Myhill, John (2001). Suppletion, lexical meaning, semantic primitives, and translation data. Linguistics, 39, 761-802. DOI: 10.1515/ling.2001.031
The Natural Semantic Metalanguage (NSM) theory has taken SAY to be a semantic primitive without specifying exactly which semantic domains it covers. Empirical research shows, however, that the words that might intuitively be taken as the exponents of SAY in English (say), Hebrew (’amar), Japanese (iu), and Spanish (decir) very frequently do not translate as each other, so that we cannot simply say that
SAY=say=’amar=iu=decir. In order to reconcile the translation data with a theory of universal primitive meaning, it is necessary to posit a
number of cases of suppletion and polysemy, the former to account for automatic, non-meaning-based alternations between allolexes reflecting a single semantic primitive, the latter to account for meaning-based alternations between different lexemes. The resulting analysis is relatively straightforward and plausible, though some aspects of it may seem initially surprising, for example that tell has two distinct meanings (tell a story vs. tell me what you want), that tell in the latter meaning alternates allolexically with say, and that ask can serve as an exponent of SAY when the clause it introduces is formally a question but does not pragmatically function as a question.
Note: While some insights in this paper are worth considering, others are much more debatable and result from an incorrect understanding of the basic principles of the NSM approach.
