Perhaps the most fundamental contribution of human language is the capacity to convey meaning. Humans effortlessly map language input to rich, nuanced representations of the information conveyed. In this course we will examine approaches to meaning in computational linguistics -- particularly, efforts to capture meaning and "understanding" in artificial intelligence. Recreating a capacity for linguistic meaning in artificial intelligence is fraught with challenges: not only the engineering challenges of simulating the robustness and nuance of human language comprehension, but also the challenge of defining what it means to comprehend language, and of assessing the extraction and representation of meaning in complex black box systems used in modern artificial intelligence. In exploring these problems, we stand not only to improve the performance of artificial intelligence systems, but also to gain insight into how semantic processing and language comprehension operate in the human mind. In this seminar we will discuss a range of literature tackling the problem of extracting, processing, and representing meaning in machines, and we will explore in detail how these approaches relate to the functioning of linguistic meaning in humans. We will also explore current standards by which "language understanding" is evaluated in artificial intelligence, and critique these methods from the perspective of human language comprehension and semantic processing.
Allyson Ettinger -- Spring 2021