Sumner, M., & Samuel, A. G. (2005). Perception and representation of regular variation: The case of final /t/. Journal of Memory and Language, 52, 322–328.
Presentation: Sarah
Summary: Shelly
One important topic in speech perception is how listeners process the considerable variation in spoken words. Some previous studies suggested that listeners view these variations as products of the interaction between the canonical forms and the context, therefore they can figure out the correct phones by assuming that the variation has underwent some assimilation rules under the influence of its environment. This is called the one abstract representation model, which suggests that listeners need to decode the surface form into canonical representation during perception. However, evidence also showed that not all variation resulted from coarticulation or induced by explicit context, but still, listeners process them with no difficulties. This phenomenon implied that the information contained in the variation is enough, and listeners do not need an abstract canonical form when they process. This is called the fine-grained representation model. In order to figure out which model is at work in speech perception, the present study conducted three experiments, using the semantic priming paradigm. In Experiment 1, they varied the forms of /t/ in the stimuli, which were [t] (canonical), [ʔt ̚] (coarticulation), and [ʔ] (glottal), and then they went on to see how well these variants exhibit semantic priming effect for listeners. Results showed that all the three variants can equally activate semantic priming, which suggested that listeners need not resort to canonical forms during process. However, in order to make sure that the fine-grained representation model is at work, two more experiments were conducted. In Experiment 2, listeners also did the semantic priming test for two blocks, but in Block 1, they would only hear one of the variants, either [t], [ʔt ̚], or [ʔ], and in Block 2, they would encounter all the three variants. If, in Block 2, the priming effect is only better when subject hears the same variant as in Block 1, then this implies that what is being stored by listeners is the details of each variant, not the canonical form. If, in Block 2, all the variants are equally well in activating priming effect, then there should be the canonical form involved in the process. Results of Experiment 2 showed that the priming effect in Block 2 is only better in the basic [t] for those listeners who also heard the basic [t] in Block 1. This result does not support fine-grained representation model, because the identity priming is not found. However, it does not fully support one abstract representation model either, because more priming effect should be found in Block 2 if the canonical form has been activated in Block 1. Experiment 3 tackled this question in a more direct way. The same material as in Experiment 2 was used, and the subjects were asked in Block 2 to judge whether they have heard the word in Block 1. Results showed that only those who heard [t] in Block 1 had the highest accuracy when they also heard [t] in Block 2. The similar results in Experiment 2 and Experiment 3 suggested in the long term, not all variants are effective in accessing representation. The basic form is better in this regard. To conclude from the above results, it is inconclusive in which model is at work in speech perception, but it seems that listeners can extract relevant info during immediate recognition, without much influence of the stored representation.
