Yearly Archives: 2016

Rickford to give Freeman Lecture in Linguistics Feb. 17th

Professor John Rickford (http://www.johnrickford.com/) of Stanford University will deliver this year’s annual Freeman Lecture in Linguistics on Friday February 17th at 2:30 in ILC N151. The title of Professor Rickford’s talk will be “Justice for Jeantel (and Trayvon): Fighting Dialect Prejudice in Courtrooms and Beyond.”

Professor Rickford is a world-renowned expert in the structure, history, and dialectology of African American English. He is the author of numerous books on AAE, including ‘Spoken Soul: The Story of Black English’, ‘African American Vernacular English’, and his most recent book, ‘African American, Creole, and Other English Vernaculars in Education’.

CS 585 Natural Language Processing at Data Science Tea Tues. Dec. 13, 2:45

——————————————————————————–
What: tea, posters and conversations about NLP
When: Tuesday, Dec 13, 2:45 – 3:45 pm
Where: Computer Science Building Rooms 150 & 151
Who: You!  Especially MS & PhD students and faculty interested in data science.
——————————————————————————–

Join us at Data Science Tea for a poster session featuring course projects from Professor Brendan O’Connor’s CS 585 Introduction to Natural Language Processing, including 60+ posters on topics like sentiment analysis, sarcasm detection, identifying portmanteaus, analyzing song lyrics, creating timelines from news, detecting bullying tweets, and more, using machine learning and computational linguistic methods.

3rd annual UMass Cognitive Science Workshop on Feb. 3rd, 2017

The 3rd annual UMass Cognitive Science Workshop will take place from 2:30 – 5 on Friday Feb. 3rd, 2017 in ILC N400. The talks listed below will be followed by a poster session and reception. If you would like to present a poster, please fill out the short form linked here. We encourage the presentation of research that has previously been presented elsewhere, though new work is of course welcome!

2:30 Rosie Cowell, Cognitive Division, Psychological and Brain Sciences

3:00 Meghan Armstrong-Abrami, Hispanic Linguistics, Languages, Literatures and Cultures

3:30 Florence Sullivan, Education.

 

Fornaciai in Cognitive Brown Bag Weds. Dec. 14 at noon

Michele Fornaciai (UMass Amherst) will be presenting “Temporal evolution of visual representation: From physical to perceived numerosity” in the Cognitive Brown Bag Wednesday Dec. 14 – all are welcome.
Time: 12:00pm to 1:15pm  Location:  Tobin 521B.
Abstract. Humans share with many animals a number sense, the ability to estimate rapidly the approximate number of items in a scene. Recent work has shown that like many other perceptual attributes, numerosity is susceptible to adaptation. It is not clear, however, whether adaptation works directly on mechanisms selective to numerosity, or via related mechanisms, such as those tuned to texture density. To disentangle this issue, we measured numerosity adaptation of 10 pairs of connected dots, as connecting dots makes them appear to be less numerous than unconnected dots. Adaptation to a 20-dot pattern (same number of dots as the test) caused robust reduction in apparent numerosity of the connected-dot pattern, but not of an unconnected dot-pattern. This suggests that adaptation to numerosity, at least for relatively sparse dot-pattern, occurs at neural levels encoding perceived numerosity, rather than at lower levels responding to the number of elements in the scene. However, little is known about the processes that make a given percept available in the content of subjective visual awareness, and when such percepts arise in the visual stream. We then exploited the connectedness illusion to investigate when the perceptual representation of numerosity emerges in the visual processing stream. We recorded brain activity by means of electroencephalogram while participants passively viewed a stream of arrays containing 16 or 32 dots, either isolated or pair-wise connected by straight lines. The results showed that the early latency visual evoked potentials (~ 100 ms) reflect physical, rather than perceived, numerosity, while the later latency potentials (~ 150 ms) reflect perceived, rather than physical, numerosity. A multivariate pattern analysis in the time domain confirmed such a pattern and further demonstrated that both the effects of physical and perceived numerosity persist until later latency (~ 400 ms). These results demonstrate that the physical information of a visual scene undergoes a series of manipulations along the visual stream that could radically change its content before being available in the subjective visual experience.

Roon in Linguistics Fri. Dec. 9th at 3:30

Kevin Roon, Post-doctoral Associate,
Speech Production, Acoustics, and Perception Lab, CUNY Graduate Center,
Friday, December 9, 3:30-4:30, in ILC N400.

Title: Modeling feature-driven modulations of phonetic output and verbal
response times

Abstract:
In this talk I will present a series of experiments that show that
verbal response times and phonetic output in various tasks can be
modulated by phonological features, but that not all features affect RTs
the same way. The focus of the talk will be on discussing possible
reasons for these differences. In Roon and Gafos (2015), we have shown
that RTs for stop-initial CV nonce syllables in a response-distractor
task are sensitive independently to voicing and primary oral
articulator. The experiments from Roon and Gafos (2015) showed another
surprising effect based on features: RTs were much slower in an
experiment where participants knew the articulator of their response but
could not predict the voicing than in an experiment where they knew the
voicing but could not predict the articulator. Yuen et al. (2010) showed
that participants reliably had more alveolar constriction when saying
/kab/ or /sab/ when they heard a /tab/ distractor than when they heard a
distractor that was the same as their utterance. In Roon and Gafos
(2016) we have proposed a dynamical computational model of phonological
planning that accounts for both the feature-level RT modulations and the
cross-experiment RT differences from Roon and Gafos (2015) as well as
for the differences in alveolar constriction found by Yuen et al.
(2010). In related work, Mouskou, Roon, & Rastle (2014) showed that in a
masked-prime experiment, participants RTs for reading CVC nonce
syllables aloud were faster when the prime and target differed only in
voicing (e.g., piz-BAF) than when they differed in many features (e.g.,
suz-BAF). However, in experiments we are running now, we do not find a
similar effect when the prime and target differ only in place (e.g.,
diz-BAF). I will discuss how various formal representations of these
features might account for the differences in RT modulation.

Cataldo at Cognitive Brown Bag Weds. Dec 7 at noon

Andrea Cataldo (UMass Amherst)

Time: 12:00pm to 1:15pm Wednesday Dec. 7  Location: Tobin 521B.
Title: A Bayesian Hierarchical Signal Detection Model for Rating Scale Data

Abstract: The rating scale is a predominant method of quantifying internal states. For instance, emotion is typically measured by asking individuals to rate how well each of several emotional words represents their current state. Group mean ratings are then compared to determine which emotion was rated highest. This method of analysis is limited in two ways: First, the finite number of stimuli available to target particular states, such as emotion, reduces within-subjects power. Second, comparing mean ratings is an insensitive measure of how individuals discriminate between possible states. This talk presents a Bayesian hierarchical signal detection model for rating scale data as a solution to both limitations: signal detection theory first offers a more sensitive discrimination measure, and the Bayesian hierarchical framework provides more reliable estimates at the individual level.

Turchin in CSSI Fri. Dec. 9th at 12:30

The UMass Computational Social Science Institute invites you to our weekly CSSI seminar.

Peter Turchin
Ecology and Evolutionary Biology, University of Connecticut

Friday, December 9, 2016 • 12:30 p.m.-2:00 p.m (lunch served at noon)

Computer Science Building, Room 150/151Title: Ages of Discord: A Structural-Demographic Analysis of Political Violence Waves 

Abstract:  A useful approach to thinking about outbreaks of political violence (scaling up to revolutions and civil wars) is to separate their causes into structural conditions and triggering events. Specific triggers of political upheaval, such as self-immolation of a Tunisian fruit vendor, are very hard, perhaps impossible to predict. On the other hand, structural pressures build up slowly and predictably, and are amenable to analysis and forecasting. Quantitative historical analysis reveals that complex human societies are affected by recurrent — and predictable — waves of political violence. The structural-demographic theory suggests that such seemingly disparate social indicators as stagnating or declining real wages, a growing gap between rich and poor, overproduction of young graduates with advanced degrees, and exploding public debt, are actually related to each other dynamically. Historically, such developments have served as leading indicators of looming political instability. In my presentation I will describe a dynamical model based on structural-demographic theory and illustrate it with data on economic, social, and political dynamics in nineteenth century America, including the most violent episode of political instability in the U.S. history, the American Civil War. I also discuss what this theory tells us about the U.S. today.

Bio:  Peter Turchin is a scientist and an author who wants to understand how human societies evolve, and why we see such a staggering degree of inequality in economic performance and effectiveness of governance among nations. Peter’s approach to answering these questions blends theory building with the analysis of data. He is the founder of a new transdisciplinary field of Cliodynamics, which uses the tools of complexity science and cultural evolution to study the dynamics of historical empires and modern nation-states. Peter has published two hundred articles, including a dozen in such top journals as Nature, Science, and PNAS. Turchin has authored seven books, including Secular Cycles (with Sergey Nefedov, Princeton, 2009), and War and Peace and War (Penguin, 2005). Currently Peter’s main research effort is directed at coordinating the Seshat Databank—a massive historical database of cultural evolution that is gathering and systematically organizing the vast amount of knowledge about past human societies, held collectively by thousands of historians and archaeologists.

Madsen in Data Science Tea Monday 4-5 pm December 5

Please join us for Data Science Tea on Monday December 5!

——————————————————————————–
What: tea, refreshments, presentations and conversations about topics in data science
Speakers: Miriam Madsen
When: Monday 4-5 pm December 5
Where: Computer Science Building Rooms 150 & 151
Who: You!  Especially MS & PhD students and faculty interested in data science.
——————————————————————————–

Designing a Novel Manual Communication System for Mechanically Ventilated (MV) ICU Patients

In the hospital Intensive Care Unit (ICU), available communication methods for patients using MV – e.g., breathing tubes – are insufficient. While a number of basic communication methods are often tried with these patients (such as writing on whiteboards/clipboards, use of letter boards, and mouthing words), patients and caregivers consistently report dissatisfaction with available methods and limited success in their use.1 Furthermore, patients consider the lack of successful communication strategies to be extremely stressful.2 The ICU setting is more complicated for assistive communication technology use than other settings due to patient population heterogeneity; variation in individuals’ physical/cognitive capabilities over time; robustness and hygiene concerns for communication tools; and a lack of available training time prior to patients’ need for communication assistance.
A novel communication system is under development, with the goal of being easily accessible by MV patients who lack sufficient dexterity to write clearly due to complications of critical illness. Using this system, a patient will manually operate a hand-held component that communicates in real time with a tablet computer, producing audiovisual content specific to ICU patient needs.

Three sets of system requirements have guided the design of this technology: features required by the ICU setting, by the patient, and by the nurses/care team. Features required by the ICU setting include the presence of ICU-specific topics and cost-effective design that is appropriately hygienic. The patient-related system requirements involve a short learning curve; hardware and software that is adaptable to individual patients; and inclusion of non-medical topics of value to patients and their families. Synthesized speech output and some form of tactile feedback are also planned for the final system version to meet patient needs. In considering the requirements of the nurses and care team, the system should be accessible despite physical restraints and should demonstrate a patient’s level of responsiveness, allowing for a clearer assessment of a patient’s cognitive state.

This system has been demonstrated in its prototype form to physicians, nurses, researchers, and engineers; based on their feedback, the system has undergone over two dozen iterations in preparation for deployment with MV ICU patients.  Patient testing in a proof-of-concept pilot trial began in November of 2016, and the system is under active development and revision.

Biography
Miriam received her bachelor’s and master’s degrees at MIT (2009, S.B. in Computer Engineering; 2010, M.Eng. in Autism Technologies). At MIT, she worked with the Speech & Communications Laboratory and the Smart Wheelchair Project in the Department of Electrical Engineering & Computer Science, as well as with Affective Computing, Biomechatronics, and the Voting Technology Project in the MIT Media Lab. She then completed a premedical preparation program at UMass Boston worked as a software engineer in industry.

She is currently in the MD/PhD program at UMass Medical School in Worcester. In her PhD program, she is in the departments of Neurology and Anesthesiology, as well holding a position as a Visiting Research Fellow in Brown University’s School of Engineering.