research sponsored by NKFIH (#139271)
home description publications software
In the suffixation of Hungarian nouns, vowel–nothing alternation can be observed (e.g. dal : dalt, but hal : halat). The occurrence of the so-called connecting vowel appearing at the boundary between the stem and the suffix depends on many factors, including the interacting aspects of the phonological and morphological structure of the given word; in fact, the occurrence of connecting vowels cannot be entirely deduced from such conditions, but is to some extent lexically, i.e. determined on a case-by-case basis. In addition to the fundamental dilemma of occurrence or non-occurrence, there are also cases where both conceivable realizations of the same paradigm cell are accepted. It is not trivial to predict the places where such oscillations occur (in which word stems and which paradigm cells).
According to our research group’s observations, in the orderly listing of the paradigms of the individual stems and suffixes, we obtain a two-dimensional pattern that reveals the stochastic regularity of the distribution of connecting vowels: in such a table, we find two clearly separable areas (one with all connecting vowels, the other with all connecting vowels), but the boundary line between the two areas takes on an unusual stepped shape. We also find fluctuating cells along the boundary line. This suggests that the current arrangement of Hungarian connecting vowels evolved from an earlier simpler, more unambiguous arrangement – namely, in the past, connecting vowels did not appear in a given word form only if both the stem and the suffix were “anti-connecting vowels”. We suspect that the present-day, disordered distribution of connectives may have arisen as a result of the slow erosion of analogical attractive forces between adjacent stems and adjacent affixes.
We attempt to account for this gradual transformation in our simple, intuitive computer model in terms of the assimilation attractive forces between directly adjacent cells in the paradigm system. The simulation confirms the preliminary idea that the logically and geometrically simpler initial arrangement almost always breaks down under the influence of these forces; and that the areas that separate along the stepped border remain distinct. This result, if not proof, is a compelling argument that analogical mechanisms may be of decisive importance for certain phenomena of morphology.
on github, usage guide, description of theory
An implementation of a component of an analogy-based syntactic theory. This component searches for one-word syntactic analogies for two-word expressions that do not occur in the training data, i.e. words that can be grammatically substituted for the given two-word expression in as many contexts as possible. For example, in the case of English training data, let’s assume that the phrase their giraffe did not occur in the training data; in this case, the program should find syntactic analogies for this phrase, e.g. the personal pronouns he, she, it.
An implementation of an analogy-based morphological theory. The implemented algorithm invents word forms that do not occur in the training data based on the surface word forms in the training data, by noticing systematic similarities between the endings of known word forms: it observes what other word form types can be obtained by changing the endings of the initial word forms in a certain way.
home description publications software