This seminar is advanced. Available to tudents with soem background in cognitive science and/or computational linguistics.
The objective of various cognitive mechanisms is to interpret the world in one and one way only (give it only one meaning) when reality presents us with options for interpretations: input ambiguities. Many important natural language inferences can be viewed as problems of resolving ambiguitiy (either semantic or syntactic, or pragmatic) based on properties of surrounding context. Lexical ambiguity resolution is an omnipresent problem in natural language processing. Thus, the objective of building an adequate A.I. system is to be able to (a) predict the intended meaning of the input; (b) create an algorithm that would be "able" to point to the ONLY "desired" output".
The main question we'll tackle (and tickle) is: how and where do humans defeat the machines when it comes to language (perceptual) processing?
During our meetings, we will look at how different cognitive mechanisms working sequentially, simultaneously, and separately resolve ambiguities; and how these mechanisms are translated into the language of "algorithms". We will rely on both theoretical findings and hands-on contemporary work (including, Google, Siri, Amazon Alexa, etc*.)
Topics include (but are not limited to):
(1) cognitive mechanisms and ambiguity resolution
(2) "mind" of the machine
(3) statistical learning vs. intuition (data vs. guess)