The Retinas

Retina is the term we use for our word space models. In a word space [SAHLGREN2006], one can compute distances between terms, and these distances can be used to determine the degree of similarity between terms. We currently have two Retinas available with different characteristics. You can get a description of the available retinas by browsing to the /retinas endpoint in our interactive API documentation, or by querying this URL (remember to include the api-key):

http://api.cortical.io/rest/retinas

The output will be similar to this:

[
  {
    "retinaName": "en_synonymous",
    "description": "An english language retina focusing on synonymous similarity.",
    "numberOfTermsInRetina": 502141,
    "numberOfRows": 128,
    "numberOfColumns": 128
  },
  {
    "retinaName": "en_associative",
    "description": "An english language retina balancing synonymous and associative similarity.",
    "numberOfTermsInRetina": 854376,
    "numberOfRows": 128,
    "numberOfColumns": 128
  }
]

As can be seen, the Retinas both contain semantic fingerprints of more than half a million English terms. The difference between the Retinas lies in the way they represent semantic relations. Let’s first distinguish between two types of similarities:

Synonymous similarity defines similarity based on common features of the items under comparison. For example, the similarity between the term pair car:truck or mouse:rat is of this nature. Associative similarity is on the other hand based on common relations or associations. A good example of this would be spider:web or mouse:cat.

The Retinas were not crafted strictly according to such definitions, but the nature of their respective similarity flavor has in the creation process been guided towards these semantic characteristics. The Retina en_synonymous thus operates in the direction of synonymous similarity, whereas the associative Retina is balancing synonymous and associative similarity, meaning that the Retinas do not differ radically.

Let’s compare the list of similar terms for the term self from the two Retinas. The outputs are shown in this table:

en_synonymous

mind, emotions, consciousness, conscious, beings, ego, essence, thoughts,

understand, emotion, desires, thinking, metaphysical

en_associative

consciousness, mind, thoughts, thinking, beings, things, psychological,

moral, truth, desires, conscious, humanity, mental

Without overinterpreting the outputs, one can say that the synonymous Retina suggests terms related to capabilities of being a human (emotions, consciousness, ego, desires, thinking) whereas the associative Retina adds terms that are a product of reflecting over (associations of) these abilities: things, moral, truth, humanity.

References

[SAHLGREN2006]Sahlgren, M. (2006) The Word-Space Model: Using distributional analysis to represent syntagmatic and paradigmatic relations between words in high-dimensional vector spaces. Ph.D. dissertation, Department of Linguistics, Stockholm University.