You are viewing the SNL 2017 Archive Website. For the latest information, see the Current Website.

Poster E69, Friday, November 10, 10:00 – 11:15 am, Harborview and Loch Raven Ballrooms

A computational account of word representation and processing in bilingual individuals

Claudia Penaloza1, Uli Grasemann2, Risto Miikkulainen2, Swathi Kiran1;1Boston University, 2The University of Texas at Austin

Bilingualism is exponentially increasing worldwide due to mass immigration and globalization. Nonetheless, there are no guidelines for the optimal rehabilitation of bilingual adults with aphasia (BAA). The possible language combinations in BAA, their relative competency in the two languages, and the effect of brain damage on their bilingual language representation among other factors contribute to the paucity of research in this field. It is, however, unfeasible to examine these issues clinically without large scale longitudinal studies in this population. As a potential solution to this problem, we have developed a computational model that will make a systematic examination of lexical access and treatment effects in BAA possible. The model is based on DISLEX, a simulation of lexical access based on self-organizing maps. The maps are trained to represent either semantic or phonological symbols, and are linked by adaptive associative connections that translate between semantic and phonetic representations. Following the revised hierarchical model (Kroll & Stewart, 1994), the extended bilingual DISLEX model consists of three interconnected maps: one for semantic symbols, and one phonological map for L1 and L2. By varying the timing and intensity of network training, the model can simulate a bilingual language system in which language representations and lexical access vary by age, age of acquisition (AoA), and relative proficiency (e.g., AoA effects are simulated by delaying L2 training to match healthy bilinguals). Using our preliminary computational model as the starting point (Kiran, Grasemann, Sandberg, & Miikkulainen, 2013), we extended the basic architecture of the model to accommodate a larger semantic/lexical representation database. Using data from MTURK and another ongoing research collaboration (Kiran, CoPI, Title: Funding: ASHA multicultural activities grant), we incorporated a database of 651 unique words in English (and corresponding translations in Spanish). Each word belongs to one of 13 categories (e.g., animals, body parts, furniture) represented in the semantic space of the computational model. Corresponding English and Spanish phonetic representations (transcribed in IPA) are represented in the English and Spanish representations of the model. Additionally, each of the 651 items is represented by a set of semantic attributes (10-20 features per item) normed on healthy adults using MTURK. Thus, for each feature for each item (e.g. ant), we have a value of the probability of certainty of the feature being applicable to that particular item (e.g., moves: yes; 100%; has legs: yes, 94%; swims: no, 94%). The computational model is currently fully functional and can simulate healthy bilinguals. We have also collected data from healthy bilingual participants and utilized existing data from BAA to validate the computational architecture. In future research, the resulting models of healthy bilinguals can then be lesioned systematically to simulate the damage leading to bilingual aphasia. Retraining lesioned models in one language or the other could help understand the influence of pre-morbid language proficiency, AoA, and other factors on treatment outcomes.

Topic Area: Computational Approaches

Back to Poster Schedule