Seminar on word meaning in context, Spring 2021
Spring 2021 | LIN 393 | Instructor: Katrin Erk | Tuesday and Thursday, 2-3:30
Word meaning in context is incredibly flexible. In some cases, it is difficult to clearly specify sense boundaries -- which raises the question how we should think of word meanings, how we should represent them. Problems of word meaning in context have been considered in different disciplines, with quite different perspectives: in formal semantics, cognitive semantics, psychology, and computational linguistics. The difficult puzzles of word meaning in context have been around a long time. But recently a multitude of new formalisms have been proposed for representing word meaning. Some of them cross disciplines and combine ideas from formal semantics and cognition. And many of them have been inspired by ideas from probability theory and machine learning -- areas that are particularly good at dealing with phenomena that involve fuzziness and uncertainty.
In this seminar, we look at the puzzles of word meaning in context, as well as formalizations that seek to address them. We will discuss papers from formal semantics and cognitive semantics, psychology, and computational linguistics. The seminar is designed to be accessible to students from linguistic as well as computational backgrounds, and designed to to engender discussions across discipline boundaries. To that end, the seminar includes "background" sessions that provide the necessary high-level background on linguistic, cognitive, or computational material.
The seminar is divided in three main topic blocks: Phenomena, Representation, and Challenges. Topics that will be discussed include:
How can we decide how many senses a word has, and when is this difficult to do?
In what ways can word senses be related, systematically or otherwise?
What is the relation between word meaning and human concept representation?
How are the senses of polysemous words represented in the mind, and what can we learn from this about polysemy in general?
"Two-tier" formalisms that draw on ideas from both formal semantics and cognition
Formalisms that assume a "meaning space" in which word meanings are located, where meaning in context is determined by manipulating points or regions in the space
Formalisms that connect categorization to the idea of classification from machine learning
How does discourse context influence meainng in context?
Can the meaning of a word be arbitrarily variable?
There is no textbook. Readings for the seminar will be made availble through links on the schedule page, and files on Canvas.