Description of EE model from Barcroft & Sommers (2014):

“The three circles on the left side of the model (Word Learning) represent the degree to which the formal component of a developing lexical representation is distributed or robust [see robust representation (RR) model of vocabulary learning]. Therefore, what is depicted here is how the more distributed (robust) representations associated with having processed acoustically varied input lead to better cued recall of target words …

… we can observe why the more distributed representations of words learned with more variability are more likely to be retrieved when cued by a picture. The reason is that the more distributed representations are more likely to come into contact with the lines emanating from (or simply being part of) the semantic representation. As can be seen in the model, a word learned with no variability makes contact only once whereas a word learned with moderate variability makes contact three times, and a word learned with high variability makes contact six times. Each contact constitutes a successful meaning-to-form mapping and allows the participant to produce the word form in question. Therefore, as indicated by the larger number of arrows in the output for high variability (six arrows) over moderate (three arrows) and low variability (one arrow) and the larger number of arrows in the output for moderate over no variability, the developing representation of word forms learned in more acoustically varied formats (as more acoustically varied input) is naturally retrieved more often, allowing them to be produced more often as output.

The right side of the model, on the other hand, depicts the effects of phonetically relevant acoustic variability on word identification and the continuing development of the formal component of lexical representations. In contrast to the left side of the model, the right side concerns cognitive procedures that happen long after the word to be identified has been learned. Input is depicted on the bottom and output on the top as it was deemed easier to follow visually in this manner, but it would make no difference if the two were inverted, provided that the arrows always go from input toward output. The three circles represent the canonical representation of the form of any given word. The lines going upward represent instances of the word form in question being produced in spoken input. Instances in which an upward-moving line makes contact with the solid part of the circle represent instances in which a word is accurately perceived and identified.

The degree to which the upward-moving lines are spread apart represents the degree to which the listener is being exposed to acoustically varied spoken input. In the case of no variability, the lines are close together, depicting the lack of variability in the input to which the listener is exposed. In this case, all six lines make contact with the solid circle and therefore are successfully retrieved and identified. In the case of moderate variability, the lines are not as close together, representing more acoustically varied input, and two of the lines do not make contact with the solid circle, leading to less successful retrieval and decreased performance in word identification. Finally, in the case of high variability, the lines are even more spread apart, representing even more acoustically varied input, and three of the lines do not make contact with the solid circle, leading to even less successful retrieval and even worse performance in word identification. In this way the model depicts why incremental increases in phonetically relevant acoustic variability lead to incremental decreases in spoken word identification performance.

In light of the fact that language users are language learners who are responsive to properties of the input, the model also depicts how acoustically varied input for words can impact canonical representations of word forms, even for words learned years and decades previously. The three solid circles represent the canonical form of a word. When listeners are presented with acoustically varied input that does not match this existing canonical form, the canonical form reshapes in response to the noncanonical forms to which the listener (listener-learner) has been exposed. The dotted circles represent this reshaping. As can be seen in the model, assuming that input without acoustic variability is consistent with the existing canonical form, no reshaping will take place. Moderately variable input, on the other hand, can cause the canonical form to reshape to a certain degree, as represented by the single dotted circle around the solid circle. Finally, highly variable input can cause the canonical form to reshape even further, as represented but the two dotted circles.

Of course we are assuming a distributed representation of word form in both the word-learning and word-identification sides of the model, therefore, the size of the circles on the word-identification simply represent that a certain degree of reshaping will take place in the overall distributed representation of the word form in question. The reshaping may be very minor in some cases, but in other cases it may be more substantial, such as in cases when one is exposed to a new variety (dialect) of their native language over an extended period of time and the process of accommodation gradually takes place.

One of the strengths of the model presented here is that it provides a mechanistic account of the effects of phonetically relevant acoustic variability across the lifespan, beginning with how it affects vocabulary learning when we are first exposed to new lexical items and extending to how it affects the manner in which we process speech during word identification. The more distributed lexical representations associated with acoustically varied input during vocabulary learning depict why more instances of target word form can be retrieved when cued by the activation of the conceptual/semantic space associated with the referent of the target word in question. The lack of one-to-one mapping of acoustically varied forms of a known (previously acquired) word and the canonical form of the word depict why acoustically varied input poses costs during word identification. Clearly, future work can help to provide more fine-grain accounts of the processes involved in both word learning and word identification, including a more quantitative account to how canonical word forms respond to different degrees of non-canonical acoustically varied input, but in our estimation, the present model provides a well-founded framework that is consistent with the current body research on the effects of variability on both vocabulary learning and speech processing.”

Citation from: Barcroft, J., & Sommers. M. (2014). A theoretical account of the effects of acoustic variability on word learning and speech processing. In Torrens, V., Escobar, L. (Eds.), The processing of lexicon and morphosyntax (pp. 7-24). Newcastle: Cambridge Scholars Publishing.