Modeling the learning signals that help children find structure in speech
Acquiring our first language from environmental evidence is one of the most difficult
learning problems we ever face, and we face it in early childhood, when we are most
cognitively limited. The learning signals that children use to extract linguistic
regularities from speech have been the subject of intensive theorizing, but they
are challenging to study experimentally.
One of my core lines of research complements experimental work in this domain by
simulating the process of learning from speech using unsupervised deep neural
networks. These computational simulations allow both (1) fine-grained interventions
on the network architecture and learning objectives and (2) detailed inspection
of acquired representations, enabling us to test causal links between the structure
of the learner and the result of the learning process.
Related publications
CoNLL
Best Paper Award
Acquiring language from speech by learning to remember and predict
@inproceedings{shainelsner20,author={Shain, Cory and Elsner, Micha},title={Acquiring language from speech by learning to remember and predict},booktitle={Proceedings of the 24th Conference on Computational Natural Language Learning},year={2020},pages={195--214},doi={10.18653/v1/2020.conll-1.15},pdf={shain_elsner_20.pdf},slides={shain_elsner_20_slides.pdf},code={https://github.com/coryshain/dnnseg},recording={https://slideslive.com/38939477/acquiring-language-from-speech-by-learning-to-remember-and-predict?fbclid=IwAR3AtIO57f7EyfTt2ObQycp0bu5EPnP2l99OBEydXIAZXLaTzMEIvfJJ8Lk},best_paper={true},abbr={CoNLL},section={papers},selected={true},project={phonacq},bibtex_show={true}}
NAACL
Measuring the perceptual availability of phonological features during language acquisition using unsupervised binary stochastic autoencoders
In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)
2019
@inproceedings{shainelsner19,author={Shain, Cory and Elsner, Micha},title={Measuring the perceptual availability of phonological features during language acquisition using unsupervised binary stochastic autoencoders},booktitle={Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)},year={2019},pages={69--85},pdf={shain_elsner_19.pdf},slides={shain_elsner_19_slides.pdf},code={https://github.com/coryshain/dnnseg},recording={https://vimeo.com/353440477},abbr={NAACL},section={papers},project={phonacq},bibtex_show={true}}
EMNLP
Speech segmentation with a neural encoder model of working memory
@inproceedings{elsnershain17,title={Speech segmentation with a neural encoder model of working memory},author={Elsner, Micha and Shain, Cory},booktitle={Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing},year={2017},pdf={elsner_shain_17.pdf},code={https://github.com/melsner/neural-segmentation},abbr={EMNLP},section={papers},project={phonacq},bibtex_show={true}}