The Weekly Newsletter of MIT Linguistics

Ling-Lunch 10/31 - Pauline Jacobson

Speaker: Pauline Jacobson (Brown)
Title: The Myth of Silent Linguistic Material
Date/Time: Thursday, Oct 31, 12:30-1:45p
Location: 32-D461

The literature abounds with arguments for the claim that there is actual silent (or deleted) linguistic material in a variety of “ellipsis” constructions – call this the Silent Linguistic Material (SLM) hypothesis. While obviously not all such arguments can be deconstructed over lunch, this talk aims to show that the reasoning behind many of them is fallacious, and that there is no reason to think there is linguistic material which is silenced or deleted under identity.

First we begin with the broader question at issue. The reason for doubting the SLM hypothesis is not driven by a stubborn allergy to silent material; rather we will put this hypothesis in the context of Direct Compositionality. Direct compositionality (see, e.g., Montague’s English as a Formal Language) maintains that the syntax and semantics work in tandem locally building expressions and assigning them a meaning. While mapping an expression into a silent version of that expression can be done quite locally, what is difficult to reconcile with this architecture is the idea of material being silenced under some sort of identity with something else in the discourse context, as this kind of identity condition is not a local property of an expression. I will briefly mention alternative accounts of both fragment answers and VP Ellipsis that don’t make use of SLM, although time precludes details arguments for the alternatives. Here then I can only level the playing field and show that SLM has no real advantage.

The arguments for SLM to be considered (and deconstructed) here fall into two classes. The first is based on the idea that the “remnant” acts as if it were surrounded by additional material with respect to certain grammatical processes/generalizations. But I will show that this itself relies on non-direct compositional and non-local account of the relevant generalizations, and that for an interesting class of such cases there are indeed alternative accounts “on the market” which undermine the rationale for SLM. Moreover, facts about indexicals known about since at least as early as Hankamer and Sag (1984) make it clear that the requisite “identity” condition cannot be formal. But if that is the case, some of the arguments for SLM also collapse as they crucially assume formal identity. The second type of argument is often implicit but seems to underlie much of the reason that SLM seems at first glance like a commonsense view: this is that the “meaning” of constructions with ellipsis becomes trivial to account for if there is SLM. But we don’t know the actual meaning – only the likely understanding in a discourse context, so this view only makes sense if put in terms of processing. But I will argue that positing SLM makes the job of the processor no easier than not positing SLM. In fact, work on processing often (or at least occasionally) makes the mistake of assuming that there is SLM, that the processor has access to the quiet material, and that processing proceeds from there. In other words, some claims about the processing of ellipsis make sense only if the processor already knows what it meaning it is “trying” to compute. As a case study I will consider an argument from Hackl, Koster-Hale and Varvoutis (2010) concerning the interaction of ACD, processing, QR, and de re vs. de dicto readings. My discussion of this point is based on joint work with Ted Gibson, Ev Fedorenko, Steven Piantadosi and Peter Graff.