Studying word order through iterative shuffling

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 10351–10366 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 10351 Studying word order through iterative shuffling Nikolay Malkin1,2,5 Sameera Lanka3 Pranav Goel4,5 Nebojsa Jojic5 1Mila, Université de Montréal 2Yale University 3Microsoft 4University of Maryland 5Microsoft Research kolya.malkin@yale.edu, pgoel1@umd.edu, {sameera.lanka,jojic}@microsoft.com Abstract As neural language models approach human performance on NLP benchmark tasks, their advances are widely seen as evidence of an increasingly complex understanding of syntax. This view rests upon a hypothesis that has not yet been empirically tested: that word order encodes meaning essential to performing these tasks. We refute this hypothesis in many cases: in the GLUE suite and in various genres of En- glish text, the words in a sentence or phrase can rarely be permuted to form a phrase car- rying substantially different information. Our surprising result relies on inference by iterative shuffling (IBIS), a novel, efficient procedure that finds the ordering of a bag of words having the highest likelihood under a fixed language model. IBIS can use any black-box model without additional training and is superior to existing word ordering algorithms. Coalescing our findings, we discuss how shuffling infer- ence procedures such as IBIS can benefit lan- guage modeling and constrained generation. 1 Introduction Is a model’s understanding of syntax a precondition for its understanding of natural language? Recent work on large language models (Devlin et al., 2019; Tenney et al., 2019; Rogers et al., 2021) has made this a popular hypothesis. Yet, models that con- sume only bag-of-words features but rival those that understand syntax have surprised researchers time and again (Iyyer et al., 2015; Joulin et al., 2017). New concerns have emerged that natural language understanding benchmarks may not be challenging enough to make sentence structure rel- evant (McCoy et al., 2019; Niven and Kao, 2019). Syntax is an essential aspect of language (Chom- sky, 1965). Sentence structure can be quite impor- tant: two sentences with very different meanings may use the same set of words (Fig. 1). But how much does syntax, as realized in word order, matter in typical English text? Given the words that make The scared mouse chased the hungry cat. The hungry cat chased the scared mouse. Figure 1: Which word order is more likely? up a sentence, but not their order, is the order usu- ally recoverable? If so, word order rarely encodes more information than is found in the bag of words. In the past, linguists could not have answered this question empirically. Manually ordering words into sentences is too laborious, and when there are multiple orders that satisfy grammatical constraints, one needs a way to choose among them. With the power of large language models, we can reduce this question to a computational one and resolve both issues: given the bag of words, find the word order that is most likely under a trained LM. To make this search tractable, we develop inference by iterative shuffling (IBIS), a procedure inspired by techniques …

Content truncated to 500 words