the task of identifying the weakest readings can be done algorithmically (Koller and Thater, 2010). How- ever, the pragmatic task of identifying the preferred scop- ing has not, as yet, been reduced to a general-purpose algo- rithm.1 To model it statistically requires training data that incorporate the relevant psycholinguistic cues: text coher- ence (Dwivedi, 2013), linear order of scope-bearers, syn- tactic structure, choice of quantifiers (as each vs. every), and presumptions of background knowledge (AnderBois et al., 2012) similar to those found in an explanation. 1Highly successful algorithms are available for certain special cases (Evang and Bos, 2013; Schuler and Wheeler, 2014). VanLehn (1978) studied syntactic influences on quantifier scoping to improve scope disambiguation in the Lunar Sci- ences Natural Language Information System (Woods …

Content truncated to 500 words