Computers can help us access our lexicons
The project 'Dynamic networks for lexical access: Design, navigation and interface' (DYNNETLAC) was set up to tackle the tip-of-the-tongue phenomenon, a kind of anomia. The research took as its starting point Newman's complex network theory (which can help in both organising and navigating the lexicon) and semantics of natural languages. The overall goal was to create dynamic networks that can help users overcome momentary blocks to accessing a desired word. To this end, the team set and achieved several sub-goals. The first entailed building lexical networks capable of simulating humans when associating words. Using Wikipedia abstracts and the British National Corpus as corpora, graphs were designed containing only verbs, nouns and adjectives, and considering only immediate 'neighbours'. The team built on research initiated by WordNet to design lexical networks that approximate the mental lexicon. DYNNETLAC also evaluated the potential for modelling navigation in complex networks using methods from DNA regulatory networks. Investigations considered how methodological interactions between biology and mathematical systems can be tailored to a linguistic model. A lesser developed objective was the design of a man-machine interaction system for searching the graphs for clues. Work did not progress in this area primarily for reasons of poor technical compatibility when trying to integrate the chosen programming language (Python) in the dynamic web page design. This is a known issue that other (non-project) researchers are addressing. DYNNETLAC tested its system by comparing its best-ranked words with primary responses retrieved by humans in the Edinburgh Association Thesaurus (EAT). Although the results were positive, the team acknowledges it would be better if the system matched their first word with the one given by the EAT. Project activities also involved a training component relevant to the research topic, new research methodologies, and developing skills relevant to computational implementation of formal and mathematical linguistic theories. Other efforts included initiating interactions and collaboration with the Natural Language Processing community. Overall, project results open a new line of research: the use of automatically generated non-annotated resources to simulate human cognitive capabilities. Although more research is needed, DYNNETLAC demonstrated that very simple techniques can be used to describe major processes of natural language generation.
