Due to the COVID-19 pandemic, this symposium will be held remotely, and can be viewed publicly over Zoom. It will take place on Monday, April 19th, from 9:45AM to 3PM (EST), and will feature presentations by:
Barbara Partee (UMass Amherst)
Gennaro Chierchia (Harvard University)
Pauline Jacobson (Brown University)
Florian Schwarz (University of Pennsylvania)
Seth Cable (UMass Amherst)
Christopher Potts (Stanford University)
The website for the symposium, which includes the full program (with abstracts) as well as the Zoom link for the remote presentations, can be found at the link below:
Max Nelson, Joe Pater and Brandon Prickett presented “Representations in neural network learning of phonology” in the UCLA colloquium series Friday October 9th. The abstract is below, and the slides can be found here.
Abstract. The question of what representations are needed for learning of phonological generalizations in neural networks (NNs) was a central issue in the applications of NNs to learning of English past tense morphophonology in Rumelhart and McClelland (1986) and in following work of that era. It can be addressed anew given subsequent developments in NN technology. In this talk we will present computational experiments bearing on three specific questions:
Are variables needed for phonological assimilation and dissimilation?
Are variables needed to model learning experiments involving reduplication (e.g. Marcus et al. 1999)?
What kind of architecture is necessary for the full range of natural language reduplication?
Brandon Prickett successfully defended his PhD dissertation “Learning Phonology with Sequence-to-Sequence Neural Networks” on Wednesday October 7th. Gaja Jarosz and Joe Pater were the co-chairs, and John Kingston and Mohit Iyyer (CICS) were the other committee members. Congratulations Brandon!
Shravan Vasishth (vasishth.github.io), University of Potsdam, will present “Twenty years of retrieval models” in the Linguistics colloquium series at 3:30 Friday September. An abstract follows. All are welcome!
After Newell wrote his 1973 article, “You can’t play twenty questions with nature and win”, several important cognitive architectures emerged for modeling human cognitive processes across a wide range of phenomena. One of these, ACT-R, has played an important role in the study of memory processes in sentence processing. In this talk, I will talk about some important lessons I have learnt over the last 20 years while trying to evaluate ACT-R based computational models of sentence comprehension. In this connection, I will present some new results from a recent set of sentence processing studies on Eastern Armenian.
Carolyn Anderson will be defending her dissertation on 6/25 at 1PM. Her dissertation is entitled ‘Shifting the Perspectival Landscape: Methods for Encoding, Identifying, and Selecting Perspectives.’ In her thesis, Carolyn explores formal, computational, and experimental models of perspective representation and processing.
Please join us, virtually, to hear Carolyn present her thesis work! Her dissertation defense will be hosted over Zoom, and we ask that people register for this meeting in advance at the link below. See you there
The Bridging AI and Cognitive Science workshop at ICLR 2020 is being held virtually on April 26th. UMass’s Carolyn Anderson and Tessa Patapoutian will be presenting a paper titled “Can Neural Network Models Learn Spatial Perspective from Text Alone?” and Max Nelson will be presenting a paper titled “Learning Hierarchical Syntactic Transformations with Encoder-Decoder Networks.” Full papers, and details on virtually attending, can be found at: https://baicsworkshop.github.io/
Tal Linzen, Johns Hopkins University, will present “What inductive biases enable human-like syntactic generalization?” in the Linguistics zolloquium series at 3:30 Friday April 17. An abstract follows. All are welcome! The Zoom link has already beensent out on department mailing lists. If you did not receive it and would like attend, please email Brian Dillon for the link.
Humans apply their knowledge of syntax in a systematic way to constructions that are rare or absent in their linguistic input. This observation, traditionally discussed under the banner of the poverty of the stimulus, has motivated the assumption that humans are innately endowed with inductive biases that make crucial reference to syntactic structure. The recent applied success of deep learning systems that are not designed on the basis of such biases may appear to call this assumption into question; in practice, however, such engineering success speaks to this question in an indirect way at best, as engineering benchmarks do not test whether the system in fact generalizes as humans do. In this talk, I will use established psycholinguistic paradigms to examine the syntactic generalization capabilities of contemporary neural network architectures. Focusing on the classic cases of English subject-verb agreement and auxiliary fronting in English question formation, I will demonstrate how neural networks with and without explicit syntactic structure can be used to test for the necessity and sufficiency of structural inductive biases, and will present experiments indicating that human-like generalization requires stronger inductive biases than those expressed in standard neural network architectures.