Whamit!

24.964: Topics in Phonology

Giorgio Magri

Monday, 10-1, 32-D461

Phonological theory has recently extended its empirical coverage from
categorical to quantitative probabilistic data. What is the proper
model of probabilistic phonology? This question is difficult. In fact,
while categorical phonology can be done with paper-and-pencil,
probabilistic phonology cannot. Sophisticated tools are needed to
analyze probabilistic grammars and spell out their phonological
predictions. On the **practical side**, the goal of this class is to
develop some of these tools, and thus to be able to do theoretical
phonology in the probabilistic setting with the same degree of
maturity we have been doing categorical phonology up until now. The
class comes with a companion piece of software (CoGeTo, available at
https://cogeto.stanford.edu/home), that implements these tools and
thus allows class participants to apply them on test cases of
interest.

On the **theoretical side**, I will use these tools to defend a very
specific proposal about the right model of probabilistic phonology. I
will distinguish two classes of models. ‘Intrinsically’ probabilistic
models (such as maximum entropy; ME) postulate phonological grammars
that directly assign probabilities and are therefore very different
from traditional categorical grammars. ‘Extrinsically’ probabilistic
models (such as stochastic harmonic grammar; SHG) instead rely on
traditional categorical grammars and derive the probabilistic behavior
indirectly from the assumption that speakers are uncertain as to which
categorical grammar to use. The class will develop an argument (based
on inference and typological predictions) in favor of SHG and against
intrinsically probabilistic models such as ME. If the argument goes
through, it shows that probabilistic natural language phonology is
intrinsically categorical, after all.

This paper (which is short, if you ignore the
appendix) illustrates some of the tools that will be developed in the
class and sketches the theoretical argument that I will try to make.

Speaker: Soo-Hwan Lee (NYU)
Title: Introducing arguments in and out of the thematic domain: Evidence from Korean case markers
Time: Tuesday, September 13th, 1pm - 2pm

Abstract: Extensive research has focused on how VoiceP (Kratzer 1996), ApplP (Pylkkänen 2008), and i* (Wood & Marantz 2017), an overarching term for Voice and Appl, establish argument structure inside the thematic domain (below TP). A question arises as to whether argument structure can be established outside the thematic domain (above TP). This work provides empirical evidence from Korean in arguing that an argument can be introduced by Voice/Appl (i*) in the left periphery. Specifically, it lends support to the claim that the discourse participant ‘addressee’ is represented in syntax (Hill 2007; Haegeman & Hill 2013; Miyagawa 2017; 2022; Portner et al. 2019 among others). In this regard, this work draws parallels between the thematic domain and the speech act domain, which have been considered to be two separate domains.