Change search
ReferencesLink to record
Permanent link

Direct link
A Potential Surface Underlying Meaning?
University of Borås, Faculty of Librarianship, Information, Education and IT. (DIGRES)
University of Borås, Faculty of Librarianship, Information, Education and IT.
2015 (English)Conference paper, Abstract (Other academic)
Abstract [en]

Machine learning algorithms utilizing gradient descent to identify concepts or more general learnables hint at a so-far ignored possibility, namely that local and global minima represent any vocabulary as a landscape against which evaluation of the results can take place. A simple example to illustrate this idea would be a potential surface underlying gravitation. However, to construct a gravitation-based representation of, e.g., word meaning, only the distance between localized items is a given in the vector space, whereas the equivalents of mass or charge are unknown in semantics. Clearly, the working hypothesis that physical fields could be a useful metaphor to study word and sentence meaning is an option but our current representations are incomplete in this respect.For a starter, consider that an RBF kernel has the capacity to generate a potential surface and hence create the impression of gravity, providing one with distance-based decay of interaction strength, plus a scalar scaling factor for the interaction, but of course no term masses. We are working on an experiment design to change that. Therefore, with certain mechanisms in neural networks that could host such quasi-physical fields, a novel approach to the modeling of mind content seems plausible, subject to scrutiny.Work in progress in another direction of the same idea indicates that by using certain algorithms, already emerged vs. still emerging content is clearly distinguishable, in line with Aristotle’s Metaphysics. The implications are that a model completed by “term mass” or “term charge” would enable the computation of the specific work equivalent of sentences or documents, and that via replacing semantics by other modalities, vector fields of more general symbolic content could exist as well. Also, the perceived hypersurface generated by the dynamics of language use may be a step toward more advanced models, for example addressing the Hamiltonian of expanding semantic systems, or the relationship between reaction paths in quantum chemistry vs. sentence construction by gradient descent.

Place, publisher, year, edition, pages
Keyword [en]
word semantics, meaning, field theory, force, potential, gravity, gradient descent, machine learning
National Category
Language Technology (Computational Linguistics) Other Physics Topics Other Computer and Information Science
Research subject
Library and Information Science
URN: urn:nbn:se:hb:diva-8533OAI: diva2:894327
Yandex School of Data Analysis Conference, Machine Learning: Prospects and Applications
EU, FP7, Seventh Framework Programme, FP7-601138
Available from: 2016-01-14 Created: 2016-01-14 Last updated: 2016-03-04Bibliographically approved

Open Access in DiVA

fulltext(1129 kB)41 downloads
File information
File name FULLTEXT01.pdfFile size 1129 kBChecksum SHA-512
Type fulltextMimetype application/pdf

Other links

Search in DiVA

By author/editor
Darányi, SándorWittek, Peter
By organisation
Faculty of Librarianship, Information, Education and IT
Language Technology (Computational Linguistics)Other Physics TopicsOther Computer and Information Science

Search outside of DiVA

GoogleGoogle Scholar
Total: 41 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Total: 280 hits
ReferencesLink to record
Permanent link

Direct link