Whamit!

Speaker: Adam Albright (MIT) & Donca Steriade (MIT)
Title: Discussion: Rasin and Katzir (2016, 2020)
Time: Monday, November 15th, 2pm - 3:30pm

Abstract: (This is a continuation of the discussion that took place on 10/25)

We will discuss two recent papers by Ezer Rasin and Roni Katzir. Rasin and Katzir (2016) On Evaluation Metrics in Optimality Theory (LI 47) describes an application of the principle of Minimum Description Length (MDL) as an evaluation metric for OT. MDL is a framework that rewards analyses that can encode the analysis and data as compactly as possible. In the original formulation of OT (Prince and Smolensky 1993/2004), grammars are total rankings of a universally fixed set of constraints (CON), so do not differ in size in any interesting way; Rasin and Katzir propose to allow language-particular constraint sets, and show that MDL learning can favor grammars and lexicons that restrictively characterize the data. Rasin and Katzir (2020) A conditional learnability argument for constraints on underlying representations (J. Linguistics 56) examines more closely the way in which the MDL approach achieves a restrictive grammar. It argues that the evaluation metric favors analyses that include constraints on underlying representations, contrary to the usual OT assumption of Richness of the Base.

You are invited to participate in our discussion this week, Tuesday, November 9, 2-5pm EST, on “Linguistics and Social Justice: Language, Education & Human Rights”  (MIT Linguistics, Graduate Seminar, 24.S96).  Please contact Michel <degraff@mit.edu> for information about Zoom link and readings.  NB: We are committed to creating an inclusive and accessible environment in our seminar. If you need assistance for accommodations or accessibility in order to fully participate, please email degraff@MIT.EDU so that we can work out adequate arrangements.

This Tuesday, November 16, 2021, Kevin Scannell will help us understand the potential of digital technology for deminoritizing, revitalizing and normalizing endangered languages — and other kinds of minoritized languages:

Language from Below:
Grassroots efforts to develop language technology for minoritized languages

Kevin Scannell
November 16, 2021, 2-5pm EST

Seminar: “Linguistics & social justice” (24.S96 @ MIT Linguistics)

Technology plays a key role in revitalization efforts in many language communities. Without the ability to use one’s native language on computers, mobile devices, social media sites, etc., speakers are forced to shift to a dominant language in contexts where computing plays an important role, most notably in schools and in the workplace.

Kevin Scannell is professor of mathematics and computer science at Saint Louis University.  He works with language communities around the world to develop computing resources that help them use their native language online, with a particular focus on Irish and the other Celtic languages. 

On Tuesday, November 16, he will share some success stories in developing language technology — both in his own work on Irish in Ireland, and in the work of his friends and collaborators on the Māori language in Aotearoa / New Zealand.

Successes in Ireland have been achieved in the face of some huge obstacles: little or no funding for these initiatives, general disinterest from the big tech companies, a very fast-moving software landscape, and technical challenges that arise when applying machine learning approaches to languages that lack sufficiently-large datasets for training.

For Irish, some of these obstacles have been overcome through a community-based, grassroots approach to tech development, and occasional (sometimes uneasy) collaborations with big tech. 

Kevin and colleagues’ work in Ireland can be viewed as part of a long history of language activism “from below.”

Title: Anchor displacement readings of X-marked epistemic modals.

Abstract: X-marking is well-known from the literature as counterfactual morphology or fake past. However, von Fintel and Iatridou (2007, 2020) have shown that this morphological marking is often used beyond counterfactual or subjunctive conditionals. They have shown that X-marking has two different functions: (i) domain widening in the case of X-marked desire predicates and (ii) ordering source addition in the case of X-marked necessity modals. In this talk I go beyond the scope of von Fintel and Iatridou (2007, 2020) by investigating X-marked epistemic modals. I show that similarly to X-marked necessity modals X-marked epistemic modals have two possible readings: (i) a reading that is about possibilities in the actual world (wild guess reading) and (ii) a reading that is about possibilities in a world that is not the actual world (counterfactual reading). Furthermore, I present evidence from Hungarian that X-marked epistemic modals can have a third reading in some contexts. These special contexts are what we find in the so-called judge examples (cf.\ Egan et al. (2005), Stephenson (2007)) like in (1).

(1) Context: Chris and Burt are sitting by the window in a cafe.  When a bus goes by they see their friend Ann, who is angry with Burt, jumping behind a bush. Chris asks Burt why Ann did that. Burt responds:
I might be on that bus.

The Hungarian equivalent of these examples can contain either O-marking or X-marking. I claim that X-marking in these examples signals that the anchor of the modal is someone other than the speaker, while O-marking is neutral with respect to who the judge is. I hypothesize that there is a competition between the O-marked and X-marked forms of epistemic modals that leads to a strong preference for the X-marked forms when it is clear that the modal cannot be interpreted from the speaker’s perspective.

Hopefully by the end of the talk I have shown that by looking at the judge examples in morphologically rich languages we not only find evidence for a third reading of X-marked epistemic modals, but also for a third function of X-marking: anchor displacement.