Abstract It is a well-known fact of computational linguistics that syntax is mildly context-sensitive and thus highly complex —- certainly more complex than phonology or morphology. This complexity is at odds with the ease of language acquisition and the impressive speed of human sentence processing. If syntax is so complex, why doesn’t this complexity cause any issues?
In this talk, I argue that syntax is actually remarkably simple from a computational perspective if one picks derivation trees as the central representation/data structure. In this case, syntax becomes subregular. Subregularity is already known to play a central role in phonology and morphology, where it has led to new learning algorithms and typological generalizations. Derivation trees allow us to extend the subregular approach from these domains to syntax. This has intriguing repercussions for
- the cognitive architecture of language, and
- the relation between Merge and Move, and
- the difference between movement and constraints, and
- efficient parsing via sensing tree automata, and
- learning from positive data via limited semantic bootstrapping.
This shows that the subregular approach to syntax has great potential for language science broadly construed: theoretical linguistics, cognitive science, and language technology.
Files [pdf]
@misc{Graf18Penntalk,
author = {Graf, Thomas},
title = {The Surprising Simplicity of Syntax: Derivation Trees, Subregular Complexity, and What It Implies for Language and Cognition},
year = {2018},
note = {Invited talk, November 26, Integrated Language Science and Technology Seminar, University of Pennsylvania, Philadelphia, PA}
}