PET Studies of Phonetic Processing ofSpeech: Review, Replication, andReanalysisRobert J. Zatorre, Ernst Meyer, Albert Gjedde, and Alan C.EvansNeuropsychology/Cognitive Neuroscience Unit andMcConnell Brain Imaging Centre, Montreal NeurologicalInstitute, McGill University, Montreal, Quebec, Canada H3A2B4Positron emission tomography was used to investigate cerebral bloodflow (CBF) changes associated with the processing of speech. In afirst experiment normal right-handed volunteers were scanned undertwo conditions that required phonetic processing (discrimination offinal consonants and phoneme monitoring), and one baseline conditionof passive listening. Analysis was carried out by paired-image sub-traction, with MRI overlay for anatomical localization. Comparison ofeach phonetic condition with the baseline condition revealed in-creased CBF in the left frontal lobe, close to the border between Bro-ca's area and the motor cortex, and in a left parietal region. A secondexperiment showed that this area was not activated by a semanticjudgment task. Reanalysis of data from an earlier study, in which var-ious baseline conditions were used, confirmed that this region of leftfrontal cortex is consistently involved in phonetic tasks. The findingssupport a model whereby articulatory processes involving a portion ofBroca's area are important when phonetic segments must be extractedand manipulated, whereas left posterior temporal cortex is involvedin perceptual analysis of speech.The phonological component of auditory linguistic process-ing has been the subject of several recent studies using PETfunctional brain imaging techniques (for reviews, see Demo-net et al., 1993; Iiotti et al., 1994). These studies have yieldedpartially consistent, but also partially contradictory evidencewith respect to the cerebral regions that may be involved inphonological processes.One of the principal controversies centers around the roleof the left frontal cortex, in or near portions of Broca's area,in phonetic analysis. Several authors (Demonet et al., 1992,1994; Zatorre et al., 1992; Paulesu et al., 1993; Fiez et al., 1995)have reported increases in cerebral blood flow (CBF) in thisvicinity in tasks requiring phonetic judgments. One interpre-tation of these findings supports a role for articulatory recod-ing in phonological processes, based upon the putative artic-ulatory contribution that is represented by activation of Bro-ca's area (Zatorre et al., 1992). However, other investigators(e.g., Petersen et al., 1989; Rumsey et al., 1992; Sergent et al.,1992) have not observed Broca's area activation, but insteadreported left posterior temporal or temporoparietal CBF in-creases in tasks involving rhyme judgments. These divergentresults have lead to a debate in the literature, and doubts havebeen expressed about the validity or generality of the conclu-sions (e.g., Sergent et al., 1992; Pinker, 1994; Poeppel, inpress).Several issues are raised by this debate, and it is the aimof this article to address them. First, there are empirical dif-ferences in the results of the various studies, which must beclarified and understood. There is also a clear need for repli-cation and extension of the existing functional imaging da-tabase on auditory phonetic processes. Finally, there are someconceptual misunderstandings in the literature, which mustbe discussed.Among these misunderstandings is the nature of the tasksthat have been used to study auditory phonological coding.A distinction may be drawn between tasks that requirements of rhyme (as in the words "house" and "mouse," -whichrhyme by virtue of their similar syllabic structure), as con-trasted with tasks that require manipulation of specific pho-netic segments (e.g., the words "bag" and "pig," which bothend with the same phonetic segment [g], but do not rhyme).It is precisely the latter type of tasks that have most consis-tently resulted in CBF increases in a portion of Broca's area.In the study by Zatorre et al. (1992), for example, pairs ofmonosyllabic words were presented, all of which had differ-ent vowels. By definition therefore none of the items rhymed.In the phonetic condition, subjects responded only if thewords ended with the same phoneme. Their judgment thusrequired them to dissect the speech sound into its constituentelements, extract the relevant segment, and compare it to an-other segment. It was in the comparison of this phonetic dis-crimination condition to a passive listening baseline that ac-tivation in Broca's area was observed, together with a leftparietal region.In the study of Demonet et al. (1992), the task was morecomplex, because a sequence of phonemes had to be iden-tified, but it also required phonetic units to be segmented andcompared. This task also yielded activity within Broca's areawhen compared to a condition involving a sequence of pure-tone triplets. In a follow-up study, Demonet et al. (1994) at-tempted to dissociate the effect of phonetic ambiguity fromthe sequential processing requirements of their earlier task.These investigators reported the strongest Broca's area activ-ity when both elements are present, but they also observedweaker CBF increases in this region, even in tasks that did nothave a sequential processing component, so long as phoneticambiguity was present.In contrast to these experiments, Petersen et al. (1989)presented pairs of visual words and asked subjects to deter-mine if they rhymed or not; this was compared to passiveviewing. Sergent et al. (1992) asked subjects to determine ifa single visual letter sounded with "ee" or not, and this wascompared to either a letter spatial task, or an object catego-rization task. Petersen et al. identified a left temporoparietalarea as showing increased CBF. Sergent et al. also found acti-vation in the left temporal lobe, somewhat more anteriorlythan did Petersen et al. (1989) in one subtraction condition.Neither study found Broca's area involvement in these tasks;although Sergent et al. did observe activation