The Effects of Phonological Complexity on Past Tense Processing: An FMRI StudyLisa L. Conant, Rutvik Desai, Jeffrey R. Binder

Department of Neurology, Medical College of Wisconsin, Milwaukee, WI

BACKGROUNDBehavioral dissociations between regular and irregular past tense morphology have been frequently

observed. Such dissociations have been taken as evidence of the existence of separate mechanismsunderlying the processing of regular and irregular verbs. In such dual mechanism accounts, thetransformation of a regular verb stem into its past tense form is thought to involve the application of aset of rules whereas irregular past tense forms, which have been learned by rote, are retrieved through alexical-associative memory mechanism [1,2]. However, behavioral dissociations may also be explainedby an alternative, single-mechanism account. In this view, all verbs are processed through a singleintegrated system, but component processes within the network such as phonology and semantics havedistinct representations within the system and may differentially contribute to the processing ofindividual words depending on multiple features of those words [3]. One feature that tends todistinguish regular and irregular verb forms is phonological complexity (PC). Because thetransformation from present to past tense in regular verbs requires the addition of phonemes while this israrely true in irregular forms, the inflection of regular verbs involves greater phonological demands andmay therefore be more vulnerable to disruption with phonological impairment [4]. This theory hasreceived support from a connectionist simulation [3] as well as a behavioral study in which regularityeffects in patients with nonfluent aphasia were no longer evident when regular and irregular past tenseforms were matched on PC [4].

In the current study, fMRI was used to examine activation patterns associated with regular andirregular past tense processing and to investigate the effect of PC on these patterns. We hypothesizedthat areas of greater activation for regular past tense processing would be minimized when regular andirregular past tense forms were matched on PC. Furthermore, we predicted that similar patterns ofactivation associated with PC would be seen when subjects were simply reading rather than generatingpast tense forms.

METHODSSubjects20 healthy, right-handed, English speaking adults (15 women), mean age=27.5, age range=20-47.

Experimental Paradigm• Two tasks were used for the current analyses. A Read Stem Task was also completed but will not be

discussed here.• Past Tense Generation Task

• Regular and irregular verb stems were presented visually.• Subjects generated the past tense of each stem aloud.

• Read Past Tense Task• Regular and irregular past tense forms were presented visually.• Subjects read aloud the past tense form.

• Generate and read tasks were presented in alternating runs. Within each run, regular and irregularitems, along with baseline fixation trials, were interspersed pseudorandomly. No more than threeitems of either past tense type were presented contiguously.

Stimuli and Subsets• For the Generate Irregular (GI)-Generate Regular (GR) contrast, the stimuli consisted of 50

irregular and 50 regular verbs matched on letter length of the stem, number of syllables in the stemand past tense forms, log frequency of the past tense form, and friend-enemy ratio. The past tenseforms of GR items were more phonologically complex than those of the GI items as indicated by ahigher number of phonemes in the GR past tense forms.

• For the GI-Read Irregular Past (RIP) and GR-Read Regular Past (RRP) contrasts, each comparisonconsisted of 50 generate and 40 read items. The generate and read conditions for each past tensetype were matched on number of syllables in the past tense form, log frequency of the past tenseform, and friend-enemy ratio.

• For the RIP-RRP contrast, the stimuli consisted of 40 regular and 40 irregular verbs matched on logfrequency of the past tense form and friend-enemy ratio. They were not matched on PC.

• In order to specifically examine the effects of PC in past tense generation, two subsets containingequal numbers of items and matched on log frequency were used.

• A subset consisting of 40 irregular and 40 regular items matched not only on number of phonemes butalso on the exact CV structure of the past tense form.

• A subset consisting of 40 irregular and 40 regular items mismatched on number of phonemes.• In order to specifically examine the effects of PC in reading of past tense forms, two subsets

containing equal numbers of items and matched on log frequency were used.• A subset consisting of 25 irregular and 25 regular past tense items matched on number of phonemes.• A subset of 25 irregular and 25 regular past tense items mismatched on number of phonemes.

Image AcquisitionFunctional Data Anatomical Data1.5T GE Signa scanner, T2*-weighted GE-EPI 3D spoiled-gradient-echo-sequence21 contiguous sagittal slices 124 contiguous sagittal slicesTE=40ms, voxels=3.75 x 3.75 x 6.5mm T1-weighted imagesClustered Acquisition, TR=7s, Acquisition Time=2300ms

Image Analysis• AFNI software package• Within-subject multiple linear regression with reference functions for each condition• Incorrect trials removed from the analyses• Individual SPMs resampled in standard stereotaxic space and smoothed with a 5mm

FWHM Gaussian kernel• Random effects analysis• Threshold at uncorrected p<0.001, corrected mapwise p<0.01

RESULTS

Left RightGenerate Irregular>Generate Regular• Bilateral inferior frontal gyrus (L>R) and

left precentral gyrus• Bilateral posterior parietal (L>R) including

the angular gyrus and intraparietal sulcus• Left ventral fusiform• Bilateral caudate and thalamus

Generate Regular>Generate Irregular• Left superior temporal gyrus (LSTG)• Right anterior cingulate

Phonologically Mismatched GI>GR• Similar to above

Phonologically Mismatched GR>GI• More extensive LSTG/STS (BA41, 42/22)• Small focus in left postcentral gyrus

PC Matched GI>GR• Similar to above

PC Matched GR>GI• None

RIP>RRP• None

RRP>RIP• LSTG

PC Mismatched RIP>RRP• None

PC Mismatched RRP>RIP• LSTG/STS (BA 41, 42/22)• Left postcentral gyrus

DISCUSSIONIn the current study, generation of irregular relative to regular past tense was associated with greater

activation of a bilateral, distributed group of primarily frontal and parietal brain regions. In contrast, relativeto irregular past tense processing, regular past tense processing was associated primarily with a single focus ofgreater activation in the LSTG, including the planum temporale and extending laterally and ventrally alongthe STG to the upper bank of the superior temporal sulcus. This region was highly related to PC and was notspecific to morphological processing as indicated by its presence during both generation and reading of theregular past tense but only when PC was greater for regular than irregular verbs.

The specific role this area may play in speech processing is not fully understood, but it contains regionspotentially associated with the early stages of acoustic to phonetic recoding, possibly including the accessingof phonemic representations necessary for this process [5]. In the current study, greater activation forphonologically complex regulars was not seen in areas involved in later stages of phonological processing,such as the IFG. In fact, there was actually greater activation in the IFC associated with irregular past tensegeneration, which is seemingly inconsistent with the lesion data suggesting preferential disruption of regularpast tense processing with damage to this area [2]. However, several factors are important to consider wheninterpreting this finding. First, these frontal regions are involved in multiple aspects of phonological andsemantic processing, thereby complicating efforts to isolate IFG activation specifically associated withmodulations in phonological complexity. In addition, preliminary analyses with the current data suggest thatthe preferential activation in both frontal and parietal areas associated with the irregulars may be significantlyreduced when reaction time is considered, suggesting that the increased activation in these areas may be atleast partially related to the greater difficulty associated with irregular past tense processing. In addition,when generation of each past tense type is contrasted with reading of that type, substantial overlap is seen infrontal regions, suggesting that these areas are important in the processing of both regular and irregular pasttense. This overlap is consistent with the previous finding that, although lesions in this area may beassociated with greater deficits in regular past tense morphology, irregular past tense processing is alsosignificantly impaired in these patients [2]. Thus, these frontal regions may be important for the processingof both regular and irregular past tense but are potentially more critical for the former. In contrast, althoughthere was some overlap in posterior temporoparietal regions as well, the activation in these regions wassignificantly greater in the contrast between generation and reading of irregular past tense forms. This findingmay reflect the greater contribution of semantics for irregular past tense processing and is consistent withreports of more selective impairment of irregular past tense morphology with posterior lesions [2].

Overall, these results support a single-system account of past tense morphology in which all verbs areprocessed through a single integrated system, but component processes within the network such as phonologyand semantics have distinct representations within the system and may differentially contribute to theprocessing of individual words depending on specific features of those words such as phonologicalcomplexity.

GI>RIP• Extensive activation of left IFG, SFG, precentral gyrus,

medial SFG, SMA, cingulate cortex, angular gyrus,supramarginal gyrus, superior parietal lobule, precuneus,inferior temporal/fusiform gyrus, basal ganglia, andthalamus.

• Right medial frontal, basal ganglia, and thalamicactivation with lesser activation in right IFG, inferiorparietal lobule, STS, and inferior temporal/fusiform.

RIP>GI• Small focus in right posterior MTG

PC Matched RIP>RRP• None

PC Matched RRP>RIP• None

REFERENCES1. Pinker S. Rules of language. Science 1991;253:530-535.2. Ullman MT, et al. A neural dissociation within language: evidence that the mental dictionary is part of a declarative memory, and that

grammatical rules are processed by the procedural system. J Cog Neurosci 1997;9:266-276.3. Joanisse MF, Seidenberg MS. Impairments in verb morphology after brain injury: a connectionist model. Proc Natl Acad Sci 1999;96: 7592-

7597.4. Bird H, Lambon Ralph MA, Seidenberg MS, McClelland JL, Patterson K. Deficits in phonology and past-tense morphology: what’s the

connection? J Mem Lang 2003;48:502-526.5. Binder J., Price C. Functional imaging of language processing. In: Cabeza R, Kingstone A, eds. Handbook on functional neuroimaging of

cognition. Cambridge:MIT Press, 2001, p. 187-251.

GR>RRP• Extensive activation of left IFG, SFG, precentral gyrus,

medial SFG, SMA, cingulate cortex, and lentiformnucleus

• Lesser activation in left inferior parietal lobule andinferior temporal/fusiform

• Right medial frontal, caudate, and thalamic activation,and small foci in right IFG and precentral gyrus

RRP>GR• Small bilateral foci in posterior cingulate (L>R) and

posterior MTG

Flattened perisylvian patchshowing the BA 41/42/22

activation

Acknowledgments: Supported by NINDS RO1 NS33576

In the contrasts between the Read Past conditions, there were no significant areas ofactivation in the right hemisphere, therefore only left hemisphere images are presented here.

STS

STG

STG

STS

Left Right