Time: 12:00pm to 1:15pm Wednesday Oct. 12.

Location:  Tobin 521B

Roger Levy (M.I.T.)
https://bcs.mit.edu/users/rplevymitedu

Probabilistic models of human language comprehension

Human language use is a central problem for the advancement of machine intelligence, and poses some of the deepest scientific challenges in accounting for the capabilities of the human mind.  In this talk I describe several major advances we have recently made in this domain that have led to a state-of-the-art theory of language comprehension as rational, goal-driven inference and action. These advances were made possible by combining leading ideas and techniques from computer science, psychology, and linguistics to define probabilistic models over detailed linguistic representations and testing their predictions through naturalistic data and controlled experiments.  In language comprehension, I describe a detailed expectation-based theory of real-time language understanding that unifies three topics central to the field — ambiguity resolution, prediction, and syntactic complexity — and that finds broad empirical support. I then move on to describe a “noisy-channel” theory which generalizes the expectation-based theory by removing the assumption of modularity between the processes of individual word recognition and sentence-level comprehension.  This theory accounts for critical outstanding puzzles for previous approaches, and when combined with reinforcement learning yield state-of-the-art models of human eye movement control in reading.

Florian Schwarz from the University of Pennsylvania (UMass 2009 PhD) will give a Department Colloquium on Friday, October 7th, at 3:30 PM and will also lead a joint session of the Semantics and Psycholinguistics Workshops on Thursday, October 6th, from 6:00 to 7:15 PM. Both events will take place in N400. Everyone is welcome. Here is a link to a folder with background readings and more information.

 One of the main questions discussed in the recent presupposition literature is whether we need to differentiate different sub-classes of triggers (or reconsider classifications of different aspects of meaning more broadly altogether). A number of theoretical proposals have been put fourth, most prominently a pragmatic reanalysis of certain triggers based on reasoning over alternatives, akin to implicatures (Simons 2001, Abusch 2010, Romoli 2014). Other proposals distinguish triggers based on their anaphoric properties or the way in which they contribute to their local context in embedded environments. An alternative view, advanced by Abrusan (2011, 2016), is that there is no one fundamental underlying difference between types of triggers, and that observed variations in their behavior can be accounted for on the basis of various orthogonal factors. In this talk, I present results from several series of experiments aiming to assess these different theoretical perspectives. First, I present comparisons between implicatures and so-called `soft’ triggers that pose a serious challenge to a variety of pragmatic approaches of the latter. Next, I turn to experimental studies directly comparing (allegedly) different types of triggers, which do provide evidence suggesting that inherently distinct trigger types need to be distinguished, rather than explaining all variation in independent terms. I conclude with a discussion of how well the remaining theoretical options for distinguishing triggers fit with the existing data, and point to future directions for further refining our theoretical understanding of differences between triggers.

Many current theoretical accounts of presupposition projection are crucially based on specific assumptions about how presuppositions are interpreted in incremental processing. However, little is known about the actual processing mechanisms involved in comprehending triggers in embedded environments. I discuss two experiments looking at presuppositions in disjunctions, using covered box picture selection tasks combined with visual world eye tracking. The results provide evidence for Rapid Incremental Presupposition Evaluation (RIPE). We will then discuss how this finding relates to theoretical options for capturing projection. Finally, we will consider other recent results on incrementality in processing presupposition projection and how they can be integrated into the larger emerging picture of the role of processing considerations in our understanding of presuppositional phenomena.

(1) The UMass Philosophy Department is pleased to present Miriam Solomon from Temple University who will be speaking on The Historical Epistemology of Evidence-Based Medicine Thursday, 10/6/2016 at 4:00 PM in the Integrative Learning Center S140. Please join us! All are welcome to attend.

Abstract: What evidence determines a medical consensus? What methods generate good evidence? Does it matter who conducts the experiments? How are patients impacted? Miriam Solomon explores these questions and how evidence-based medicine has superseded other accounts of objectivity in medicine resulting in responses from the areas of translational medicine, personalized medicine, and precision medicine.

(2) Prof. Solomon will be giving a colloquium talk, sponsored by the Philosophy Department, titled “Agnotology, hermeneutical injustice, and scientific pluralism: the case of Asperger Syndrome”, on Friday, Oct. 7, at 3:30 PM in Bartlett 206.

Abstract: Agnotology and hermeneutical injustice are among the most fruitful new ideas in social epistemology. When the ideas were first presented, they came with examples that have become canonical: lost knowledge of abortifacients and climate change denial (for agnotology) and postpartum depression, sexual harassment and sexual identity (for hermeneutical injustice). These examples have been useful for introducing the concepts of agnotology and hermeneutical injustice, but they oversimplify the epistemology. The purpose of this paper is to explore a case—the diagnostic category of Asperger Syndrome , embraced in the late 1980s and early 1990s, and then jettisoned in 2013—in which it is essential to acknowledge the more complex epistemic situation.

Joshua Hartshorne

Assistant Professor, Boston College

http://l3atbc.org/index.html

A critical period for second language acquisition: A study of 700,000 English speakers

Children learn language more easily than adults, though when and why this ability declines have been obscure for both empirical reasons (underpowered studies) and conceptual reasons (measuring the ultimate attainment of learners who started at different ages cannot by itself reveal changes in underlying learning ability). I address both limitations with a dataset of unprecedented size (669,498 native and non-native English speakers) and a computational model that estimates the trajectory of underlying learning ability by disentangling current age, age at first exposure, and years of experience.  These reveal that grammar-learning ability is preserved almost to the crux of adulthood (17.4 years-old) and then declines steadily. The results support the existence of a critical period for language acquisition, but the age of offset is much later than previous estimates. We show that the results are nonetheless consistent with published data, and consider possible explanations and experimental paths forward.

Date: Thursday October 6^th

Time: 1:00-2:00pm

Location: Tobin 423

What: tea, refreshments, presentations and conversations about topics in data science
Event: Deep Learning Whiteboard Talks
When: 4-5 pm October 4
Where: Computer Science Building Rooms 150 & 151
Who: You!  Especially MS & PhD students and faculty interested in data science.

The presenters include:

James Atwood (PhD Student advised by Prof. Don Towsley)
Diffusion-Convolutional Neural Networks 

Abstract: We present diffusion-convolutional neural networks (DCNNs), a new model for graph-structured data.  Through the introduction of a diffusion-convolution operation, we show how diffusion-based representations can be learned from graph-structured data and used as an effective basis for node classification. DCNNs have several attractive qualities, including a latent representation for graphical data that is invariant under isomorphism, as well as polynomial-time prediction and learning that can be represented as tensor operations and efficiently implemented on the GPU.  Through several experiments with real structured datasets, we demonstrate that DCNNs are able to outperform probabilistic relational models and kernel-on-graph methods at relational node classification tasks. 
Daniel Cohen (MS/PhD Student advised by Prof. Bruce Croft)
Memory Networks for First Sentence Detection 

Abstract: In passage retrieval, detecting the beginning and end of a passage at the sentence level granularity is a challenging problem. Due to the varying levels of relevance for each sentence, We employ a deep memory network to capture long term dependencies which a variety of RNN’s fail to recognize.
Huaizu Jiang (PhD Student advised by Prof. Erik Learned-Miller)
Learning to predict object occlusions in videos

Abstract: Given first two frames of a video containing a set of moving objects, the goal is to predict where they are going (locations) and their visibility (fully visible, partially visible, or fully invisible) in the next immediate frame. Due to the perspective effect of the camera, the object closer to the camera (thus more likely to be visible) tends to be larger and moving faster. In this project, we are interested in training Convolutional Neural Networks (CNNs) to learn such common sense knowledge.  Additional information, such as  appearance and shading, may also provide cues to the object occlusions prediction. 

12:00pm to 1:15pm Location:  Tobin 521B

Nate Kornell (Williams College)
http://sites.williams.edu/nk2/

Title: Learning is from Mars, student perceptions of learning are from Venus

Abstract
People learn more when they engage in productive struggle than when learning is easy. Students often think the opposite, that they have learned less when they have been made to struggle. This research suggests that students are not always reliable judges of how much they have learned. I will discuss this research as well as research on a related question: Do the best teachers get the best ratings from their students?