Somsukla Banerjee & Achla M. Raina & Harish Karnick - Modelling Concept Space

The present paper proposes a model for generation and organization of concepts in the human cognitive architecture, focusing on language processing. Presenting empirical evidence from cognitive neuropsychology and language acquisition research, we argue for autonomy of concepts, positing amodal conceptual representations abstracted from representations of experience grounded in specific modalities. Concepts are defined as a non-empty collection of attributes and/or functions a1, …, an or f1, …, fm. The set of attributes and functions forms its representational structure. An attribute is a three tuple <name, value, type> and a function is dynamic in nature consisting of the three tuple <argument, process, returned value>. It is possible to store the knowledge of individual concepts in the Concept Space in a declarative manner or generate them with the use of a finite set of operations. The operations include the addition, removal, widening, narrowing and projection of attributes and addition or removal of functions. Any operation or sequence of operations successively applied to a concept leads to the generation of a new concept. It is proposed that the Concept Space evolves through these operations working on the existing representations, leading to generation of new concepts.

The paper outlines some of these operations and illustrates their functioning. We posit a non-hierarchical organization of concepts in which the derivational links across concepts are captured through the attribute derivation_account. Further, we claim language to be a part of the concept system: linguistic units such as words/signs and other non-compositional expressions are viewed as Word Concepts. Word Concepts are also concepts. The attribute linguistic_referent is the link between a concept and the corresponding word concept(s). In a similar manner, the attribute conceptual_referent is the link between a word concept and the corresponding concept(s). Production and comprehension of language and language development are claimed to involve creation, instantiation and activation of concepts and the referential links across them.
The predictions of the model regarding bilingual processing of language are discussed for the issues of accessing concepts, processing cross-linguistic polysemy and representing and accessing multi-word expressions. These predictions are tested through a series of empirical studies conducted on Hindi-English bilinguals. The re-sults of the empirical studies are discussed in the paper.