ABSTRACT
In computational linguistics, generalization by analogy can be defined as the inferential process by which an unfamiliar linguistic object O (called the “target object”) is seen as an analogue of more familiar objects (the “base objects”), so that whatever piece of knowledge is acquired about the latter can be used to deal with the former too (Owen 1990). A fundamental property of generalization by analogy is its flexibility: that is, its capacity to deal successfully with a variety of different problems (in this context, a variety of linguistic phenomena), with not much built-in knowledge being required (e.g. concerning general structural properties of words, sentences, etc.). The problem is that, even in a relatively well-understood linguistic domain such as morphology, it is not as yet clear how powerful generalization by analogy should be for the huge variety of word formation processes attested across different languages to be conveniently and adequately captured.
