Polysemy

Polysemy indicates that a semiotic sign can have multiple meanings.

Polysemy poses a particular challenge for machine learning models used in NLP. Many of those models rely on statistical measures, such as word frequency, that match the form of a linguistic unit but not its semantics. The determination of the meaning of a polysemic word instead requires an understanding of the meaning of the surrounding context. This is very hard to do for machines, but there is some progress in recent transformer models that create more context-aware embeddings, such as BERT.

Polysemy differs from what Derrida called dissemination in the sense that polysemy considers multiple discrete meanings of some linguistic unit. Derrida's concept of dissemination instead challenges the idea that the art of interpretation can, at the limit, be exhausted by a combinatorics of all the possible meanings that all the possible signifiers can take on. The idea that meanings can be "summed up" into a whole assumes that the whole project of interpretation moves within an hermeneutic horizon that, ultimately, always presupposes meaning (but also fundamentally remains underway; never completes the sum). (E.g. see "The Sharing of Voices" by Jean-Luc Nancy). Derrida's work challenges this idea from many angles, but e.g. in Limited Inc. he speaks of "the necessity of disengaging from the concept of polysemics what I have elsewhere called dissemination" (Limited Inc. 8-9)).