lexemean

Lexemean is a term used in linguistics and computational linguistics to denote a measure that summarizes the typical or central semantic value of a lexeme. The word is a blend of lexeme and mean, reflecting a central tendency of meaning rather than a single sense. The term is informal in many contexts and has no single standardized definition across all fields.

In practice, lexemean can refer to different but related ideas depending on the approach. In distributional semantics, lexemean may be the mean vector of all context-derived embeddings for a given lexeme, serving as a prototype semantic representation. In sense inventories or sense-based embeddings, it can denote the average of embeddings across a lexeme’s senses, yielding a central representation that approximates overall usage.

Calculation methods vary. For vector-based lexemean, one collects multiple embeddings of the lexeme from diverse contexts or senses and computes their arithmetic mean. For sense-based lexemean, each sense is mapped to an embedding, and these are averaged across senses. The resulting vector provides a compact representation that can be used for downstream tasks.

Applications include providing a simple baseline representation for word meaning in NLP tasks, initializing or guiding word sense disambiguation, and aiding clustering of polysemous lexemes. Lexemean can also be used to study semantic shift by tracking changes in the central representation over time or across corpora.

Limitations arise from its simplicity. A lexeme with many distinct senses may have a blurred central vector, and the quality depends on the underlying embeddings or sense inventory. It remains one of several approaches to summarizing lexical meaning and is not universally standardized. Related concepts include centroids of word vectors and prototype theory in semantics.