Using novel word context measures to predict human ratings of lexical proficiency

Introduction While most researchers agree that lexical proficiency is an important component of second language (L2) language competence and academic achievement (Alderson, 2005; Daller, Van Hout, & Treffers-Daller, 2003; Laufer, 1992), the construct of lexical proficiency itself is still poorly understood and the field of L2 research lacks a unified theory of vocabulary acquisition (David, 2008; Schmitt, 2010). This is troubling given the need to understand how L2 lexicons develop in order to allow principled decisions regarding language pedagogy, student placement, and curricula. Past studies investigating L2 lexical proficiency have examined the intrinsic difficulty of lexical items (e.g., Laufer, 1997), the development of lexical automaticity (e.g., Hulstijn, Van Gelderen, & Schoonen, 2009), receptive vs. productive lexical knowledge (e.g., Melka, 1997), and the distinction between breadth and depth of knowledge (e.g., Read, 2000). Another approach to conceptualizing and assessing L2 lexical proficiency examines the manner in which L2 lexical items are stored, processed, and retrieved from the mental lexicon (Aitchison, 1994). The assumption behind such an approach is that newly acquired lexical items will need to assimilate into a network of already known words, resulting in restructuring of the network as whole. Lexical proficiency is thus understood as the ability to not only differentiate between semantically related words, but to recognize the variety of ways in which lexical items may be connected to one another (Read, 2004; Singleton, 1999). Presumably, the L2 lexicon strengthens as learners develop stronger links between items and are able to more easily accommodate new words in the network (Haastrup & Henriksen, 2000). The traditional approach to investigating mental lexicons analyzes online language processing in order to gain insights into the mental lexicon and how lexical items are stored, processed, and retrieved (e.g., Conklin & Schmitt, 2008; Ellis & Beaton, 1993; Laufer, 1997). While word frequency is generally considered one of the best predictors of language processing (Balota & Chumbley, 1984; Whaley, 1978), in the current study we investigate the variability of word context to better understand lexical proficiency from a network perspective. A promising method for better understanding L2 lexical proficiency lies in the use of natural language processing (NLP, Meurers, 2013) tools, such as the Tool for the Automatic Analysis of Lexical Sophistication (Kyle & Crossley, 2015), Coh-Metrix (Graesser, McNamara, Louwerse, & Cai, 2004) and AntWordProfiler (Anthony, 2014). Such NLP analytics allow researchers and educators to operationalize many of the constructs related to lexical proficiency and quantify them in learner-produced data. The aim of the current study is to determine if computational indices related to associative, lexical, and semantic word context can increase our understanding of L2 lexical knowledge. To do so, we analyze the spoken lexical output of both L2 English learners and native English speakers in relation to human ratings of lexical proficiency using five novel indices made available in a recently updated version of TAALES (version 2.0). The key questions that motivate our study are as follows: * What is the relationship between word context and human ratings of lexical proficiency? * Can measures of lexical, semantic, and associative word context predict human ratings of lexical proficiency? Our goal is to predict human ratings of holistic lexical proficiency in spoken language to better understand the construct of lexical proficiency and the role that word context may play in identifying and predicting lexical proficiency in L2 populations. Background Natural language processing analytics and lexical proficiency An emerging approach to better understanding lexical proficiency involves the computational analysis of language produced by language learners (Meurers, 2013). …