Isabelle Charnavel of Harvard University will present a colloquium in the Department of Linguistics in ILC 400, on Friday Dec. 2 at 3:30. All are welcome!

Title:  Logophoricity in Adjunct Clauses

Abstract:

It has long been observed that perspective sensitive elements such as logophoric pronouns or long distance reflexives can occur in some circumstantial subordinate clauses. The goal of the talk is to explore the consequences of this observation for the syntax and semantics of adjunct clauses. Based on the case study of causal clauses in English (and French), I argue that such adjuncts are intrinsically

perspectival: an adjunct conjunction (e.g. because/since) expresses a relation (e.g. cause) between an argument A (roughly, the main clause) and its complement B, which is established by a judge. The identity of the judge, which can be diagnosed by the use of elements oriented towards the speaker or the event participant in the adjunct clause, depends on the nature of A (namely event/state, proposition or speech act).

(1)  Liz has a fever because she has malaria. (cause of state according to Liz or speaker)

(2)  Liz has malaria since she has a fever. (reason for truth of proposition or cause of assertion according to speaker)

Moreover, the scope and judge possibilities of causal clauses increase in the presence of attitude clauses: an attitude holder can be the causal judge for an embedded state/event or an embedded proposition.

Ultimately, the correlations between the scopal and the perspectival possibilities of causal clauses show that the perspective parameter (the

judge) can be analyzed as a silent argument of the conjunction that must be bound within its clause.

who: Pedram Rooshenas, University of Oregon
when: 11:30am, Wednesday, Nov 30
where: Computer Science Building rm150
food: wraps from The Works

Learning Tractable Graphical Models

Abstract:
Probabilistic graphical models have been successfully applied to a wide variety of fields such as computational biology, computer vision, natural language processing, robotics, and many more. However, in probabilistic models for many real-world domains, exact inference is intractable, and approximate inference may be inaccurate.  In this talk, we discuss how we can learn tractable models such as arithmetic circuits (ACs) and sum-product networks (SPNs), in which marginal and conditional queries can be answered efficiently.

We also discuss how we can learn these tractable graphical models in a discriminative setting, in particular through introducing Generalized ACs, which combines ACs and neural networks.

Bio:
Pedram Rooshenas is a Ph.D. candidate at the University of Oregon working with Prof. Daniel Lowd. Pedram’s research interests include learning and inference in graphical models and deep structured models.

Pedram has an MSc. degree in Information Technology, with a thesis on data reduction in wireless sensor networks, from Sharif University, Tehran and an MSc. degree in Computer Science from the University of Oregon.

Pedram also maintains Libra, an open-source toolkit for learning and inference with discrete probabilistic models.

Amanda Rysling (UMass Linguistics) will be presenting in this week’s cognitive brown bag – all are welcome!

Title: Preferential early attribution in incremental segmental parsing

Time: 12:00pm to 1:15pm Wednesday Nov. 30. Location:  Tobin 521B.

Abstract: Recognizing the speech we hear as the sounds of the languages we speak requires solving a parsing problem: mapping from the acoustic input we receive to the phonemes and words we recognize as our language. The literature on segmental perception has focused on cases in which we as listeners seem to successfully un-do the blending of speech sounds that naturally occurs in production. The field has identified many cases in which listeners seem to completely attribute the acoustic products of articulation to the sounds whose articulation created them, and so seem to solve the parsing problem in an efficient and seldom-errorful way. Only a handful of studies have examined cases in which listeners seem to systematically “mis-parse,” attributing the acoustic products of one sound’s articulation to another sound, and failing to disentangle the blend of their production. In this talk, I review the results of six phoneme categorization studies that demonstrate that such failure to completely un-do acoustic blending arises when listeners must judge one sound in a string relative to the sound that follows it, and the acoustic transitions between the two sounds are gradual. I then report the results of studies that demonstrate that listeners persist in attributing the acoustic products of a second sound’s articulation to a first sound even when the signal conveys early explicit evidence about the identity of that second sound, and so could have been leveraged to begin disentangling the first from the second before the second sound was fully realized. I go on to argue for a shift in our perspective toward segmental parsing. Attributing the product of a later sound’s articulation to an earlier sound seems inefficient or undesirable when we understand the goal of segmental parsing to be the complete attribution of acoustic products to exactly the sounds whose articulations gave rise to them. But when we consider the fact that listeners necessarily perceive the evidence for events in the world at a delay from when those events occurred, it is adaptive to prefer attributing later-incoming acoustic signal to earlier speech sounds.

Tianfan Xue will present “Visual Dynamics Probabilistic Future Frame Synthesis Via Cross Convolutional Networks” in the Machine Learning and Friends Lunch this Thursday at noon. Details, abstract and short bio are below.

Visual Dynamics Probabilistic Future Frame Synthesis Via Cross Convolutional Networks

Abstract:
We study the problem of synthesizing a number of likely future frames from a single input image. In contrast to traditional methods, which have tackled this problem in a deterministic or non-parametric way, we propose to model future frames in a probabilistic manner. Our probabilistic model makes it possible for us to sample and synthesize many possible future frames from a single input image. To synthesize realistic movement of objects, we propose a novel network structure, namely a Cross Convolutional Network; this network encodes image and motion information as feature maps and convolutional kernels, respectively. In experiments, our model performs well on synthetic data, such as 2D shapes and animated game sprites, as well as on real-world video frames. Please refer to our website for more details: http://visualdynamics.csail.mit.edu/.

Short bio:
Tianfan Xue is currently a fifth-year Ph.D. student in MIT CSAIL, working with William T. Freeman. Before that, he received his B.E. degree from Tsinghua Universtiy, and M.Phil. degree from The Chinese University of Hong Kong. His research interests include computer vision, image processing, and machine learning. Specifically, he is interested in motion estimation and image and video processing based on the motion information.

Time: 12:00pm to 1:15pm Wednesday Nov. 2. Location:  Tobin 521B. All are welcome!

Kevin Potter (University of Massachusetts)

Title:
Testing a perceptual fluency/disfluency model of priming with a model of response time and choice

Abstract:
With immediate repetition priming of forced choice perceptual identification, short prime durations produce positive priming (i.e., higher accuracy when the target is primed, but lower accuracy when the foil is primed). In contrast, long prime durations reverse this pattern. The dynamic time course of this transition from positive to negative priming is well explained by the nROUSE model of Huber and O’Reilly (2003), which includes neural habituation. This model assumes that the speed of perceptual identification is used to decide which choice word was seen most recently as the briefly flashed target. Thus, short duration primes induce faster identification (perceptual fluency) for the primed choice and a bias for the primed alternative whereas long duration primes induce slower identification (perceptual disfluency) for the primed choice and a bias against the primed alternative. This account makes specific predictions regarding perceptual identification latencies, and yet a test of these predictions is difficult with forced choice testing, which reflects a comparison decision process. To address this limitation, we collected forced-choice and single-item same-different responses in the same priming paradigm. We then applied a diffusion-race model to the data, transforming the response time and choice data into ‘observed’ drift rate parameters (i.e., the rate of evidence accumulation). Remarkably, the drift rates were inversely proportional to the identification latencies of the nROUSE model even though each model was applied independently to the data and even though the nROUSE model was only applied to the accuracy data. This convergence of the models confirms key predictions of the nROUSE model regarding perceptual fluency and disfluency.

Lera Boroditsky will present “How the languages we speak shape the ways we think” on Thursday, October 27, 4:30 PM, McConnell Foyer 103, Smith College.

Lera Boroditsky is an Associate Professor of Cognitive Science at UCSD and Editor in Chief of Frontiers in Cultural Psychology. She previously served on the faculty at MIT and at Stanford. Her research is on the relationships between mind, world, and language (or how humans get so smart). She has been named one of 25 Visionaries changing the world by the Utne Reader, and is also a Searle Scholar, a McDonnell scholar, recipient of an NSF Career award, and an APA Distinguished Scientist lecturer.

Time: 12:00pm to 1:15pm Wednesday Oct. 26. Location:  Tobin 521B.

Amy Criss (Syracuse University)

http://memolab.syr.edu/

Title: Memory across Tasks, Items, and Individuals

Memory researchers have spent the past several decades drilling down into memory. For example, many empirical investigations and theoretical developments focus on a single task and/or a single effect. We propose that the field is well-positioned to benefit from zooming back out. Here we present one attempt to do so. 462 participants completed 5 memory tasks with a fixed set of words that varied on many dimensions. We extract factors that are shared across memory tasks and we identify properties of items that support memory.  Finally, we sketch the beginnings of a conceptual model based on these findings.

Jason Q. Ng, Data Analyst Tumblr; Research Fellow, Citizen Lab
A data-driven approach to researching censorship and sensitive conversations on social media 

Like all nations, China has been profoundly affected by the emergence of the Internet, particularly new forms of social media which allow individuals themselves to be independent broadcasters of news. However the rise of “We Media” has also led to a corresponding rise in the filtering and blocking of online content in China. Identifying and explaining these disruptions comes with a host of challenges for researchers–ranging from technical ones like developing methodologies for tracking online censorship to non-technical ones like even defining what online censorship is. 

In this talk, we’ll look at a number of different ways online censorship can be defined as well as various data-driven techniques for revealing its occurrence in social media–as well as the various ways social media companies attempt to mask or justify it. However, just as important as identifying the mechanisms for how censorship is implemented is trying to understand the motivations for such behavior. Knowing both how and why online censorship occurs is key for not only academic researchers who hope to better understand content moderation and filtering practices, but it is essential information for the activists, journalists, and advocates who utilize such findings in their work.

About the speaker: Jason Q. Ng is currently a Research Fellow at the University of Toronto’s Citizen Lab, Data Analyst at Tumblr, and author of Blocked on Weibo, a book on Chinese social media. He is also a research consultant at China Digital Times where he develops censorship monitoring tools and teaches a digital activism course at Columbia SIPA. His writing and research projects can be found at www.jasonqng.com.