The key tasks outlined in the NSF grant proposal have been completed and have been published as we detail in what follows. The numbers refer to the section numbers in the original proposal.

5.1 Task 1: Pure mediated semantic priming: Our proposed model, the Hierarchical Holographic Model (HHM), is unable toreplicate the pure mediated priming effect documented by Jones (2010). We did not pursue this task further. Instead, we pivoted to syntactic priming, which HHM can capture,though it is outperformed by transformer-based neural language models. We document ourfindings in Kelly et al. (2020).

5.2 Task 2: Ordering words to form grammatical sentences:HHMs successfully performed the word ordering task. We document our findings inKelly et al. (2017) and in Kelly et al. (2019).

5.3 Task 3: Judgements of acceptability: HHM is able to do the judgement of acceptability task, but not nearly as well asconventional neural language models. It also appears that much of the variability injudgements of acceptability is accounted for by spelling errors, rather than the syntacticrelationships HHM is designed to capture. We have a paper on our findings set in Wang et al. (2020).

5.4 Task 4: Part of speech tagging and “super-tagging”:HHM is able to perform tagging or "super-tagging" of part of speech, as documentedin in Kelly et al. (2017) and in our upcoming paper Kelly et al. (2019).

6.2 The semantic and syntactic relationship continuum:Additionally, we test HHM on Chomsky (1956)’s classic nonsense sentence “Colorless green ideas sleep furiously” and find that HHM is able to detect that “Colorless green ideas sleep furiously” is more grammatical than the inverted alternative “Furiously sleep ideas green colorless.” This work is documented in our paper Kelly (2019),currently under review.

To extend Kelly et al. (2020) will require additional syntactic priming data from human participants. Ph.D. candidate Zixin Tang explored an extension of the proposed work. HHM is a computational model of high-level, abstract, syntactic representations. Second-language speakers tend to apply first-language syntactic structures to their second-language,suggesting that in the mind of the speaker, there are syntactic representations that areshared between the first and second language. Zixin’s work investigates what syntactic representations are shared between the first and second language in the mind of a second language speaker and how these sharedrepresentations can be modelled. HHM is one modelling approach that could be used tocapture the shared representations. Other approaches include formalisms from theoretical linguistics, such as minimalist grammar, harmonic grammar, or combinatory categorical grammar (Steedman & Baldridge, 2011). Another approach would be to apply a deep neural language model, such as BERT (Devlin et al., 2019). Zixin’s research is in an early,exploratory stage, and the precise formalism, theory, or modelling technique to be used has not yet been settled.

References

Chomsky, N. (1956). Three models for the description of language. IRE Transactions onInformation Theory, 2, 113–124. https://doi.org/10.1109/TIT.1956.1056813

Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deepbidirectional transformers for language understanding, In Proceedings of the 2019Conference of the North American Chapter of the Association for ComputationalLinguistics: Human Language Technologies, Minneapolis, Minnesota, Association forComputational Linguistics. https://doi.org/10.18653/v1/N19-1423

Jones, L. L. (2010). Pure mediated priming: A retrospective semantic matching model.Journal of Experimental Psychology: Learning, Memory, and Cognition, 36 (1), 135.

Kelly, M. A., Reitter, D., & West, R. L. (2017). Degrees of separation in semantic andsyntactic relationships (M. K. van Vugt, A. P. Banks, & W. G. Kennedy, Eds.). In M. K. van Vugt, A. P. Banks, & W. G. Kennedy (Eds.), Proceedings of the 15thInternational Conference on Cognitive Modeling. Warwick, U.K., University ofWarwick.https://iccm-conference.neocities.org/2017/ICCMprogram_files/paper_42.pdf

Kelly, M. A., Reitter, D., West, R., & Ghafurian, M. (2019). Indirect associations inlearning semantic and syntactic lexical relationships. PsyArXiv.https://doi.org/10.31234/osf.io/ytnjp

Kelly, M. A., Xu, Y., Calvillo, J., & Reitter, D. (2020). Predicting syntactic priming fromsentence embedding vectors (S. Denison, M. Mack, Y. Xu, & B. C. Armstrong,Eds.). In S. Denison, M. Mack, Y. Xu, & B. C. Armstrong (Eds.), Proceedings of the42nd annual conference of the cognitive science society, Austin, TX, CognitiveScience Society. https://clcs.sdsu.edu/pubs/cogsci2020_whichsentence_0529.pdf

Steedman, M., & Baldridge, J. (2011). Combinatory categorial grammar (R. Borsley &K. Borjars, Eds.). In R. Borsley & K. Borjars (Eds.), Non-transformational syntax:Formal and explicit models of grammar. Wiley-Blackwell.

Wang, J., Kelly, M. A., & Reitter, D. (2020). Do we need neural models to explain humanjudgments of acceptability? (S. Denison, M. Mack, Y. Xu, & B. C. Armstrong,Eds.). In S. Denison, M. Mack, Y. Xu, & B. C. Armstrong (Eds.), Proceedings of the42nd annual conference of the cognitive science society, Austin, TX, CognitiveScience Society. https://arxiv.org/abs/1909.08663

Last Modified: 05/02/2023
Modified by: Prasenjit Mitra