Input and Age‐Dependent Variation in Second Language Learning: A Connectionist Account

Abstract

Language learning requires linguistic input, but several studies have found that knowledge of second language (L2) rules does not seem to improve with more language exposure (e.g., Johnson & Newport, 1989). One reason for this is that previous studies did not factor out variation due to the different rules tested. To examine this issue, we reanalyzed grammaticality judgment scores in Flege, Yeni‐Komshian, and Liu's (1999) study of L2 learners using rule‐related predictors and found that, in addition to the overall drop in performance due to a sensitive period, L2 knowledge increased with years of input. Knowledge of different grammar rules was negatively associated with input frequency of those rules. To better understand these effects, we modeled the results using a connectionist model that was trained using Korean as a first language (L1) and then English as an L2. To explain the sensitive period in L2 learning, the model's learning rate was reduced in an age‐related manner. By assigning different learning rates for syntax and lexical learning, we were able to model the difference between early and late L2 learners in input sensitivity. The model's learning mechanism allowed transfer between the L1 and L2, and this helped to explain the differences between different rules in the grammaticality judgment task. This work demonstrates that an L1 model of learning and processing can be adapted to provide an explicit account of how the input and the sensitive period interact in L2 learning.

Language ERPs reflect learning through prediction error propagation

“…In most of these models, turning off the computation of summary signals of mean activation or activation update disrupts the model's ability to explain ERPs, but no other functions are affected (e.g., comprehension of meaning can still take place). In the Error Propagation account, turning off the error computation that supports ERPs also prevents the model from learning, which means that it can no longer explain adaptation effects like structural priming (Chang et al, 2006) or second language acquisition in adults (Janciauskas & Chang, 2018). In contrast, deactivating the computation of the Semantic Update in the model of Rabovsky et al (2018) would not affect any of the comprehension results obtained with the Sentence Gestalt model in St.…”

“…This complex architecture is motivated by the need to explain a range of different behaviors from production (structural priming, Chang et al, 2006; heavy NP shift, word order effects, Chang, 2009;aphasia, Chang, 2002) and acquisition studies (auxiliary inversion, Fitz & Chang, 2017; acquisition of verb classes, Twomey et al, 2014;accessibility hierarchy, Fitz et al, 2011). It has also been applied to learn syntactic constraints from different languages (German, Chang, Baumann, Pappert, & Fitz, 2015;Japanese, Chang, 2009; English-learning Korean speakers, Janciauskas & Chang, 2018). In this work, we are testing whether ERP effects in comprehension can arise out of an architecture that was designed originally for production and acquisition.…”

“…Since the model does next word prediction, all syntactic knowledge is first encoded into the weights to the NEXTWORD layer and only later are syntactic categories developed in the COMPRESS and HIDDEN layers that provide a more abstract encoding of the rules of the language. Janciauskas and Chang (2018) argued that L2 learners make greater use of their lexical system for syntactic distinctions in the L2, so that might help to explain the persistent L2 N400 in bilingual adults (Weber & Lavric, 2008). Thus, the model can help to explain developmental N400 effects for syntactic stimuli in children and L2 speakers.…”

Towards computational models of multilingual sentence processing

“…It requires no architectural adaptations, apart from the minimal addition of language-control units that steer the network towards producing a sentence in one of the two languages. Janciauskas and Chang (2018) used the bilingual Dual-path model to study the effects of L2 age of acquisition (AoA) and length of L2 exposure. They first trained the model on Korean-like sentences.…”

Simulating Spanish-English Code-Switching: El Modelo Está Generating Code-Switches

“…Several NLP applications have emerged, e.g., to detect code-switches (Solorio and Liu, 2008;Guzmán et al, 2017), or to automatically recognize code-switched speech (Yılmaz et al, 2016;Gonen and Goldberg, 2018). Moreover, there are a small number of cognitive computational models relevant to code-switching: Filippi et al (2014) developed a model of code-switched word production and Janciauskas and Chang (2018), while simulating age of acquisition effects on native Korean speakers of English, reported that the models that had been exposed to English later produced code-switches, i.e., occasionally used Korean words in their predominantly English production.…”