Researchers describe a computational model of bilingual naming built from laterally-connected self-organizing maps (SOMs). Their abstract argues that traditional machine learning needs more training data than is usually available when modeling a single individual's cognition. Alternative architectures such as laterally-connected SOMs may offer a more practical path to explanatory, person-specific models.
The source is a single arXiv abstract, so it does not report experimental results or compare the model against other approaches. The stated motivation is to give computational models explanatory power for cognition in individuals, and the specific focus is on bilingual naming performance. Details of how lateral connections contribute to naming are not included in the abstract.