22.07.2019

Linguistic Modeling and its Interfaces

Vortrag von Patrick Rebuschat und Simón Ruiz, 26. Juli

LEAD's Distinguished International Professor Patrick Rebuschat (Lancaster University, UK) and LEAD graduate Simón Ruiz will hold a talk presenting joint work with Padraic Monaghan (University of Amsterdam):

"The Role of Exposure Condition on the Cross Situational Learning of Vocabulary and Morphosyntax"
Friday, July 26, 10:15AM, Blochbau 1.13 (Wilhelmstr. 19)

ABSTRACT. The use of mixed-effects regression modeling as a tool for statistical  analysis of  second language  data has  been increasingly recommended in  recent years  (see, e.g.,  Cunnings, 2012;  Cunnings & Finlayson,  2015; Linck  &  Cunnings, 2015;  Murakami, 2016;  Plonsky, 2017; Plonsky  & Oswald, 2017)  because it enables  variability across individuals  and  across particular  stimuli  to  be included  in  the model. Thus, design effects as well as inter-individual differences can be revealed within a single analysis.  However,  a less  often exploited  property  of these  analyses  is  that the  more  immediate context in which a stimulus is processed can also be included, such as which structure has just been experienced by the individual  and its effect on current learning.

In this  paper, we illustrate  the potential  utility of this  form of analysis by  presenting an experimental  study on the  second language acquisition  of  vocabulary  and   morphosyntax.  In  this  artificial language  study, 90 participants learned nouns, verbs, adjectives, morphological markers, and word order by keeping track of cross-situational statistics. During learning, participants viewed two scenes each comprising two aliens performing a transitive action presented on a computer. Participants simultaneously heard a sentence, composed of two nouns, a verb,  two marker words indicating which noun was  the  subject and which the object, and optional adjectives describing the aliens’  colour. Word order varied across SOV and OSV. One of  the  scenes  was described  by  the  sentence, and  over training participants were required  to learn  co-occurrences between the sentence  and the target  scene (through sensitivity  to relations between  particular  words  and   their  referents). During testing, knowledge of words and of  word order was individuated. We manipulated global  properties  of  learning,  by varying  exposure  across  three conditions. An implicit condition  provided no information  about the language  structure and  no  feedback about performance. An  explicit condition provided  information about word order and the presence of the marker words.  Finally, a feedback condition provided no information about the language structure but feedback in the form of a ringing bell played  by  the computer  when  participants  correctly selected the target  scene. Thus, this study extends previous studies of explicit versus implicit learning  of grammar (e.g., Krashen, 1982; Nazari, 2013) by testing both grammatical and vocabulary learning, and by investigating the  additional feedback  condition, which  provides information to the participant about how their learning is proceeding, but without explicit information about the structure to be acquired.

We found that  learning was effective in all  conditions. However, the feedback condition resulted in better learning of vocabulary than both implicit and explicit instruction conditions, which did not differ. In contrast, word  order syntax  was better learned  in the  implicit and explicit conditions compared to  the feedback condition. Thus global properties of learning affected  acquisition of vocabulary and syntax, with  performance  on each  aspect  of  learning boosted  by  distinct instruction  conditions. To acquire  vocabulary,  information  about learning performance was  most effective, but to  acquire syntax, this feedback  on performance  was slightly  detrimental. Focusing  only on instructions for grammatical learning (e.g., Nazari, 2013) thus limits our understanding of how language learning proceeds. In terms of local properties of learning, we analysed whether the learner’s performance on  the previous  trial containing each  piece of information –  the nouns, the verb, and the adjectives – affected accuracy on the current learning trial. We found that the precise context of the participants’ learning of  verbs and adjectives,  but not nouns,  influenced current performance. These results  show the  importance not  only of  global instructions, but also contingently taking into account the learner’s recent learning  experience in  determining effectiveness of language learning. Our paradigm thus illustrates how both vocabulary and syntax can be learned in the laboratory, and using GLME analysis, we show how learning  different aspects of language  are affected  differently by instruction and by context of learning.