A series of talk by Shravan Vasishth

Tue Mar 3 19:50:54 UTC 2009

de la part d'Asaf Bachrach :


> Dear all,
> Shravan Vasishth (Chair of Psycholinguistics & Neurolinguistics,
> Potsdam university) will be visiting the unicog (Neurospin) lab in  
> the coming
> week. He  will be giving a  series of seminars (
> meeting next Tuesday).
> The  first (dealing with prediction and working memory) will be  
> held on  Friday
> morning in la  Maison de la Recherche. The second (on Aphasia)   
> will be held in
> the Pitié-Salpetriere at 15:00 this Friday. The third one  will  
> take  place
> during the regular LSCP lab meeting at the DEC.  See below  for  
> details,
> Asaf
>
>
>
> Shravan Vasishth (Chair of Psycholinguistics & Neurolinguistics,
> Potsdam university)
>
>
> Title: Prediction and retrieval in dependency resolution: Models  
> and  data
> Friday 6/3/2009 11:00 ? 12:30 am
> Location: Salle D040, Maison de la Recherche, 28, rue Serpente, 75006
>
> Title:  A case against chance performance: Evidence from eye movements
> of agrammatic aphasics
> Friday 6/3/2009 15:00-16.30
> Location: Pitié-Salpetriere, Pavillon de l'Enfant et de l'Adolescent,
> (dans le batiment, aller à droite, prendre l'ascenseur "monte malades"
> jusqu'au troisième. étage La salle de conférence est au fond à gauche)
>
> Title: The Integration Advantage due to Clefting and Topicalization
> Tuesday 10/3/2009   14:00-16:00
> Location: Laboratoire de Sciences Cognitives et Psycholinguistique
> Ecole Normale Supérieure, 29 rue d'Ulm
>
>
> Dans les trois cas; le labo organisateur est l'unité de neuroimagerie
> cognitive INSERM-CEA.
>
> Merci!
>
> ABSTRACTS:
>
>
> Prediction and retrieval in dependency resolution: Models and data
> Resolving a head-dependent relationship a central act in online
> sentence comprehension; without it, comprehension is impossible.
> Fillers must be connected to gaps, antecedents to pronouns and
> reflexives, licensors to polarity items, incoming phrases with one of
> multiple possible attachment sites; and so on. Although such
> dependency resolution processes often proceed smoothly, at other times
> difficulty occurs: targeted items are retrieved slowly or not
> retrieved at all, or the wrong item is retrieved. What causes such
> breakdowns? Understanding the nature of online dependency resolution
> remains a central open problem in sentence processing research.
>
> Prediction and retrieval based difficulty have been invoked to explain
> dependency resolution difficulty. Retrieval theory has argued
> primarily for decay and/or interference as explanatory primitives
> (Gibson 2000, Lewis and Vasishth 2005, Van Dyke and McElree 2007,
> among others), while predictive or expectation-based accounts rely on
> quantifying the uncertainty about the upcoming word in a sentence
> (some recent examples are: Hale 2001, Levy 2008, Boston, Hale, Patil,
> Kliegl, Vasishth 2008, Demberg and Keller 2008).
>
> Retrieval theory and expectation-based explanations are,
> theoretically, orthogonal aspects of the incremental parsing process.
> This raises the possibility that both classes of explanation should
> operate more or less simultaneously to determine parse difficulty. In
> order to test this hypothesis, I present a scalable computational
> model of online parsing that delivers retrieval and prediction cost in
> a unified framework for arbitrary sentences (Boston et al 2009; cf.
> Patil et al 2009). I show how retrieval and prediction based accounts
> can be combined to explain a range of benchmark data gathered in my
> laboratory using methodologies such as self-paced reading, eyetracking
> and event related potentials.
>
> A case against chance performance: Evidence from eye movements of
> agrammatic aphasics
> (joint work with Sandra Hanne, Irina Sekerina, Frank Burchert, Ria De
> Bleser)
> Broca aphasics often perform at chance level in the comprehension of
> reversible noncanonical sentences (Grodzinsky 1995, Burchert et al.
> 2003). The Trace-Deletion-Hypothesis (Grodzinsky 1995) argues that
> patients erroneously assign thematic roles based on an agent-first
> heuristic. In an eye-tracking study, Dickey et al. (2007) observed a
> mismatch between aphasics? online sentence processing and their
> offline responses: While patients exhibit normal-like online
> processing, they often do not succeed in offline comprehension: chance
> performance in offline tasks does not necessarily reflect chance level
> performance during incremental online processing. Dickey and
> colleagues argue that the normal online performance of aphasics can be
> explained in terms of the slowed processing hypothesis: aphasics build
> structure more slowly than normals, and are therefore sometimes unable
> to converge on the correct syntactic structure in time to parse the
> sentence correctly.
> We provide further evidence for the slowed processing account from
> German. A sentence-picture-matching experiment was conducted with 8
> controls and 7 German Broca aphasics. We used German canonical SVO
> sentences (1) and noncanonical OVS sentences (2).
> (1) Der Sohn fängt den Vater.
> (2) Den Sohn fängt der Vater.
> Two pictures were presented side-by-side simultaneously with the
> spoken sentence: the target (agent and patient acting according to the
> sentence) and the foil picture (semantically reversed action).
> Accuracy, reaction times and eye-movements were recorded.
> For accuracy, controls were at ceiling for both conditions (SVO: 98%,
> OVS: 95%) while patients were impaired for SVO (80%) and at chance for
> OVS (46%). Patients were twice as slow as controls but SVO was
> processed faster than OVS in both groups. Controls? eye movements
> reflected a preference for the target picture from the der/den-NP
> region onwards in both conditions. Patients? fixation patterns in the
> canonical SVO condition were very similar. When we analyzed all trials
> (correct and wrong ones) for the noncanonical OVS condition, we
> observed a persistent preference for the foil picture. This might
> suggest the application of an agent-first heuristic. Following Dickey
> et al. (2007), we looked at correct and incorrect trials separately
> and found that, in correct trials, patients (just like controls)
> showed an early preference for the target picture. This is new
> evidence for a dissociation between patients? online processing and
> their offline performance, similar to claims regarding children
> (Sekerina et al. 2004).
> In addition to the above findings, I demonstrate how individual
> differences among aphasics' performance (a crucial issue that is
> avoided in aphasia research) can be modeled statistically using linear
> mixed effects models.
> The Integration Advantage due to Clefting and Topicalization
> (joint work with Rukshin Shaher, Felix Engelmann, Pavel Logacev,
> Narayanan Srinivasan)
> What is the functional motivation for the existence of elaborate
> syntactic markers such as clefts, left-dislocated topicalizations, and
> given-new ordering? Although it is clear that they are syntactic
> markers that facilitate effective information-packaging that
> communicates a message to the hearer/reader, it is less clear how this
> kind of restructuring impacts processing in real-time sentence
> comprehension. Two eyetracking studies involving Hindi address this
> question. We show that such information structure markers drive the
> re-allocation of attention for facilitating comprehension; this
> re-allocation has the consequence that the message is processed faster
> and more efficiently.
> Previous work on information structure marking has shown that readers
> detect focused information more quickly and accurately, and remember
> it better than non-focused information. For example, in a probe
> recognition and naming task, Birch and Garnsey (1995) showed that
> clefted nouns ('It was a?') can be named faster than non-clefted ones.
> In an eyetracking experiment, Foraker & McElree (2007) showed that
> clefting a noun improves its availability in online sentence
> comprehension. Birch and Rayner (1997) provided evidence that
> processing a clefted noun is computationally costly. They claim that
> the costly processing operations on the clefted noun reflect a more
> robust encoding in memory which explains the facilitation during
> retrieval found by Birch and Garnsey.
> Our eyetracking experiments extend on this previous work by showing:
> (i) there is an initial processing cost (encoding cost) associated
> with encoding a focused element; (ii) but this results in richer
> encoding, which facilitates later processing (integration advantage);
> (iii) the integration advantage interacts with the widely accepted
> given-before-new ordering preference.
> An eyetracking study (n=32) involving Hindi clefted sentences was
> carried out; the factors clefting and given-new order were
> manipulated. Subjects saw two sentences, a context sentence followed
> by a target sentence, which collectively described the relative
> position of three objects. They were then presented with a picture and
> had to indicate whether the layout in the picture matched the
> description.
> Context sentence:
> banduuk duurbiin kii baayii taraf hai.
> ?The gun is to the left of the binoculars.?
> Target sentence: (cleft marker in italics, given material in bold)
> ek  tijorii  bhii  hai,  aur/lekin  [NP1 jhanDaa]  (hai jo)
> One  safe  also  is,  and/but  [NP1 flag]   (is that)
> [NP2 duurbiin]  [INT kii  daayii   taraf]   hai.
> [NP2 binoculars]  [INT gen  right   side]   is
> ?There is a safe, and/but (it is) the flag (that ) is to the right of
> the binoculars.?
> We found higher first pass regression probabilities for clefted nouns
> as opposed to non-clefted nouns. This can be interpreted as the
> encoding cost for clefted nouns. We also found shorter re-reading
> times at the clefted noun and fewer regressions to it from the
> integration site. We interpret this as evidence for the integration
> advantage.
> In addition, although whole sentence total reading time showed a
> given-new advantage (confirming the accepted opinion that given-new
> order is easier to process), the clefted word itself was read faster
> when it was new rather than given.
> We replicated the above findings through a second eyetracking
> experiment (n=32) involving Hindi left-dislocated topics that
> superficially resemble clefts but have a different
> information-structuring function.
> In sum, we present new evidence from online sentence comprehension
> that syntactic information-structure markers such as clefts and
> left-dislocated topics serve to facilitate retrieval of the
> clefted/topicalized element and that the initial cost of encoding can
> be minimized by providing context information.
>


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/parislinguists/attachments/20090303/9886eea8/attachment.htm>