- Friday 28 September 2018
- LUCL Colloquium 2018-2019
2311 BD Leiden
The Aristotelian meaning of a sentence is what we – silently, paradoxically – converge upon before diverging in other dimensions of language-in-use. It covers the very backbone of language communication: you understand what she is saying, and therefore, you can agree or disagree or wonder and get angry, confused, impressed or flattered. Yet, both in grammars of natural language and in natural language processing the nature of this semantic level is rarely addressed, if not bluntly avoided. At the same time, the meaningfulness of natural language as such has become a principle target of artificial intelligence. Except for linguists, it seems, everybody is aiming for this - with little success for the time being, though.
In this talk, I will try to substantiate some insights into the mechanics of Aristotelian meaning which emerged during long term efforts to construct a cyclic (language-to-language) semantic parser and generator for Dutch. These insights boil down to the following conjectures:
(a) the meaning of a sentence is the outcome of an operation on that sentence, rather than a rigid attribute of it (avoiding semantic naivety)
(b) in order to compute the meaning of a sentence and to verify that computation, we need an explicit formal, analytic, granular semantic algebra, covering almost classic semantic concepts by Montague, Wierzbicka, Dowty, Van Benthem, Zwarts and others (avoiding grammatical naivety)
(c) the computation of meaning both exploits and disregards fine-graded syntactic analysis (avoiding big data hubris)
(d) syntactic combinatorics does not project semantic space, and propositional structure does not project syntactic patterns; syntax and semantics entertain different levels of entropy; thus, meaningful parsing and meaningful generation are incongruent processes (avoiding algorithmic naivety).
Mainly by referring to meaning-driven generation, the interference of grammar, algorithms, data structures and the empirical cycle is demonstrated by the present state of the Delilah language automaton. The limits and the challenges of algebraic approaches to meaning engineering – in particular issues of scale and sustainability - will be addressed and discussed.