research
Some of the projects I have been working on are briefly outlined below. The Resources page has links to Praat/shell scripts, Python/Matlab code, references and more.
-
Focus on quantifiers: prominence and contrast
- Preliminary report on contrast types and acoustics of the quantifier "some" in a web-harvested corpus of utterances [manuscript 2014]
- K-means clustering shows strong effect of focus on the acoustic prominence of "some" [abstract ETAP2015]
-
The prosody of Romanian clitics
- Romanian pronominal clitics can form an independent prosodic word [article] [slides] [abstract PWPL 2013]
- Romanian pronominal and auxiliary clitics have different preferences for their hosts. When an appropriate host is not present, they can both form independent prosodic words. [poster] [handout] [abstract MFM 2014]
-
Semantic/pragmatic effects of Romanian Clitic Doubling with or without Differential Object Marking
Like in some of the other Balkan languages, Romanian pronominal clitics may co-occur with the object they refer to. However, the exact conditions for this co-occurrence are still somewhat murky, especially for the direct object. With certain types of NPs in argument position, clitic doubling may be obligatory, while with others it triggers specificity effects. With some NPs it may be impossible, except when syntactic movement intervenes or when the head noun is missing and a demonstrative or quantifier carries the phrase through. Finally, clitic doubling of the direct object is almost always accompanied by differential object marking, a preposition-like word accompanying the associate NP. Differential object marking has also been claimed to trigger specificity effects, but since it often accompanies clitic doubling, it is hard to separate their respective functions. A preliminary exploratory corpus study suggests that looking at associates with demonstratives or quantifiers could allow us to tease differential object marking apart from clitic doubling. Some major challenges are the subtlety of the semantic/pragmatic judgments that come with these constructions and the large number of syntactic and semantic/pragmatic factors that need to be integrated into a full account.
-
Modelling German nested and Dutch crossed dependencies
Embedded infinitival CPs in Dutch have the unusual property of creating (theoretically boundless) crossed dependencies. In common syntactic parlance, a limited set of "restructuring verbs" can take nonfinite CP complements with zero complementizer. When these are embedded, they give rise to discontinuous constituency. Instead of the German-style nested case [NP1 NP2 NP3 V3 V2 V1], Dutch has the "crossed" [NP1 NP2 NP3 V1 V2 V3]. This construction has generated interest in mainstream syntax, but also in mathematical and computational linguistics due to its implications for generative power. Typologically as well, such types of discontinuous constituents are rare, compared to the more common German-style nested ones. Additionally, a psycholinguistic study suggests that Dutch crossed order is in some sense easier to process at higher levels of embedding than the German nested order.
I wrote two grammars for each of the Dutch and German infinitival embedding constructions. One set of grammars is written in the Combinatory Categorial Grammar formalism, using the OpenCCG interface and based on the analysis sketched out in Steedman 2000: ch6 and Baldridge 2002. The other set is in Stablerian Minimalist Grammar, loosely adapting a remnant movement analysis of Dutch and German restructuring verbs described in Koopman and Szabolcsi 2000 for Hungarian. The MGs were developed in Prolog using Stabler's MG CKY parser and utilities. MGs and CCGs are mathematically well-defined syntactic formalisms with mildly context sensitive generative power (CCGs are in fact a proper subset of MGs (Vijay-Shanker and Weir 1994)). Taken together, the MG and CCG analyses allow me to approach this syntactic problem from two different analytic standpoints, while also allowing me to work towards a syntactically-informed psycholinguistic model to try to determine if the claimed processing effects observed in Dutch versus German might have a structural basis.