Sentence generation for artificial brains: A glocal similarity-matching approach

A novel approach to sentence generation – SegSim, Sentence Generation by Similarity Matching – is outlined, and is argued to possess a number of desirable properties making it plausible as a model of sentence generation in the human brain, and useful as a guide for creating sentence generation components within artificial brains. The crux of the approach is to do as much as possible via similarity matching against a large knowledge base of previously comprehended sentences, rather than via complex algorithmic operations. To get the most out of this sort of matching, a certain amount of relatively simple rule-based processing needs to be done in pre- and post-processing steps. However, complex algorithmic operations are required only for the generation of sentences representing complex or unfamiliar thoughts. This, it is suggested, is the sort of sentence generation approach that makes sense in a system that – like a real or artificial brain – combines the capability for effective local application of logical rules with the capability for massively parallel, scalable, inexpensive similarity matching.