Skip to content

Brief Literature Review

The Doom Lab edited this page Jun 25, 2020 · 2 revisions

Welcome to the word2manylanguages wiki!

  1. Distributional Models of Memory

    a. General overviews of these models:

    1. https://www.researchgate.net/profile/Jon_Willits2/publication/264825283_Models_of_semantic_memory/links/53f299cd0cf2bc0c40ecadc0.pdf
    2. https://ir.lib.uwo.ca/cgi/viewcontent.cgi?article=1124&context=psychologypub

    b. The main paper I am basing a lot of this idea on: http://crr.ugent.be/papers/Mandera_et_al_JML_2016.pdf Word2vec

    1. https://arxiv.org/pdf/1301.3781.pdf
    2. https://arxiv.org/pdf/1402.3722.pdf
  2. What you might use as corpora for these models: The Subtitle Projects

    a. http://crr.ugent.be/programs-data/subtitle-frequencies

    b. https://link.springer.com/content/pdf/10.3758%252FBRM.41.4.977.pdf

    c. https://www.frontiersin.org/articles/10.3389/fpsyg.2011.00027/full

    d. https://biblio.ugent.be/publication/599589/file/738788.pdf

    e. There are bunch of languages, but this paper brings it together: https://psyarxiv.com/fcrmy/

    f. Python package: https://github.com/jvparidon/subs2vec

  3. What you might use as dependent variables to assess if these models "work"

    a. The Lexicon Projects - in several languages, but get started with English: https://link.springer.com/content/pdf/10.3758/BF03193014.pdf

    b. The Semantic Priming Project: https://www.aggieerin.com/pubs/hutchison%2013.pdf

    c. The Small World of Words: https://smallworldofwords.org/articles/2018.DeDeyne.The%20Small%20World%20of%20Words%20English%20word%20association%20norms%20for%20over%2012,000%20cue%20words.BRM.pdf

Goal/hypothesis/contribution to the literature:

  1. How comparable are models of semantic spaces built on different languages, given a similar type of corpora input?

  2. Is there a general rule about the dimensionality or window size that we might consider "universal"?

  3. Are there common pairs that occur with the most frequent words that we might consider a "universal set" for studying similarity?

Part 2: Can we make these models better by combining subtitles and simple wikipedia: https://dumps.wikimedia.org/simplewiki/20200601/

Clone this wiki locally