What determines the order of adjectives in English? Comparing efficiency-based theories using dependency treebanks

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 2003–2012 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 2003 What Determines the Order of Adjectives in English? Comparing Efficiency-Based Theories Using Dependency Treebanks Richard Futrell University of California, Irvine rfutrell@uci.edu William Dyer Oracle Corporation william.dyer@oracle.com Gregory Scontras University of California, Irvine g.scontras@uci.edu Abstract We take up the scientific question of what determines the preferred order of adjectives in English, in phrases such as big blue box where multiple adjectives modify a following noun. We implement and test four quantita- tive theories, all of which are theoretically mo- tivated in terms of efficiency in human lan- guage production and comprehension. The four theories we test are subjectivity (Scon- tras et al., 2017), information locality (Futrell, 2019), integration cost (Dyer, 2017), and in- formation gain, which we introduce. We eval- uate theories based on their ability to predict orders of unseen adjectives in hand-parsed and automatically-parsed dependency treebanks. We find that subjectivity, information locality, and information gain are all strong predictors, with some evidence for a two-factor account, where subjectivity and information gain reflect a factor involving semantics, and information locality reflects collocational preferences. 1 Introduction Across languages, there exist strong and stable constraints on the order of adjectives when more than one is used to modify a noun (Dixon, 1982; Sproat and Shih, 1991). For example, in English, big blue box sounds natural and appears relatively frequently in corpora, while blue big box sounds less natural and occurs less frequently (Scontras et al., 2017). In this paper, we take up the scien- tific question of what explains these constraints in natural language. To do so, we implement quanti- tative models that have been proposed in previous literature as explanations for these constraints, and compare their accuracy in predicting adjective or- dering data in parsed corpora of English1. In the last few years, adjective order has become a crucial testing ground for quantitative theories 1All code and data are available at https://github. com/langprocgroup/adjorder. of syntax. These theories provide mathematical models that can describe the distribution of words in sentences and the way those words combine to yield the meaning of a sentence, in a way that cap- tures the fine-grained quantitative patterns observ- able in large text datasets (Manning, 2003; Bres- nan et al., 2007; Chen and Ferrer-i-Cancho, 2019). Quantitative syntactic theories are often efficiency-based, meaning that they model word distributions as the result of a process that tries to maximize information transfer while minimizing some measure of cognitive cost; as a result, they often use the mathematical language of informa- tion theory. Such theories promise not only to describe distributions of words, but also to explain why they take the shape they do, by viewing human language as an efficient code subject to appropriate constraints. This work informs NLP by providing a theory of language structure that integrates with data-driven, optimization-based machine learning models. Adjective order is a fruitful empirical target for quantitative theories of …

Content truncated to 500 words