Transposed letter effect

Switching the position of adjacent letters in the base word is a close transposition. This type of transposition creates the greatest priming effect. For example, an effective prime for the word computer would be the TL non-word comptuer.

Forming a prime word by switching the position of nonadjacent letters in the base word is a distant transposition. There is significantly less priming effect in a distant transposition than a close transposition, no matter how distant the two letters are from each other.

Switching the position of adjacent words in the base word is a transposed word effect. This transposition creates a great effect. For example, the white cat ran slowly becomes the cat white ran slowly.[1] Participants are shown to take longer and were more error prone to reject the transposed word sentence than control sentence (e.g. the white cat ran helpfully).[2][3][4] This type has been used to provide arguments in favour of parallel of serial processing.[1][2]

There are a number of theories that have been challenged by the effects shown with transposition-letter priming. These theories mainly have to do with how letters are used to process words.

Slot-based coding theory states that each letter in a word is connected to a particular location, or slot, within that word. One of the major theories that predict letter slots is the interactive activation model created by McClelland and Rumelhart (1981).[8] This model assumes that people are letter-position specific when detecting words, so our recognition of words is based on what letters it contains, where the letters are placed within that word and the length of the word itself. Another example is the Bayesian reader model created by Norris (2006) which also assumes that the letters in a word are associated with their specific location.[9] Many experiments (for example, Perea and Lupker, 2003)[10] have shown if transposed-letter priming is used, (for example, priming judge with jugde) a priming effect will be seen, but if the prime is a word where two letters are changed to other letters (for example, priming judge with jupfe), there is no priming effect. According to the two models discussed, jugde would be assumed to be no more similar to judge than jupfe because the letters are not in the correct slots. In another experiment, TL non-words were more likely to be (incorrectly) classified as actual words than are nonsense control words (Andrews, 1996).[11] Andrew’s results are also inconsistent with the many slot-based coding models because even with two of the letters in the incorrect spot, they read the word as if spelled correctly.

The parallel distributed processing model proposed by Seidenberg and McClelland (1989) also uses a portion of words but instead of letters they are a small group of letters in the same order as in the word.[12] For example, the word judge would have these groupings: _ju, jud, udg, dge, ge_. This predicts that if part of two words match there will be some priming, but this model still depends on the position of the letters to some extent, so it is not compatible with results from transposed-letter priming.

There are a number of theories that are supported by the results shown by the transposed-letter effect.

The SERIOL model (sequential encoding regulated by inputs to oscillations within letter units) described by Whitney (2001) explains processing of words as five levels, or nodes: retinal level, feature level, letter level, bigram level and word level.[13] In the bigram level, the letters detected are turned into a number of pairs. For example, the word cart has the bigrams ca, ar, rt, ar, at and ct. The bigrams that more closely represent the location of letters in the words are given more weight. The pairs are then used to form the word. Within this model, letter location is still a factor but is not a defining feature of word processing, so the transposed-letter effect is consistent with this model.

In the SOLAR model (self-organizing lexical acquisition and recognition) described by Davis (1999) each letter is associated with its own level of activation.[14] The first letter in the word has the highest level of activation and so on until the last word has the lowest level of activation. In this model, position does describe the level of activation for that particular letter but because the activation is successive, two letters beside each other would have a similar activation level. The SOLAR model is consistent with the results of the transposed-letter effect priming because with this effect experiments have shown priming when two adjacent letters are switched but not when two letters farther apart in the word are switched.

Transposed-letter priming was used by Christianson, Johnson and Rayner (2005) on compound words to test the role of morphemes in word processing.[15] They switched the letters either within the morphemes (for example, snowball to snowblal) or between morphemes (for example, snowball to snobwall) in the primes and found a greater priming effect within the morphemes than between. This supported the theory that morphemes are used during the processing of compound words because the priming effect was only reduced when the letters were switched over the morpheme boundary and were unable to separate into their separate parts.