Timothy J. O'Donnell: Computation, Storage, and Generalization in Language
Timothy J. O'Donnell
- McGill University
April 28, 2017, 2:30 p.m. - April 28, 2017, 3:30 p.m.
A much-celebrated aspect of language is the way in which it allows us
to express and comprehend an unbounded number of thoughts. This
property is made possible because language consists of several
combinatorial systems which can be used to creatively build novel
words and sentences using inventories of stored, reusable units.
For any given language, however, there are many more potentially
storable units of structure than are actually used in practice ---
each giving rise to many ways of forming novel expressions. For
example, English contains suffixes which are highly productive and
generalizable (e.g., -ness; Lady-Gagaesqueness, pine-scentedness) and
suffixes which can only be reused in specific words, and cannot be
generalized (e.g., -th; truth, width, warmth). How are such
differences in generalizability and reusability represented? What are
the basic, stored building blocks at each level of linguistic
structure? When is generalization possible when is it not?
How can the child acquire these systems of knowledge?
I will discuss how tools from machine learning, artificial
intelligence, and computational linguistics can address these
problems. The general approach is based on the idea that the problem
of computation and storage can be solved by using a probabilisitc
tradeoff between a pressure to store fewer, more reusable units and a
pressure to account for each linguistic expression with as little
computation as possible. This tradeoff is grounded in foundational
principles of inductive inference, but has surprisingly far reaching
implications across multiple levels of linguistic structure. I will
discuss several specific models based on this framework and provide
examples of how the approach can help solve long standing empirical
puzzles, simplify existing accounts, and connect theories from
linguistics, psychology, and computer science.