Below I will be investigating the psychological concept of similarity. This subject relates to NL Nature because a new section of this project (yet to be released) will rely on species features for identification.

To understand the relevance, imagine we have observed a bird, but don’t know what species it belongs to. It reflects an underlying principle of cognition. We first learn about the phenomenon and then establish links between that phenomenon and the rest of our knowledge base. The links we create can subsequently evolve into relationships and classifications. We know we observed a bird, and without knowing what species the bird is, we cannot use traditional NL Nature interface (figure 1) without considering what features of the unknown bird are common to the features of the birds on the NL Nature’s dropdown list. Even for a professional biologist it is not Linnaean taxonomy, but the attribute similarity that will allow classifying a black and white bird with long bill and pointed wings as a gannet.

Sjöberg, L. (1969). The dimensionality of similarity judgments. The American Journal of Psychology, 82(4), 441-456.

"the basic problem is whether it is at all possible to measure the similarity...for one individual one one occasion" (p. 441)

"the possibility of measuring similarity has seldom been questioned" (p. 441)

------------------

This is an exploratory article on dimensionality of similarity judgements. With reservation the author assumes “it is possible to measure similarity” (p. 441). Acknowledging the prior research, Sjöberg notes that most of it concentrated on similarities between single objects (p. 444).  The author investigates relationships among scaling methods, with the objective of testing similarity on paired objects. The article presents classification fo similarity diving it into three groups: quantitative ("intrasitivity of paired comparisons" - p. 444), qualitative ("second order similarity spaces" p. 444), and mixture of both. The method for hypothesizing similarity categories involved pictures with faces showing three emotions - happiness, depression and anger, and measured paired comparison, category judgement and magnitude estimation.  As the result of the test, "the three expected kinds of similarities did not emerge as separate factors" (p. 452). Instead, the test suggested a "clustering of similarities" (p. 456) based on variations in content. The test supported both uni- and multidimensional scaling of similarity. The author also warned that the test results were not entirely reliable and called for futher investigation of the phenomena. 

Cutujian, F. C. (1942). The rôle of identical stimulus-elements in judgments of similarity. The American Journal of Psychology, 55(1), 18-32.

---------------

A foundation article that attempts to define the concept of silimarity. The author departs from the traditional definition of similarity "as partial identity" (p. 18), and investigates the theory proposed by Karl Duncker who believed that "similarity is not nessesarily due to partial identity" (p. 18). Duncker suggested two types of similarity "similarity between sums and similarity between gestalten" (p. 18). The sum similarity is based on "proportional ... number of identical elements," while gestalten arises from "functional structures" (p. 18). Cutujian develops an experiment to test Duncker's theory. The experiment uses four groups of proverbs: one with complete identity, second with complete verbal identity, but different order of words, third with partial verbal identity, and fourth with non verbal identity (meaning similarity). As th eresult of the experiment, Cutujian observes the two types of silimarities - "similarity of sums" and "similarity of menaings" and concludes that the choice between them is congitevely deterimed by the element identity of the stimuli (p. 29). Article also notes that when using similarity of meanings, the element identity does not play significant role in judgement (p. 29). Supporting Duncker's findings, the article redefines similarity as similarity of meanings. It is similarity of meanings between two fields of meaning that creates "the process of generalization in" subject's mind which results in "an awareness of similarity" (p. 30). Similarity of sums, on the other hand, does not trigger generalization process and evokes a "sense of identity" (p. 30). Article belives that similarity of sums, which is based on process of generalization and transfer better reflects similarity concept.

Rumelhart, D. E., & Abrahamson, A. A. (1973). A model for analogical reasoning. Cognitive Psychology, 5(1), 1-28.

Similar to Cutujian (1942), Rumelhart and Abrahamson viewed similarity judgement, a basic form of reasoning, as a process that relied on structure, rather than content (1973, p. 1).

essintially argues that similarity is attribute matching!

-----------------------------------

The article attempts to fill the theoretical gap in the process of analogy formation by presenting a model of analogical reasoning. Similar to Cutujian (1942), Rumelhart and Abrahamson views similarity judgement, a basic form of reasoning, as a process that relies on structure, rather than content (1973, p. 1). While content comparison relies on remembering specific information, similarity reasoning is a "process of information retrieval" which is "derived from the memory structure" (p. 2). Similarity concept is a spatial one. The closer a concept to another in memory, the more similar it is. The authors adopt Euclidean multidimensional space model that was developed by prior research. This model representing psychology space is easy to measure (vector distance being the main variable) and visualize. Using "parallelogram rule" (p. 11, p. 28), the authors experimentally observe that a pair of concepts are related to another pair though the proximity of their respective attributes. The article concludes that while the model does not fully describe the process of analogy formation, it is a foundation for future research.

Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124-1131.

An article that presents psychological concept of judgement from the point of view of statistical validity, discusses biases in judgement under uncertainty. The main argument of the article is that people place too much focus on the perceived distance between the two concepts, and their perceived similarity, and fail to consider other known factors in judgement of even probabilities (p. 1124-1125). Authors present several common “systematic errors” in assessing probabilities: representativeness, availability of instances, and adjustment from an anchor. Representativeness is a concept relevant to similarity because it reflects the probably of one concept to belong to another, broader concept. Representativeness has a number of judgement biases.  “Insensitivity to prior probabilities of outcome” describes an error when prior given probabilities are ignored. A description that is stereotypical of a lawyer will most likely be labelled as the description of a lawyer, even when responded is aware that the number of lawyers is very small in the given sample (p. 1125). “Insensitivity to sample size” reveals the failure to consider sample size in predicting outcomes. It is based on the mistaken belief that “chance is … a self-correcting process in which a deviation in one direction induces a deviation in the opposite direction to restore equilibrium” (p. 1125). In reality deviations are diluted through large sample size. “Insensitivity to predictability” demonstrates tendency to see future similar to the present situation, even though the later may present few clues to the future events (p 1126). People select outcomes that are more similar to the inputs. The more consistent the inputs, the more confident people are of the outcomes. Yet, statistically, “redundancy among inputs decreases accuracy” (p. 1126). The author concludes that while the judgement biases contribute to cognitive economy, they “lead to systematic and predictable errors” (p. 1131).

Imai, S. (1977). Pattern similarity and cognitive transformations. Acta Psychologica, 41(6), 433-447.

The author attempts to create a transformation model for pattern similarity. Similarity is one of the “most basic judgements”, which is indirectly related to generalization, and is useful in understanding pattern cognition (p. 434). Using the stimuli of two sets of black and white beads, the following four transformations are detected: mirror-image, phase, reversal, and wave length (p. 435). Based on the experiment, several conclusions were drawn. The abstract nature of the beads showed that “meaningful predictions” are possible “without referring to the identity of the features” (p. 445). The may be more transformations than the four proposed that are part of cognition process. Finally, further investigation is required to clarify the role of transformations in pattern similarity judgement.