Similarity & Relatedness presentations by Marten Postma, Minh Le, Alessandro Lopopolo, and Emiel van Miltenburg.
July 01, 2015 at VU Amsterdam
Similarity and relatedness lie at the core of semantics. At CLTL, we study similarity and relatedness from different perspectives: linguistic perspective using WordNet hierarchy, mathematic perspective about evaluation, neurological perspective using brain wave recoding, and perceptual perspective using distributional and multimodal approach.
— Similarity — Is-a relation
Presentation by: Marten Postma, MA.
In this talk, WordNet-based similarity and relatedness measures are discussed. Originally, most measures only use the graph structure of WordNet, whereas later approaches also make use of reference corpora and sense definitions. Finally, we made an effort to define similarity and relatedness within WordNet. Similarity was defined as a subset of relatedness. Similarity measures only make use of the “is-a” relation, whereas relatedness measures make use of all semantic relations as well as the definitions.
View/download presentation by Marten Postma, MA:
Similarity & Relatedness: Similarity — Is-a relation (pdf)
— Similarity measure Evaluation
Presentation by: Minh Ngoc Lê, MSc.
Popular datasets such as SimLex-999, WordSim, RG-65 and evaluation schemes such as Spearman’s rho are convenient and intuitive. But are they reliable? It turns out that when the math is done, they start to collapse…
View/download presentation by Minh Ngoc Lê, MSc, Dr. Antske Fokkens:
Similarity & Relatedness: Similarity measure Evaluation (pdf)
— Similarity vs. Association in neural oscillatory data: expectations, hints and applications
Presentation by: Alessandro Lopopolo, MA.
I present a series of neuro-imaging experiments aimed to investigate the scientific foundations of the distinction between similarity and relatedness (or association). Preliminary results seem to suggest a link between different frequency bands of brain activity and how the two semantic dimensions are encoded and processed in the human brain. Moreover, I discuss the implications of these observations for neuro-ingeneering and for the development of semantic-based BCI’s for communication.
View/download presentation by Alessandro Lopopolo:
Similarity & Relatedness: Similarity vs. Association (pdf)
— Distributional similarity and how to evaluate your models
Presentation by: Emiel van Miltenburg, MA.
This talk provides an update of my work so far building data sets to evaluate distributional models, both text-based and image-based. On the text-based side I talk about extending current work (focusing on English) to Dutch. On the image-based side, I talk about a new data set of hybrid animals; one-off images of animals that are composed of (usually two) different animals. These images may be used for a more robust test of image understanding by machines.
View/download presentation by Emiel van Miltenburg, MA:
Similarity & Relatedness: Distributional similarity (pdf)