CC BY-NC-ND
0Available online at www.sciencedirect.com
Procedía
Social and Behavioral Sciences
ELSEVIER
Procedia - Social and Behavioral Sciences 61 (2012) 9 - 10
Academy of Aphasia 2012
Case series studies frequently evaluate a theory using quantitative measures of fit between simulations of the theory and the observed data. In this work, simulations of speech production deficits reveal that a common measure of fit fails to distinguish competing theories. To be theoretically productive, case series research must focus on data that distinguish alternative accounts.
Speech production theories assume feed-forward activation flow from semantic features ("feline") to lexical representations () followed by phonological representations (/k/). Theories differ in the extent of feedback activation flow. In the Two-Step Interactive Account (TSIA; Schwartz, Dell, Martin, Gahl, & Sobel, 2006) activation flows bidirectionally, with equal strength, between each set of representations. In contrast, the Restricted Interaction Account (RIA; Rapp & Goldrick, 2000) incorporates no feedback from lexical to semantic representations; phonological to lexical feedback is significantly weaker than feedforward connections.
Previous case series research has assessed TSIA using the fit of a simulation to each individual's overall distribution of responses (proportion correct, semantic error, etc). Does this measure of fit successfully indicate whether TSIA is the correct theory? Rather than utilize real data—where the true theory is unknown—artificial case series were generated by TSIA and RIA simulations. If this measure of fit indicates whether TSIA is correct, the fit of the baseline TSIA simulation should be better when a case series was generated by a TSIA vs. a RIA simulation.
Case series were first generated by damaging simulations with identical structure to the baseline TSIA simulation. The simulations used two 6 word neighborhoods to generate responses. Processing was disrupted by randomly lowering the strength of activation flow between representations. The results show that measures of fit to overall response distributions did not indicate whether TSIA was correct vs. incorrect. The baseline TSIA simulation exhibited an excellent fit to both case series. The average root mean square deviation (RMSD; lower indicates better fit) to the TSIA case series was 0.01; RIA, 0.017.
Above, the baseline TSIA simulation and the sources of the artificial case series all used the same small lexicon to generate responses. In contrast, in previous empirical studies of TSIA, the case series data were generated by English speakers with much larger lexicons. To examine the effect of this type of approximation, case series were generated using a 29 word lexicon. Using two 6 word neighborhoods, the baseline TSIA simulation actually exhibited a better fit for case series generated by RIA vs. TSIA simulations (mean RMSD TSIA: 0.019; RIA: 0.016).
These result show that measures of fit to overall response distributions do not successfully discriminate theories of speech production. Methods for discovering measures that can discriminate theories (Rapp & Goldrick, 2000) will be discussed.
References
* Corresponding author.
E-mail address: matt-goldrick@northwestern.edu .
© 2012 Published by Elsevier Ltd. Selection and/Tor peer-reviewunder responsibility of The Academy ofAphasia
1877-0428 © 2012 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of The Academy of Aphasia doi: 10.1016/j.sbspro.2012.10.054
Goldrick M. /Procedia - Social and Behavioral Sciences 61 (2012) 9 - 10
Rapp, B. C., & Goldrick, M. (2000). Discreteness and interactivity in spoken word production. Psychological Review, 107,460499.
Schwartz, M. F., Dell, G. S., Martin, N., Gahl, S., & Sobel, P. (2006). A case series test of the interactive two-step model of lexical access: Evidence from picture naming. Journal ofMemory and Language, 54 ,228-264.
