On Monday 11/21, John Alderete (Simon Fraser U) will give a talk entitled “Language generality in phonological encoding: Moving beyond Indo-European languages” at 1:15pm in ILC N400 (& Zoom: see below). The title and abstract of his talk can be found below. Everyone is welcome!

Dr. Alderete’s research gives close formal analyses of phonological and morphological systems, and asks how these systems are learned and represented in larger theories of cognitive science. His dissertation and related work addressed theoretical questions concerning how phonological structures, and the morphological influences on these structures, can be formalized in constraint-based theories of grammar. His later work extended these questions using fieldwork, computational, and experimental methods to probe the learnability and viability of these structures, as well as how they are encoded and accessed in online production processes. His current focus, actively pursued with his students in the Language Production Lab, investigates how language particular phonological structures shape speech production processes.

Title: Language generality in phonological encoding: Moving beyond Indo-European languages

Abstract: Theories of phonological encoding are centred on the selection and activation of phonological segments, and how these segments are organised in word and syllable structures in online processes of speech planning. The focus on segments, however, is due to an over-weighting of evidence from Indo-European languages, because languages outside this family exhibit strikingly different behaviour. We examine speech error, priming, and form encoding studies in Mandarin, Cantonese, and Japanese, and argue that these languages deepen our understanding of phonological encoding. These languages demonstrate the need for language particular differences in the first selectable (proximate) units of phonological encoding and the phonological units processed as word beginnings. Building on these results, an analysis of tone slips in Cantonese suggests that tone is processed concurrently with segments and sequentially assigned after segments to fully encoded syllables.

Zoom: https://umass-amherst.zoom.us/j/93267322657?pwd=UHdaUWNZTGFOeDZZYktlSTlMWGpMZz09

Meeting ID: 932 6732 2657; Passcode: 463946).

This week at MLFL we will host Rana Hanocka, an Assistant Professor at the University of Chicago. Dr. Hanocka’s research is focused on building artificial intelligence for 3D data, spanning the fields of computer graphics, machine learning, and computer vision. Deep learning, the most popular form of artificial intelligence, has unlocked remarkable success on structured data (such as text, images, and video), and Dr. Hanocka is interested in harnessing the potential of these techniques to enable effective operation on unstructured 3D geometric data.

who: Rana Hanocka (https://people.cs.uchicago.edu/~ranahanocka/)
when:  Thursday, 11/17/2022, 12pm-1pm
where: CS 150/151 (pizzas available), Zoom
generous sponsor: ORACLE LABS

On 11/18 (Fri) in Integrative Learning Center (ILC) Room S331 from
3:30 pm to 5:00 pm, Susi Wurmbrand (Harvard & Universität Wien) will give a linguistics colloquium talk.

Implicational complementation hierarchies: Containment and the freedom of syntax

“Typological and cross-linguistic observations show that complementation configurations can be ranked according to their semantic properties, forming an implicational complementation hierarchy along which syntactic or morphological distinctions operate. I suggest a model where the cross-linguistically stable (possibly universal) properties follow from a rigid syntax?semantic mapping of categories defined via containment, whereas variable properties indicate the points where syntax may act autonomously.  I will discuss several phenomena where implicational relations have been observed (among them finiteness, transparency, restructuring, the left periphery) and show that they can be related to truncation options (whether implemented via exfoliation, structure removal or non-projection) regulated by containment.”

This week in Cognitive Brownbag in Tobin 521B, we will hear from Eleonore Neufeld (UMass Philosophy). Dr. Neufeld received my PhD in Philosophy from the University of Southern California in 2020. Dr. Neufeld’s research interests are in philosophy of mind and cognitive science, philosophy of language, and social philosophy.

If you would like to hear her talk but can’t make it to Tobin, you are welcome to join on zoom
https://umass-amherst.zoom.us/j/97623669473?pwd=bU9aTVo1c1U3ZVBuVDJ3QmFwUmVHUT09
Meeting ID: 976 2366 9473
Passcode: Cog2223

This week at MLFL we will host Rachel Rudinger, an Assistant Professor at the University of Maryland, College Park. She was previously a Young Investigator at the Allen Institute for AI in Seattle, and a visiting researcher at the University of Washington. Her research interests include computational semantics, common-sense reasoning, and issues of social bias and fairness in NLP.

If you are interested in an individual meeting about research with Prof. Rudinger, please sign up here.

who: Rachel Rudinger (https://rudinger.github.io/)
when:  Thursday, 11/3/2022, 12pm-1pm
where: CS 150/151 (pizzas available), Zoom
generous sponsor: ORACLE LABS

Title: “Not so fast!”: Revisiting assumptions in (and about) Natural Language Reasoning

Abstract:
In recent years, the field of Natural Language Processing has seen a profusion of tasks, datasets, and systems that facilitate reasoning about real-world situations through language (e.g., RTE, MNLI, COMET). Such systems might, for example, be trained to consider a situation where “somebody dropped a glass on the floor,” and conclude it is likely that “the glass shattered” as a result. In this talk, I will discuss three pieces of work that revisit assumptions made by or about these systems. In the first work, I develop a DefeasibleInference task, which enables a system to recognize when a prior assumption it has made may no longer be true in light of new evidence it receives. The second work I will discuss revisits partial-input baselines, which have highlighted issues of spurious correlations in natural language reasoning datasets and led to unfavorable assumptions about models’ reasoning abilities. In particular, I will discuss experiments that show models may still learn to reason in the presence of spurious dataset artifacts. Finally, I will touch on work analyzing harmful assumptions made by reasoning models in the form of social stereotypes, particularly in the case of free-form generative reasoning models.

Bio:
Rachel Rudinger is an Assistant Professor in the Department of Computer Science at the University of Maryland, College Park. She holds joint appointments in the Department of Linguistics and the Institute for Advanced Computer Studies (UMIACS). In 2019, Rachel completed her Ph.D. in Computer Science at Johns Hopkins University in the Center for Language and Speech Processing. From 2019-2020, she was a Young Investigator at the Allen Institute for AI in Seattle, and a visiting researcher at the University of Washington. Her research interests include computational semantics, common-sense reasoning, and issues of social bias and fairness in NLP.

Zoom:
https://umass-amherst.zoom.us/j/92643805933?pwd=bjJjUFBjNHdCdTBoR2RMbmRWWHJtUT09

Meeting ID: 926 4380 5933
Passcode: 140150
One tap mobile
+13126266799,,92643805933# US (Chicago)
+16468769923,,92643805933# US (New York)

On 10/21 (Friday) at 3:30 in ILC S331, Liina Pylkkänen of NYU will give the following talk:

The syntactic, semantic and associative brain: How conceptual combination interacts with syntactic composition and associative processing in temporal cortex.

What is the neural basis of syntactic and semantic composition? Our lab’s research has revealed a consistent neural correlate of composition in the left anterior temporal lobe (LATL), tracking conceptual aspects of composition, and more elusive correlates of syntactic composition in posterior temporal cortex. Is the LATL fully independent of syntactic composition, or does it conceptually combine elements that also syntactically combine? That is, how does LATL activity relate to more structural aspects of composition? On the flip side, we can also investigate how the LATL interacts with *less* structural aspects of processing, namely pure association, which has also been shown to drive LATL activity, including in the monkey brain. I will report on our recent progress on these questions, as we gradually make headway in understanding how the combinatory machine works in our brains.

This talk can be watched in person in S331 in the ILC or remotely using the zoom link:

Join Zoom Meeting
https://umass-amherst.zoom.us/j/92526730268?pwd=ZDE4MytyNnRucTY1TDZCZnJYRnA1dz09

Meeting ID: 925 2673 0268
Passcode: 406498

On this Friday 10/21 11:30-1:00 in ILC Linguistics Dept R458, Tom Roeper will give a report on the Wuppertal conference “Optionality and Variation in L2 Acquisition” (Oct 5-7). He will also present his contribution to the conference:

“How Minimal Interfaces Guide L2 Acquisition and How Minimal Interfaces Express the Connection between Syntax and Cognition.”

Before Tom’s presentation, we will start the meeting with planning some sections to provide support for experimental project design from graduate and undergraduate students.

This provides an ideal opportunity to learn how to do (or hone) experiments focused on basic linguistic concepts. Bring your ideas for the meeting!

Forthcoming (Date TBD): presentation by Antonella Sorace (Univ. of Edinburgh) who will talk about bilingualism both at the theoretical and practical levels.

Everyone welcome!

[A good opportunity to learn about what is going on in L1,L2, Multiliingualism]

On Wednesday, October 19th at noon in Tobin 521B, Dr. Brown will present:  Assessing the costs of visualizing uncertainty in electronic displays.

Feel free to pass along the invitation to others.  If you would like to hear her talk but can’t make it to Tobin, you are welcome to join on zoom
https://umass-amherst.zoom.us/j/97623669473?pwd=bU9aTVo1c1U3ZVBuVDJ3QmFwUmVHUT09
Meeting ID: 976 2366 9473
Passcode: Cog2223

Uncertainty in geospatial data refers to ambiguity that can arise from incomplete data, calculation errors, noisy equipment, and scaling issues to name a few (e.g., Pang, 2001; Pang, Wittenbrink, & Lodha, 1997).  In general, uncertainty is an inaccuracy that is not accounted for but is likely to result in user error because it occurs in finished products like maps and geospatial applications. However, visualizing uncertainty adds a layer of visual information, which is likely to increase the amount of visual clutter and can negatively impact usability (Beck, Trafton, & Lohrenz, 2010; Lohrenz & Beck, 2010). Additionally, processing of information presented in geospatial displays is heavily influenced by the layout, graphic design, and purpose which need to be taken into account (e.g., Hegarty, 2013; Liben, 2009; You, Chen, Liu, & Lin, 2007). Thus, there is a general need to display or visualize uncertainty to improve usability and decision making complicated by an increased load on cognitive processes by doing so.
The current study compared three techniques for visualizing geospatial uncertainty to determine if they differentially affected the cognitive processes required to search for targets within electronic nautical charts.  We manipulated the amount of clutter in the charts by varying the range of uncertainty and including fabricated wind, pressure and electronic navigation icons as additional geospatial layers.  As predicted, performance declined as the amount of clutter increased from all geospatial variables including uncertainty. Interestingly, the type of visualization had the greatest effect on performance. The visualization that was most distinct resulted in the worst performance, as it also increased the amount of visual clutter in the chart significantly more than the other uncertainty visualization types. The results have implications for the design and use of electronic maps of all types (road maps, maritime, and aeronautical charts).

Dr. Brown received her PhD in Psychology from Louisiana State University in 2011. She initially joined the US Naval Research Laboratory as a Karles Fellow with the Geospatial Human-Computer Interaction group at the Stennis Space Center where her research focused on decision support in geospatial displays. In 2018, she transferred to the Warfighter Applied Cognition and Technology Lab in DC where her research has pivoted to focus on individual differences in cognitive ability. Her research centers on basic and applied applications of attention, memory and decision making. Of particular interest are the ways in which individual differences in attention and memory can be leveraged to support Naval recruitment, selection and training.

On 10/20 (Thursday) at noon in Computer Science Building Room 150/151, Dimitris Tsipras will give a presentation. More details can be found here. Everyone is welcome!

Machine Learning and Friends Lunch is a weekly interactive forum at UMass Amherst where friends with broad interests in machine learning methods and applications gather for presentations on cutting-edge research.

MLFL is held on Thursdays from 12:00 pm to 1:00 pm Eastern Time in Computer Science Building Room 150/151, unless otherwise noted. Talks are broadcast via Zoom. Pizzas are available starting at 11:50 am. Everyone is welcome.

Click to view this semester’s schedule, event announcements on our college website, talks from previous semesters, and our Youtube channel for recorded talks.

MLFL has been graciously sponsored by our friends at Oracle Labs.

On Thursday 10/13 @12:00 in Computer Science Building room 150/151, Daniel Munoz Huerta will give a talk. More detail can be found here.

Machine Learning and Friends Lunch is a weekly interactive forum at UMass Amherst where friends with broad interests in machine learning methods and applications gather for presentations on cutting-edge research.

MLFL is held on Thursdays from 12:00 pm to 1:00 pm Eastern Time in Computer Science Building Room 150/151, unless otherwise noted. Talks are broadcast via Zoom. Pizzas are available starting at 11:50 am. Everyone is welcome.

Click to view this semester’s schedule, event announcements on our college website, talks from previous semesters, and our Youtube channel for recorded talks.

MLFL has been graciously sponsored by our friends at Oracle Labs.