My primary research interest is in building a computationally-implementable theory of human behavior. This forces precision in theorizing, and allows the making of quantitative in addition to qualitative predictions. The two key tools I use towards this goal are bounded optimal control, and the modeling of cognitive and task architecture, which together form the core of computationally rational analysis. The central idea is to understand how behavior is jointly shaped by intrinsic constraints on the organism, and the constraints imposed by the task environment in which it operates. Providing additional empirical support for the approach, my own work shows evidence for adaptive behavior in humans in the very tasks I computationally investigate.
Adaptive eye-movement control
My dissertation is concerned with understanding adaptive eye-movement behavior in a simple wordlist-reading task. By abstracting away from sentence-level complexity, I am able to investigate the way that moment-by-moment eye movement decisions are conditioned on the ongoing process of word recognition. Some key interesting findings so far:
- Participants in our simple wordlist-reading task adapt their behavior to differences in quantitatively-expressed payoff, this adaptation is expressed in small but significant differences on the level of individual saccade decisions, and an adaptive model that is trying to optimize behavior in this task makes the same adaptation.
- Spillover lexical frequency effects — the robust fact that a high-frequency word results in the following word being read faster — cannot be solely a simple consequence of parafoveal preview alone, excluding a prominent candidate explanation for the effect (provided by the E-Z Reader model of Reichle and colleagues).
- Memory-driven sampling of past input can explain spillover effects in the absence of parafoveal preview, and doing so may in fact be adaptive under certain conditions.
Preview is not completely dead in the water, however: in work currently underway I am attempting to reconcile preview and memory-based explanations of the spillover frequency effect, and show how the effect's appearance and disappearance under different empirical parafoveal preview conditions can itself be a signature of adaptation to changes in the task environment.
Shvartsman, M., Lewis, R. L., and Singh, S. (submitted). Computationally Rational Saccadic Control: An Explanation of Spillover Effects Based on Sampling from Noisy Perception and Memory.
Shvartsman, M., Lewis, R. L., & Singh, S. (2014) Spillover frequency effects in a sequential sampling model of reading. Talk given at the 27th annual CUNY conference on human sentence processing. <10% talk acceptance rate.
Lewis, R. L., Shvartsman, M., & Singh, S. (2013). The adaptive nature of eye movements in linguistic tasks: how payoff and architecture shape speed-accuracy trade-offs. Topics in Cognitive Science, 5(3), 581–610. DOI: 10.1111/tops.12032. [pdf].
Shvartsman, M., Lewis, R. L., & Singh, S. (2013). A New Account of Spillover Effects in Reading Evidence from Parafoveal Masking. Poster presented at the 26th Annual CUNY Conference on Human Sentence Processing.
Shvartsman, M., Lewis, R. L., & Singh, S. (2012) The adaptive nature of eye-movement control in linguistic tasks. Talk given at the 25th annual CUNY conference on human sentence processing. <10% talk acceptance rate.
Shvartsman, M., Lewis, R., Singh, S., Smith, M., & Bartek, B. (2011). Predicting Task Performance from Individual Variation in Eye-Movement Control Strategies. Poster presented at the 24th annual CUNY conference on human sentene processing.
Towards a theory of memory cues for sentence processing
A large body of recent work in psycholinguistics is seeking understand so-called illusions of grammaticality: sentences which pass muster in normal use or reading, but seem obviously problematic on closer inspection, for example The key to the cabinets are on the table, where the plural verb are disagrees in number with its singular subject key. The most successful theories to date rely on a cue-based memory system: the verb seeks valid subjects by approximate cue-matching, and the salient plural cue on cabinets makes the sentence seem fine. We are trying to understand what can be a cue in this context, and when different cues can become available for matching, using the behavior stop-signal ('speed-accuracy tradeoff') paradigm.
Language representations in bilinguals
A central question in the research on bilingual speakers is how their mental language representations are organized. How much of the representations are shared, and how much are distinct? Does this vary by speaker or language pair? We are working on teasing apart whether structure or merely word order representations are connected, by using constructions where only one of the two is shared, in an on-line eyetracking task.
This is work in progress with Guadelupe de los Santos (UMich) and Julie Boland (UMich).
In this project, we were interested in how early domain-specific information is available in auditory perception. We used the oddball paradigm in MEG, a technique that picks up very early signs of expectation violation or mismatch detection. We constructed material pairs where the acoustic difference yielding the mismatch was the same, but the acoustic information (not differeing between the pair) was consistent with the stimulus being in the language domain (vowels), music domain (piano intervals), or neither (sine waves). The key finding was that early auditory mismatch detection was already impacted by domain-specific acoustic information.
Bergelson, E., Shvartsman, M., & Idsardi, W. J. (2013). Differences in mismatch responses to vowels and musical intervals: MEG evidence. PLoS One, 8(10). DOI: 10.1371/journal.pone.007675. [pdf].
Bergelson, E., Shvartsman, M., & Idsardi, W. J. (2010) Differences in Brain Responses to Vowels and Musical Intervals. In S. M. Demorest, S. J. Morrison, & P. S. Campbell (Eds.). Proceedings of the 11th International Conference on Music Perception and Cognition.
Shvartsman, M., Bergelson, E. & Idsardi, W.J. (2009) From tones to vowels: a neurophysiological investigation of sine and formant dyads. Poster presented at the first annual Neurobiology of Language Conference.
If language is the solution, then what is the problem? And why does the solution look the way it does? A reasonable answer to the first question is that the problem is communication, and that communication is generally a helpful and good thing. A reasonable answer to the second question is that it's the best thing that biological evolution and cultural adaptation managed to cobble together from what it had to work with. But those are imprecise answers. In this work, we were aiming for greater precision, by setting up an environment with agents who needed to solve a simple task, and seeing what properties of agent, environment, and task were necessary for particular properties of language to emerge as useful. Some of the properties we were interested in was in the systematicity of symbol ordering (whether symbol order matters), and ambiguity (whether a given symbol always means the same thing). We looked at how those properties were affected by constraints on the agent (like memory size and accessibility) and the environment (like the ability to communicate and act on communicated information at the same time).
*Bratman, J., *Shvartsman, M., Lewis, R. L., & Singh, S. (2010). A new approach to exploring language emergence as boundedly optimal control in the face of environmental and cognitive constraints. In Salvucci, D. and Gunzelmann, G., editors, Proceedings of the 10th International Conference on Cognitive Modeling. (*The first two authors made equal contribution to this work.)