Whamit!

This week continues a tradition that began in 2015: our tenth joint colloquium shared by the Linguistics and Philosophy halves of our department:

Speaker: Dorit Abusch & Mats Rooth (Cornell University)
Title: Possible worlds semantics for film and the problem of over-informative embedding
Time: Friday, December 6th, 3.30-5pm
Location: 32-141

Abstract: In the superlinguistics program, the possible worlds toolkit from the philosophy of language and linguistic semantics is applied to other informational artifacts. This talk develops this for film. While film has hardly any obvious syntax apart from the concatenation of shots, methodologically it is useful to posit an abstract syntax that is interpreted compositionally. We begin by reviewing prior work on geometric semantics, temporal progression, and indexing in pictorial narratives, and then develop them for film. The result is a possible worlds semantics for extensional passages in film. Intensional passages in film include shots describing the hallucinations, dreams, and recollections of a character. Thematically, they include stories about drug-induced hallucination, characters who see the dead, and schizophrenic characters who interact with imaginary ones. In the abstract syntax, a straightforward embedding strategy is adopted, involving an attitude predicate, and a discourse referent for an experiencer. For the semantics, we argue that there is a systematic problem of the embedded shots having detailed geometric, temporal, and (for sound film) acoustic information. This information is so strong that it is implausible that in described situations, the experiencing character should have strong enough information to entail the information in the embedded shot. This makes Hintikka semantics for embedding in film problematic. A starting point for a solution is the hypothesis that attitudinal alternatives for a hallucinating or mis-remembering character should be comparable to the attitudinal alternatives of a character who is perceiving their environment veridically, or remembering accurately. Since people do not pick up all the information in their visual and acoustic environments, the attitudinal state of a hallucinating character should not be required to entail all of the geometric, temporal, and acoustic information in an embedded film shot.

Speaker: Stanislao Zompì (Universität Potsdam) and Zhouyi Sun (MIT)
Title: Augmenting Vocabulary Insertion: From Monotonic to Output-driven
Time: Thursday, December 5, 5-6pm
Location: 32-D769

Abstract: It has been argued that *ABA in root suppletion patterns can be reduced to the principle of monotonicity: given a partial order among feature bundles, if two feature bundles x and z are mapped to the same value, then any y between x and z in terms of the given order must also be mapped to the same value (Graf 2019; Moradi 2021). We show that Underspecification/Overspecification combined with Pāṇinian ordering as rules for Vocabulary Insertion are both inherently monotonic. In other words, previous works deriving *ABA using these approaches all implicitly share the monotonicity thesis.

However, given the independently motivated featural successive containment relations in the domains of Case (Nom ⊂ Acc ⊂ Dat) and Gender (Neu ⊂ Msc ⊂ Fem) in Germanic languages, we find monotonicity falls short in capturing attested paradigms, and therefore both Underspecification and Overspecification undergenerate. Monotonicity in the form as explicitly posited by Graf (2019) or implicitly assumed by many others may thus not be a desirable property for language. Our survey further suggests that the empirical picture instead aligns with an concept that’s properly weaker that monotonicity—Output-driven (Tesar 2014; Magri 2018a,b), which requires monotonicity not between any two feature bundles with the same output, but only between any feature bundle and the output it maps to.

We propose an optimization-based approach to Vocabulary Insertion which allows for and penalizes feature addition and deletion, with penalties tied to specific features. Realization of a feature bundle is determined by minimizing the penalty incurred to reach a feature bundle associated with a dedicated Vocabulary Item. We argue it generates—and only generates—output-driven mappings, fitting the empirical picture.

Speaker: Keely New (MIT)
Title: On the apparent lack of multiple wh-questions in Jakarta Indonesian
Time: Tuesday, December 3, 1pm - 2pm
Location: 32-D461

Abstract: It has been claimed that in certain languages (e.g. Italian, Irish, and Somali), it is impossible to form questions that include more than one wh-constituent. This is surprising because under prominent semantic analyses of questions, single and multiple wh-questions are derived through similar mechanisms (as noted by Dayal, 2016 and Kotek, 2019). I argue that the lack of multiple wh-questions should not be attributed to a language parameter. Focusing on Jakarta Indonesian as a case study, I will show that (i) the impossibility of multiple wh-questions is construction-specific rather than a language-wide pattern, and (ii) we can derive the impossibility of multiple wh-questions from independently observable constraints on question formation.

Speaker: Eyal Marco (Joint work with Radan Nasrallah and Ezer Rasin), Tel Aviv University.
Title: Phonological derivations are not harmonically improving: Evidence from Nazarene Arabic
Time: Monday, December 2, 5-6:30pm
Location: 32-D831

Abstract: Opacity poses a well-known challenge for the classical version of Optimality Theory (OT; Prince & Smolensky 1993/2004), which applies phonological processes in a fully parallel fashion. In response to the opacity challenge, a variety of extensions to the classical model have been proposed. Our focus is on extensions that incorporate a limited kind of serialism into OT: Harmonic Serialism (McCarthy 2000, 2016) and Optimality Theory with Candidate Chains (OT-CC; McCarthy 2007). The degree of serialism permitted by these theories is restricted by a property that McCarthy has called ‘Harmonic Improvement’, according to which every derivational step induces a change that creates an output that is ‘more harmonic’ than the input of that step, i.e., every derivational step must generate a better form with respect to the constraint ranking of the language. One consequence of Harmonic Improvement is that it rules out derivations of the form /A/ → |B| → [A] (also known as Duke-of-York, see Pullum 1976, McCarthy 2003, Gleim 2019), because /A/ → |B| implies that B is more harmonic than A while |B| → [A] contradictorily implies the opposite.

We present new data regarding the distribution of stress and vowel length in Nazarene Arabic (NZA), an understudied variety of Palestinian Arabic spoken in Nazareth, Israel. We argue that this distribution is opaque and is best analyzed through a derivation of the form /A/ → |B| → [A], where the application of a vowel lengthening process is undone by a later vowel shortening process. The distribution thus poses a challenge to serial theories of phonology that obey `Harmonic Improvement’. We show this by arguing that Parallel OT, Harmonic Serialism and OT-CC have trouble generating the NZA pattern, but not rule-based phonology (Chomsky & Halle 1968) and Stratal OT (Bermúdez-Otero 1999, Kiparsky 2000), which have the freedom to undo processes in the course of the derivation.

Speaker: Omri Doron (MIT)
Title: A typological argument against lexical cumulativity
Time: Wednesday, December 4, 1pm - 2pm
Location: 32-D461

Abstract: In this talk, I develop and motivate a new implementation of an old idea about the contrast between acceptable Hurford disjunctions (HDs) like (i) and unacceptable ones like (ii) (e.g. Gazdar 1979) — the idea that (i) is good because its simpler alternative (iii) would trigger an optional ’not all’ inference that is avoided by using (i), whereas (ii) has no analogous advantage over its weaker disjunct (e.g. Gazdar 1979). A straightforward implementation of this “Manner approach” in terms of ambiguity avoidance runs into a series of problems pointed out by Meyer (2013, 2014). Meyer therefore develops an alternative account of the contrast in (i)/(ii), based on the idea that exhaustification creates a contradiction when faced with alternatives that are contextually, but not logically entailed by its prejacent.

(i) Ann did some or all of the problem sets.
(ii) #Ann went to France or to Paris.
(iii) Ann did some of the problem sets.

In this talk, I will first argue against Meyer’s logicality-based alternative to the Manner approach and then introduce an implementation of the Manner approach that avoids some (although not all) of the problems she discusses. It relies not on ambiguity avoidance, but on a trade-off between the Gricean submaxim *Be brief!* and a submaxim I call *Be precise!*, which encodes a dispreference for utterances that are ‚imprecise‘, i.e. that depend on the QUD for their truth conditions in the sense discussed by Križ & Spector (2021). To motivate *Be precise!*, I will take a short detour to the pragmatics of plurals and *all*. Returning to HDs, I will argue that it is possible to view (iii) as imprecise and (i) as a strategy to avoid imprecision, if we adopt the view that exhaustification is syntactically obligatory, but alternative pruning is restricted by the QUD in the way proposed by Bar-Lev (2020). Time permitting, I will conclude by discussing an open problem for this approach, which has to do with the status of ignorance inferences in HDs.

Speaker: Patrick Juola (Duquesne University)
Title: Forensic Linguistics: Casework and Empirical Demonstrations
Time: Thursday, December 5th, 12:30pm - 2pm
Location: 32-D461

Abstract: Forensic linguistics is a relatively undersubscribed subfield of forensic science, which focuses on developing legal conclusions from linguistic evidence. In this talk, we discuss the theory and practice of “authorship attribution” (aka “styometry”) and some of its applications (such as identifying the author of a ransom note). Examples are drawn from actual analyses drawn both from the literature and from the presenter’s own casework.