Current vector space: default








Load from a file
You can explore the vision space by looking up the nearest neighbors (i.e., most similar representations) for prototypes and individual images -- see also the Help page and Usage info. For example, type in

CASTLE_PRO
LION_007
ASTRONAUT_025Q

and press Calculate.

Note:

  • numbers "_001" up to "_015" and "_030" at the end of a word denotes an individual image with that word label (for example, LION_001 is a picture that was labelled as LION). You can use the Picture picker on the right-hand to search for a label (e.g., LION), which will show all individual images.

  • "_PRO" at the end of a word (for example, LION_PRO) denotes a prototypical vector derived from an average of all images from that category. Note that, since this is an average, there is no individual image corresponding to LION_PRO.

  • codes ending in a number followed by "Q" denote images with varying degrees of similarity to their prototype; "_100Q" is the closest image to the prototype (for example, LION_100Q is the closest image to LION_PRO), while "_000Q" is the most distant one. Again, use the Picture picker to search for a label (such as LION) to display these images.
The ViSpa system was empirically evaluated in a series of large-scale studies, described in

Günther, F., Marelli, M., Tureski, S, & Petilli, M. (2021). ViSpa (Vision Spaces): A computer-vision-based representation system for individual images and concept prototypes, with large-scale evaluation. psyArXiV preprint. https://doi.org/10.31234/osf.io/n4dmq

This website was created as part of the research project described in this paper. We kindly ask you to cite this paper when you are using this website for your own work; this will help us a lot!

The website was modelled after the snaut website, using the code provided by Paweł Mandera at https://github.com/pmandera/snaut For more information on the snaut website, check the associated paper:

Mandera, P., Keuleers, E., & Brysbaert, M. (2017). Explaining human performance in psycholinguistic tasks with models of semantic similarity based on prediction and counting: A review and empirical validation. Journal of Memory and Language, 92, 57-78. https://doi.org/10.1016/j.jml.2016.04.001
Note: To compute typicality scores, just calculate the simliarity between an image and its own prototype vector (e.g., between LION_001 and LION_PRO).

Load from a file
Load from a file
Note: To compute typicality scores, just calculate the simliarity between an image and its own prototype vector (e.g., between LION_001 and LION_PRO).

Load from a file
This is an interface that allows to do semantic arithmetic. For example, king - man + woman = queen. You can combine arbitrary number of positive and negative vectors.
In order to try this example, type 'king, woman' to the upper box (positive vectors) and 'man' (negative vector) to the lower one.

For more information see:

Mikolov, T., Yih, W., & Zweig, G. (2013). Linguistic Regularities in Continuous Space Word Representations. In HLT-NAACL (pp. 746–751).