The harmonic mind : from neural computation to optimality-theoretic grammar

By Joe Pater. 2009

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Yoshua Bengio
2. Li Deng
3. Jianfeng Gao
4. Dong Yu
5. Zhe Gan
6. Kyunghyun Cho
7. Mohit Bansal
8. Xiaodong He

Citations in the corpus

• Deep Learning: Methods and Applications (2014, background): “, they can potentially perform powerful reasoning over structures, as argued in various historical publications as well as recent essays [38, 156, 286, 288, 292, 336, 335].”
• Distributional Memory: A General Framework for Corpus-Based Semantics (2010, neutral): “…being explored in corpus-based semantics, and interesting synergies are emerging with research areas such as neural systems (Smolensky 1990; Smolensky and Legendre 2006), quantum information (Widdows and Peters 2003; Aerts and Czachor 2004; Widdows 2004; Van Rijsbergen 2004; Bruza and Cole…”
• Mathematical Foundations for a Compositional Distributional Model of Meaning (2010, methodology): “In what follows we use the steps above to assign meaning to positive and negative transitive sentences5.”
• How Limited Systematicity Emerges: A Computational Cognitive Neuroscience Approach (Author's Manuscript) (2014, methodology): “    Other Neural Network Approaches to Systematicity     There have been a number of different approaches to introducing systematicity into neural network models over the years (Bodén & Niklasson, 2000; Chalmers, 1990; Christiansen & Chater, 1994; Hadley, 1997; Hadley & Hayward, 1997; Niklasson ”
• Weighted Constraints in Generative Linguistics (2009, methodology): “Thus, an even stronger bridge between GL and other research in cognitive science might be built by using weighted constraint grammars for the generative analysis of linguistic systems, as in OT’s predecessor, Harmonic Grammar (HG; Legendre, Miyata, & Smolensky, 1990; Smolensky & Legendre, 2006; see ”
• Question-Answering with Grammatically-Interpretable Representations (2017, methodology): “The proposed capacity for distributed representation of structure is provided by Tensor Product Representations, TPRs, in which a discrete symbol structure is encoded as a vector systematically built—through vector addition and the tensor product—from vectors encoding symbols and vectors encoding th”
• Weighted constraints and gradient restrictions on place co-occurrence in Muna and Arabic (2008, unknown): “(2006) claim that HG’s cumulativity can produce implausible linguistic systems, the only published example of an unattested system produced by weighting but not ranking involves a gradient Alignment constraint (Legendre et al. 2006), which is controversial even in OT (see esp.”
• A Maximum Entropy Model of Phonotactics and Phonotactic Learning (2008, background): “…is the number of times that x violates the ith constraint, and N i=1 denotes summation over all constraints (C1, C2, . . . , CN).2 2 Our ‘‘scores’’ are closely related to the harmony values explored in Smolensky 1986 and subsequent work (Smolensky and Legendre 2006); hence the abbreviation h(x).”
• A Survey of Artificial Cognitive Systems: Implications for the Autonomous Development of Mental Capabilities in Computational Agents (2007, methodology): “…to have a strong influence on cognitive science, either in a strictly PDP sense such as McClelland's and Rogers' PDP approach to semantic cognition [83]) or in the guise of hybrid systems such as Smolensky's and Legendre's connectionist/symbolic computational architecture for cognition [84], [85].”
• Word Order and Scope: Transparent Interfaces and the ¾ Signature (2012, result): “15 Our reasoning here is in some ways parallel to the OT logic that ‘‘bans only the worst of the worst’’ (BOWOW; see Prince and Smolensky 2004, Smolensky and Legendre 2006).”
• Last Words: Computational Linguistics and Deep Learning (2015, unknown): “It led to a vast amount of work by Paul Smolensky on how basically categorical systems can emerge and be represented in a neural substrate (Smolensky and Legendre 2006).”
• Category-theoretic quantitative compositional distributional models of natural language semantics (2013, methodology): “An early proposal by Smolensky [74, 75] to use linear algebraic tensors as a composition operation solves the problem of finding non-commutative vector composition operators.”
• Conspiring to Mean: Experimental and Computational Evidence for a Usage-Based Harmonic Approach to Morphophonology (2013, methodology): “Drawing on ideas from machine learning (Daelemans & van den Bosch 2005) and harmonic grammar (Legendre et al. 1990, Smolensky & Legendre 2006), I develop a formally explicit model showing how morphophonological constructions can be extracted from the lexicon and how the competition between such…”
• Gradual Learning and Convergence (2008, unknown): “…the optimal Output mapping from a set of candidates sharing the same Input (as in some versions of HG; see Legendre et al. 2006 and other papers in Smolensky and Legendre 2006, as well as Prince and Smolensky 1993/2004, Keller 2000, 2006, Flemming 2001, Prince 2002b, Pater, Bhatt and Potts…”
• Combining Symbolic and Distributional Models of Meaning (2007, background): “(Smolensky & Legendre 2006) integrate the connectionist and symbolic models in the following way.”