Computational Linguistics & Phonetics Computational Linguistics & Phonetics Fachrichtung 4.7 Universität des Saarlandes Computational Psycholinguistics


Shakila Shayan

Visiting Researcher
Department of Computational Linguistics and Phonetics
Saarland University

I received my double degree PhD in Computer Science and in Cognitive Science from Indiana University in Bloomington, Indiana. I recently finished a postdoctoral fellowship at the Max Planck Institute for Psycholinguistics in Netherlands.


Acquisition of Verb Argument Structure

I am interested in language(s) and studying how they are learned across different cultures. My primary research interest, involves children's learning of the verb argument structure. Among many theoretical and experimental approaches to the problem of learning verb constructions, several researchers have relied on cross-situational comparison and in particular Structure Mapping Theory (Gentner 1982) as a processing mechanism for learning verbs and their argument structure (e.g., Fisher 2000, 2006; Tomasello 2003; Childers 2009). In this approach they have mainly applied the theory in aligning abstract semantic knowledge with the abstract syntactic knowledge (e.g., Agent Act-on Patient mapped to Subject Verb Object) without discussing where the abstract semantic knowledge comes from. In my dissertation I tried to provide a specific mechanistic account on how structure mapping can be applied to learning about semantic roles. This goal was pursued both experimentally and computationally.

In a set of experimental studies I investigated the emergence of Agent and Patient roles in English speaking children 3-5 years of age. The two major issues I tried to address were the nature of children's knowledge and the mechanism by which this knowledge could be acquired. I approached these issues by focusing on two aspects of Agent-Patient relations; a) The role of language in boosting a generalization process over all Agents (or all Patients) that are similarly marked in the syntax and b) the role of individual components of an action sequence in making all Agents (or all Patients) relationally similar between and within action categories. In the context of an analogy task that promoted comparison, I tested children's understanding of the relational similarity between novel Agents and between novel Patients in simple but novel transitive events. I developed a set of familiar and novel actions in the form of animations. Children's task was to match participants across events based on their relational similarity (role in the action). Results showed that children's understanding of semantic roles is dependent on the structure of the language and the components that comprise a simple action sequence. Children's ability to match roles across events was fragile and context-dependent; despite apparent competence with the syntactic frame of familiar verbs that was assessed through familiarization trials of the experiments, children had difficulty identifying Agent-Patient roles in a novel context. These findings are in line with the view that children learn action categories and their corresponding verb argument structure on a case-by-case basis using a multi sensory input and embodied cognitive skills.

Categories and Concepts Across Languages

Cross-cultural differences in conceptual development is another line of my research. I am interested in whether and how language could shape our conceptual knowledge. During my postdoctoral time at Max Planck I worked within the Categories Across Languages project bridging between the Acquisition group and the Language and Cognition group. We studied children's developing knowledge of language across cultures and investigated differences in their categorization behavior that might be language driven. Particularly I was involved in several sub projects, on Language of Perception and Development of Perceptual Categories.

In one project we looked at the development of cross-modal associations of pitch to other dimensions. Pitch is described metaphorically in some languages, extending vocabulary from the domain of space (low-high), size (big-small, also thick-thin), and strength quality(strong-weak). The question we address in this project is whether people make such associations at perceptual level: Does high (low) pitch go with a thick snake or a thin snake? We also ask whether these non-linguistic associations are language specific.