Article | Open

Intact word processing in developmental prosopagnosia

  • Scientific Reports 7, Article number: 1683 (2017)
  • doi:10.1038/s41598-017-01917-8
  • Download Citation
Received:
Accepted:
Published online:

Abstract

A wealth of evidence from behavioural, neuropsychological and neuroimaging research supports the view that face recognition is reliant upon a domain-specific network that does not process words. In contrast, the recent many-to-many model of visual recognition posits that brain areas involved in word and face recognition are functionally integrated. Developmental prosopagnosia (DP) is characterised by severe deficits in the recognition of faces, which the many-to-many model predicts should negatively affect word recognition. Alternatively, domain-specific accounts suggest that impairments in face and word processing need not go hand in hand. To test these possibilities, we ran a battery of 7 tasks examining word processing in a group of DP cases and controls. One of our prosopagnosia cases exhibited a severe reading impairment with delayed response times during reading aloud tasks, but not lexical decision tasks. Overall, however, we found no evidence of global word processing deficits in DP, consistent with a dissociation account for face and word processing.

Introduction

The recent many-to-many model of visual recognition proposes that specialised brain regions for the processing of faces and words are functionally integrated1,2,3; for example, areas specialised to recognise faces will also, to a lesser extent, contribute towards the recognition of words. The many-to-many model predicts that as a group, those with deficits in one area (e.g., face processing) should also show deficits in the other4 (e.g., word processing). Evidence for this view comes from patients with acquired prosopagnosia (AP), a disorder characterised by an inability to recognise faces following some form of trauma to the brain regions specialised for face processing; these cases have been shown to exhibit subtle word processing deficits3. Furthermore, individuals with alexia, a disorder associated with word processing deficits after damage to the brain areas specialised for processing words (typically the visual word form area: VWFA), have also been found to exhibit signs of face recognition impairment3, 5. Taken together, these findings give prima facie support to the many-to-many model’s proposal that word and face recognition are functionally integrated.

In general, however, evidence of associated deficits is not as compelling as evidence of a dissociation6. Numerous studies have shown AP and alexia cases, with unilateral damage, to be spared in their respective word and face processing abilities7,8,9,10,11. It has been suggested that the discrepancy in these results where a dissociation was found between word and face processing7,8,9,10,11, and other work that identified associative deficits between the two domains3, is due to the latter’s testing of AP cases that also suffered from object recognition impairments: these cases were likely impaired at an earlier stage of visual processing, or had damage to cortical areas that not only processed faces, but also contributed towards the recognition of words. The obvious conclusion from these results is that face and word recognition are reliant upon specialised processes that do not overlap.

Prosopagnosia can also be developmental (DP) in nature, occurring in individuals with no history of brain damage12,13,14. DP cases have been shown to exhibit reduced matter density and abnormal neural responses to faces throughout the brain’s face processing regions15,16,17,18,19. Typically, the Warrington Recognition Memory Test for Words20 has shown no evidence of word processing impairment in DP21,22,23, although it only comprises a single study-test cycle and thus may be too crude to detect subtle reading impairments. More recently it has been shown that DP cases are apparently unimpaired when reading aloud words of various lengths24, 25 and single letters24. These studies, however, comprised basic reading tasks which did not fully test word processing under a broad set of linguistic and perceptual demands.

Alexia cases exhibit abnormally slower reading latencies as word length increases, otherwise known as the word length effect (WLE)26, 27. However, these impairments are directly linked to damage in the VWFA and the confusability of a word’s constituent letters, that is, how perceptually similar (confusable) each letter in the word is to other letters in the alphabet26. For example, O is highly confusable due its similarity with C, G and Q, by contrast, X is low in confusability because of its dissimilarity to other letters28. When a word’s summed confusability is controlled for across words of different lengths, the WLE is abolished in alexia26. This suggests that alexia cases only exhibit abnormal WLEs due to the increasing confusability of a word’s constituent letters, rather than its actual length per se.

In this respect, it is maybe not surprising that DP cases evinced neurotypical reading abilities in recent studies where confusability was not controlled for24, 25; those with alexia only exhibit an abnormal WLE as confusability increases with increasing word length. This fact suggests the need for DP cases to be thoroughly tested on a battery of tasks where confusability is carefully controlled for. If alexia and DP cases share similar deficits in their early perceptual processing of faces and words, then those with DP should show similarly elevated WLEs when the sum confusability of a word increases with word length. Conversely, we should also see those with DP exhibit neurotypical WLEs when asked to read aloud words where sum confusability is held constant as word length increases.

In addition to reading aloud, lexical decision tasks, where participants are asked to quickly decide whether a presented string of letters constitutes a valid word or not, are a popular tool to test word recognition29. While neuropsychological evidence has shown that damage to the VWFA impairs reading aloud, lexical decision making is spared30, suggesting a dissociation between these two tasks. However, despite reading words aloud and alexia being directly linked to the VWFA, neuroimaging research has suggested that reading relies more on the dorsal pathway, whereas lexical decisions are associated with a stronger involvement of the occipito-temporal cortex31,32,33 which includes many of the face related cortical regions. A case could therefore be made that lexical decision tasks, rather than simple reading aloud tasks, might be better suited to testing the many-to-many model’s predictions of common word and face processing deficits in DP.

Similarly, DP cases are characterised by their very inability to retrieve confirmation that a face has been encountered before. We hypothesise that performance in lexical decision tasks, rather than naming tasks, might be more diagnostic of the common difficulties DP cases experience when judging facial identity. When participants see a word during a lexical decision task, they need to access the semantic memory system which stores facts about the world to confirm that they know that this word is a word34. Recognition memory models typically posit that recognition works the same way for different types of stimuli, with words and faces both able to elicit a familiarity signal on which a recognition decision is based35. There are a series of stages at which this type of recognition can fail for face stimuli in DP. In some cases, those with DP may fail to match the presented face to a previously stored representation due to poor perceptual processing of the face’s attributes. Other DP cases, however, are thought to be successful in perceptually activating this stored representation of the presented face36. Instead, this success in perceptual processing somehow fails to connect downstream to the semantic memory store to confirm familiarity or to episodic memory where information such as when and where the face was previously encountered is stored. In this respect, if the face recognition system is integrated with word recognition, then we should expect to see those with DP exhibiting similar failures, either through mistakenly judging the lexicality of a visually presented word or non-word, or being slower in confirming word familiarity due to degraded perceptual processing. It should be noted that individuals with DP are generally able to confirm that a celebrity’s name is known to them. For example, after a famous faces test the experimenter will check whether the DP case has failed to recognise a particular face because of their face recognition problems, or simply because they do not know who the celebrity is. While this may indicate that DP cases are unimpaired at processing the familiarity of non-face stimuli, no study has yet confirmed this fact with a lexical decision task.

Confusability and word length place distinct perceptual demands upon the visual recognition system, however, this system can also be tested in its ability to process words of changing linguistic complexity. For example, the mere frequency of a word appearing in written language can crudely index one’s level of visual experience with that word. If DP is associated with deficits in their sensitivity to experience, then such deficits should not only impact their ability to identify famous faces, but also impair performance on word processing tasks where word frequency is varied. Similarly, the age at which one acquires a word has also been shown to affect reading performance37. Age of Acquisition (AoA), however, is linked to word frequency, and both variables should therefore be examined jointly. Finally, a word’s orthographic neighbourhood is comprised of all other words that can be derived by changing one of its constituent letters (the size of a word’s neighbourhood is denoted by N)38; for example, lob has the orthographic neighbours mob, log, lot and lab. Intriguingly, activity in the brain’s right hemisphere, which exhibits many of the neural abnormalities in DP15,16,17,18,19, appears to be sensitive to N39. Under the assumptions of the many-to-many model, one might therefore expect any linguistic deficits in DP to vary with N.

We tested the many-to-many model’s account of visual recognition by examining the performance of a group of DP cases on a comprehensive battery of 7 behavioural word recognition experiments. We label tasks where we vary word length across conditions as testing the role of perceptual information in word processing due to the fact that such a manipulation varies the physical length of our stimuli between trials. By contrast, any task that maintains the physical length of words while varying linguistic properties, such as frequency or AoA, will be labelled as testing the processing of linguistic information. We should add a caveat, however, that this classification is rather crude and is only meant to facilitate discussion of the different tasks. While the many-to-many model broadly predicts word processing impairments in prosopagnosia, it may be the case that these impairments only manifest themselves when demands are placed upon perceptual, rather than linguistic, processing. If DP cases were also impaired in linguistic processing, then it might indicate a much more basic, low level visual problem where words and faces are processed prior to functionally specialised regions. We therefore wanted to examine whether this was the case across perceptual and linguistic tasks. Each set of tasks consisted of one lexical decision task and a number of word reading tests.

Methods

Participants

The 11 DP cases that participated in the behavioural tasks were aged 20–73 years old (Mean = 41.55 years, 3 males). The 37 controls comprised of 2 groups: a younger group of 18 participants aged between 20–33 years (Mean = 23 years, 6 males) and an older group of 19 participants ranging from 56–77 years (Mean = 66 years, 7 male) to be roughly comparable to DP cases aged 32 years and younger or 52 years and older respectively. Due to the small numbers in each group, they were collapsed together for our analyses. All participants had normal or corrected to normal vision and were native English speakers. All controls and DP cases were either studying at, or had completed, university education. None of the controls reported difficulties in recognising faces, a fundamental criterion for prosopagnosia, and none of the participants had dyslexia. It should be noted that due to time constraints, not all DP cases completed all 7 behavioural tasks but their data is still included where possible. The study was given ethical approval by the Swansea University Research Ethics Committee. All methods were carried out in accordance with approved guidelines and required informed consent to be obtained from all participants.

Figure 1 lists the DP cases that participated in the experiments and their neuropsychological tests of face processing impairment, which included: a shortened Famous Faces Test40 (FFT), the Cambridge Face Memory41 (CFMT), and the Cambridge Face Perception Test42 (CFPT), with further details found in the citations. We collected data for the shortened FFT from 164 participants (101 female) to ascertain normative means and SDs for the general population in the local geographical area (M = 94.6%, SD = 6.23). Normative scores for the CFMT and CFPT were taken from the cited literature. As can be seen from Fig. 1, all of our DP participants scored more than 2 SD below the control mean on the FFT and CFMT, with 4 showing impaired performance on the CFPT. As with previous DP research43,44,45, our criteria for identifying DP cases required impairment on both the CFMT and FFT.

Figure 1
Figure 1

Neuropsychological testing results of the 11 DP cases that participated in the experiments. Columns indicate: Famous Faces Test (FFT), Cambridge Face Memory Test (CFMT), Cambridge Face Perception Test upright and inverted (CFPTupr and CFPTinv).

General Procedure

The seven experiments were completed in a random sequence for each participant. We analysed our data using mixed model ANOVAs, the purpose of which was to test the prediction that individuals with prosopagnosia should, as a population, exhibit word processing deficits4. To this end, we only report main effects or interactions involving the factor Group (controls vs. DP cases), with any follow up comparisons Bonferroni corrected. All response times were for correct responses and all group analyses two-tailed. Bayesian analyses were also performed to test the weight of evidence for the null hypothesis (Supplementary Information). Slope values for the word length effect27 were calculated by regressing the response times and errors, with individual DP cases’ WLEs reported in the Supplementary Information. Additionally we used the Crawford’s t-test46 to detect any abnormalities in individual DP cases’ performance. As we were testing the many-to-many model’s prediction that DP cases would exhibit global deficits in word recognition, we used a one-tailed test with 18 degrees of freedom to produce a critical t-value of 1.737 for the older DP cases; any individual with a t-value above this score will be identified as impaired. The critical t-value for the younger group with 17 degrees of freedom was 1.743. Any variables (e.g., bigram frequency) that were matched across conditions on any given task were confirmed as not being statistically different from one another. All word lists are provided on Scientific Reports’ website.

Impact of Perceptual Information (Word Length)

Lexical Decision: Length (word confusability not controlled)

Lexical decision tasks should reveal any difficulties DP cases may have in confirming word familiarity under perceptually demanding conditions of varying word length. Stimuli comprised 120 words and non-words. The 120 words consisted of 3 groups of 3-, 5- or 7-letters in length. Groups were matched for CELEX frequency, AoA (Bristol Norms) and bigram frequency. Mean bigram frequency merely means the frequency with which any pairs of adjacent letters found in a word occur within the printed English language. It was not possible to control for N across the 3 different letter length groups due to the inverse relationship between N and word length: 3-letter words avg. 13 neighbours, 5-letter words avg. 2.25 neighbours, 7-letter words avg. 0.2 neighbours. The 160 non-words were taken from the ARC Non-Word Database47. Non-words were pseudowords matched with the respective word stimuli for string length, orthographic neighbours, and bigram frequency. Examples include treaps, grauds and guites.

Each trial began with a centrally presented black fixation cross for 2000 ms against a white background. Then one of the 160 word or 160 non-word targets was presented in black, replacing the fixation cross. Participants were required to judge as quickly and accurately as they could, whether each target was a word or non-word by pressing the appropriate response keys on a keyboard. Immediately after their response, an asterisk (*) appeared onscreen for 500 ms before the beginning of the next trial. Presentation of the stimuli was randomised and controlled using SuperLab Pro. Stimuli were presented in 24 point, lower-case Arial font. Prior to the experiment, participants were required to complete 12 practice trials (6 words and 6 non-words).

Reading Aloud: Length (word confusability not controlled)

Alexia cases are impaired when asked to read aloud words of different lengths where confusability is not controlled for. To test whether DP cases exhibit similar impairment, we designed the present task to mimic such conditions. Word stimuli were the same as in the previous task, however, the non-words were not used; all of our reading aloud stimuli lists comprised real words alone. Each trial was exactly the same as described for the previous task apart from the following details: instead of responding word or non-word by pressing response keys, participants were required to read the word aloud when it was presented on the screen. The targets remained on the screen until the participant responded. Vocal responses were detected using an SV-1 voice key (Cedrus Software). Due to the fact that the voice key could be triggered by any sound, participants’ responses were also checked for accuracy from separate recordings using a digital voice recorder. As for the lexical decision task, participants initially completed 6 practice trials.

Reading Aloud: Length (sum confusability maintained across words)

Alexia cases are spared when reading words of different lengths where the sum confusability of all words is maintained. To examine whether DP cases exhibit similar performance, we controlled the sum confusability of all words in this task so confusability was the same for each word regardless of word length. Stimuli were comprised of 120 words taken from prior work on summed confusability so that any abnormal performance by our DP cases could be interpreted with respect to their alexia cases26. Words were matched on N, summed letter confusability and frequency while varying word length, with the 120 items comprising equal numbers of 5-, 6-, and 7-letter long words. The procedure was the same as the previous reading aloud task. Participants had to complete 6 practice trials prior to the experiment.

Reading Aloud: Length (average letter confusability maintained across words)

As mentioned, alexia cases are impaired when confusability increases across words of different lengths. We decided to better control this variable than in the second length task by maintaining the average letter confusability across words of different lengths. This will have the effect of increasing the average word confusability in a linear fashion as word length increases. If DP cases have similar difficulties in reading as those with alexia, then they should exhibit elevated WLE when attempting to read words in this condition. Stimuli comprised of 120 words again taken from prior work26, and were matched on N, average letter confusability, and frequency. Length was varied with equal numbers of our 120 stimuli comprised of 5-, 6-, and 7-letter long words. The procedure was the same as for the previous reading aloud task and included 6 practice trials.

Impact of Linguistic Information

Lexical Decision: Frequency x Age of Acquisition (AoA)

Word frequency crudely indexes our visual experience with different words. As AoA influences performance where frequency is varied, we tested participants across words that varied in AoA too. Stimuli comprised 160 words and 160 non-words. The 160 words were divided into four orthogonal conditions according to AoA (early/late) and frequency of use (high/low): half of the words were early acquired (Mean = 5.37 years of age, earliest word = 3.7 years of age, latest word = 8.3 years of age), with the remainder acquired late (Mean = 9.29 years of age, earliest word = 6.7 years of age, latest word = 12.6 years of age; Bristol Norms48). While there was some overlap between the highest and lowest AoA groups, this was necessary to still enable us to have distinct high and low frequency conditions. Within each condition (early/late AoA), half were high-frequency words and the remaining half were low-frequency. High frequency words had a word frequency score of >240 per million whereas low frequency words were <30 per million (CELEX database49). Words were matched across all groups for length (in letters), number of orthographic neighbours (N) and mean bigram frequency. The 160 non-words were taken from the ARC Non-Word Database47. Non-words were also assigned into four groups and matched with the respective word stimuli for string length, orthographic neighbours, and bigram frequency. The procedure was exactly the same as the previous lexical decision task.

Reading Aloud: Frequency x AoA

We used the word, but not non-word, stimuli from the above lexical decision task crossing word frequency with AoA in a reading aloud task using the same procedures as described for previous word naming tasks.

Reading Aloud: N Confusability

N has been shown to modulate reading performance in alexia cases when letter confusability is varied50. We therefore examined reading performance across different levels of N and letter confusability. It should be noted that this task does place considerable perceptual demands upon the visual recognition system, so may not be as exclusively testing linguistic processing as our previous linguistic tasks. Stimuli consisted of 200, 4-letter long words taken from prior work on alexia cases so that our results could be comparable if our DP cases were abnormal50. The words were varied by letter confusability and N. The words were split into 4 groups: 50 high confusability high N, 50 high confusability low N, 50 low confusability high N, and 50 low confusability low N. The cutoffs were: low N < 5, high N > 8, low confusability <0.45, and high confusability >0.53. Participants had to complete 10 practice trials prior to the main task. Procedure was the same as previous reading aloud tasks.

Results

Impact of Perceptual Information

Lexical Decision: Length (word confusability not controlled)

Figure 2 presents the results for the lexical decision task where word length was varied. To test for any possible effects of lexicality between the groups, response times were subjected to a mixed model ANOVA with Stimuli (words vs. non-words) as a within subject factor and with Group (controls vs. DP) as a between subject factor. No significant effect for the factor Group [F (1, 45) = 1.47, MSE = 68909, p = 0.23] nor any significant Group x Stimuli interaction was found [F (1, 45) = 0.001, MSE = 22, p = 0.98]. A similar ANOVA performed on the errors also revealed no effect for Group [F (1, 45) = 0.75, MSE = 12, p = 0.39], nor any Group x Stimuli [F (1, 45) = 0.15, MSE = 2, p = 0.7] interaction.

Figure 2
Figure 2

Results for Lexical Decision: Length (word confusability not controlled). Data for words and non-words (NW) are shown in left panel (response time) and right panel (number of errors) for control and DP groups. Error bars indicate ±SEM. The three word conditions are words comprised of 3-, 5-, or 7-letters. As there were 3 times more items in the NW condition, the NW errors were divided by 3 to make their rates displayed in the graph proportionally comparable to the word conditions.

To examine any possible differences between response times across length of word stimuli, an ANOVA was performed with factors of Length (3-, 5- and 7-letters) as a within subject factor and Group (controls vs. DP) as between subject factors. No significant Group [F (1, 45) = 3.6, MSE = 107057, p = 0.064] effect was found, nor any significant Length x Group interaction [F (2, 90) = 1.76, MSE = 2039, p = 0.18]. Between group comparisons on the WLE slopes also indicated that there were no significant response time (DP: M = −16.44 ms/letter; Controls: M = 2.58 ms/letter, [t(45) = 1.32, p = 0.2]) or error related (DP: M = −.75 errors/letter; Controls: M = −1.23 errors/letter, [t(45) = 1.09, p = 0.28]) WLE differences between the groups. The same ANOVA performed on the errors yielded no significant effect of Group [F (1, 45) = 1.15, MSE = 6.25, p = 0.29] nor a Group x Length interaction [F (2, 90) = 0.79, MSE = 0.81, p = 0.46]. In summary, our analyses revealed that those with DP do not exhibit any deficits in lexical decisions as word length is varied.

Reading Aloud: Length (word confusability not controlled)

Figure 3 displays the results for the reading aloud task where word length was varied. A mixed model ANOVA was performed on the response times with Length (3-, 5- and 7-letters) as a within subject factor and with Group (controls vs. DP) as a between subject factor. No significant effect was found for Group [F (1, 45) = 0.41, MSE = 13187, p = 0.53] nor any significant Group x Length interaction [F (2, 90) = 0.39, MSE = 421, p = 0.68] either. A mixed model ANOVA performed on the errors revealed no main effect for Group [F (1, 45) = 0.8, MSE = 3.41, p = 0.38] nor any significant Group x Length interaction [F (2, 90) = 0.72, MSE = 0.49, p = 0.49]. Independent samples t-tests on the WLE slopes for the response times (DP: M = 14.64 ms/letter; Controls: M = 10.79 ms/letter, [t(45) = 0.38, p = 0.7]) and errors (DP: M = −0.05 errors/letter; Controls: M = 0.18 errors/letter, [t(45) = 1.02, p = 0.31]) found no significant WLE differences between the groups. In summary, the DP group exhibited no impairment in their performance when reading words of different lengths.

Figure 3
Figure 3

Results for Reading Aloud: Length (word confusability not controlled). Data are shown in left panel (response time) and right panel (number of errors) for control and DP groups. Error bars display ±SEM. The three conditions are words comprised of 3-, 5-, and 7-letters.

Reading Aloud: Length (sum confusability of letters maintained across words)

Figure 4 displays the results for the reading task where sum confusability was kept constant across varying word lengths. A mixed model ANOVA was performed on the response times with Length (5-, 6- and 7-letters) as a within subject factor and with Group (controls vs. DP) as a between subject factor. We found no significant Group effect [F (1, 45) = 0.11, MSE = 5343, p = 0.75] nor any Group x Length [F (2, 90) = 0.07, MSE = 62, p = 0.93] interaction. A similar ANOVA performed on the errors also produced no effect of Group [F (1, 45) = 1.35, MSE = 8.35, p = 0.25] nor any Group x Length [F (2, 90) = 1.01, MSE = 0.84, p = 0.37] interaction. Between group comparisons on the slopes showed no significant differences between the groups in their response time (DP: M = 20.33 ms/letter; Controls: M = 17.58 ms/letter, [t(45) = 0.31, p = 0.76]) nor error rate related (DP: M = 0.55 errors/letter; Controls: M = 0.23 errors/letter, [t(45) = 1.61, p = 0.11]) WLE. In summary, those with DP appear to have no impairment in their reading abilities across words of different lengths when controlling for sum confusability.

Figure 4
Figure 4

Results for Reading Aloud: Length (sum confusability maintained across words). Data are shown in left panel (response time) and right panel (number of errors) for control and DP groups. Error bars show ±SEM. The three conditions are words comprised of 5-, 6-, or 7-letters with sum confusability controlled for across all words.

Reading Aloud: Length (average letter confusability maintained across words)

Figure 5 shows the results for the reading task where average confusability was kept constant as word length was varied. A mixed model ANOVA was performed on the response times with Length (5-, 6- and 7-letters) as a within subject factor and with Group (controls vs. DP) as a between subject factor. No significant main effect of Group was found [F (1, 45) = 0.96, MSE = 30264, p = 0.33], nor any significant Group x Length interaction [F (2, 90) = 0.83, MSE = 609, p = 0.44]. A similar ANOVA performed on the errors revealed no significant main effect for Group [F (1, 45) = 0.54, MSE = 4.33, p = 0.47]. The Group x Length interaction was not significant either [F (2, 90) = 1.91, MSE = 2.5, p = 0.15]. Between group comparisons on the slope values yielded no significant differences between the groups in their response time WLE (DP: M = 25.41 ms/letter; Controls: M = 18.36 ms/letter, [t(45) = 0.95, p = 0.35]) but the DP group exhibited an abnormal trend in their error related WLE (DP: M = 0.6 errors/letter; Controls: M = 0.1 errors/letter, [t(45) = 1.84, p = 0.073]). Visual inspection of Fig. 5 shows that this was due to the DP cases evincing superior performance in the 5- and 6-letter long word conditions, but comparable performance to the controls in the 7-letter condition. This suggests no apparent abnormalities in the DP group despite their elevated WLE. Overall, the DP cases did not exhibit any deficits when reading words of different lengths where average letter confusability was kept constant, as shown by their neurotypical response times and errors made.

Figure 5
Figure 5

Results for Reading Aloud: Length (average letter confusability maintained across words). Data are shown in left panel (response time) and right panel (number of errors) for control and DP groups. Error bars show ±SEM. The three conditions are words comprised of 5-, 6-, or 7-letters with average confusability controlled for across all words.

Impact of Linguistic Information

Lexical Decision: Frequency x Age of Acquisition (AoA)

Figure 6 presents the results for the lexical decision task that varied word frequency and age of acquisition. In order to test for any possible differences between the groups for lexicality, response times were subjected to a mixed model ANOVA with Stimuli (words vs. non-words) as a within subject factor and Group (controls vs. DP) as a between subject factor. No significant main effect for Group [F (1, 45) = 1.38, MSE = 55017, p = 0.25] nor any Group x Stimuli interaction [F (1, 45) = 0.36, MSE = 4640, p = 0.55] was found. A similar ANOVA on the errors revealed no significant effect was found for Group [F (1, 45) = 2.45, MSE = 166, p = 0.13]. As with the response times, no significant interaction involving Group was found either [F (1, 45) = 0.12, MSE = 3.48, p = 0.74]. Overall, there appears to be no impairment in DP when making judgments of lexicality.

Figure 6
Figure 6

Results for Lexical Decision: Frequency x AoA. Data for words and non-words (NW) are shown in left panel (response time) and right panel (number of errors) for control and DP groups. Error bars indicate ±SEM. The four word conditions are high frequency/early acquisition (H/E), high frequency/late acquisition (H/L), low frequency/early acquisition (L/E), and low frequency/late acquisition (L/L). As there were 4 times more items in the NW condition, the NW errors were divided by 4 to make their rates displayed in the graph proportionally comparable to the word conditions.

We examined any possible differences between the two groups for word types by performing an ANOVA on the response time data, with factors of Frequency (high vs. low) and AoA (early vs. late) as within subject factors and Group (controls vs. DP) as a between subject factor. No significant effect involving Group [F (1, 45) = 1.29, MSE = 55402, p = 0.26], nor any significant Group x Frequency [F (1, 45) = 0.27, MSE = 440, p = 0.61] or Group x AoA interactions [F (1, 45) = 1.45, MSE = 1942, p = 0.24] were found. Similar analyses on the errors revealed no significant effect of Group [F (1, 45) = 2.44, MSE = 15, p = 0.13], nor any significant Group x Frequency [F (1, 45) = 1.04, MSE = 1.64, p = 0.31] or Group x AoA [F (1, 45) = 0.2, MSE = 0.14, p = 0.66] interactions. In summary, our analyses revealed no differences between the controls and DP cases in their response times or errors made on our lexical decision task when varying age of acquisition and word frequency.

Reading Aloud: Frequency x AoA

Figure 7 presents the results for the reading aloud task that varied word frequency and age of acquisition. Response times were subjected to a mixed model ANOVA with Frequency (high vs. low) and AoA as within subject factors and with Group (controls vs. DP) as a between subject factor. We found no significant effect for Group [F (1, 45) = 0.8, MSE = 32077, p = 0.38], nor any significant Group x Frequency [F (1, 45) = 0.4, MSE = 184, p = 0.53] or Group x AoA [F (1, 45) = 0.52, MSE = 233, p = 0.48] interaction.

Figure 7
Figure 7

Results for Reading Aloud: Frequency x AoA. Data are shown in left panel (response time) and right panel (number of errors) for control and DP groups. Error bars indicate± SEM. The four conditions are words of high frequency/early acquisition (H/E), high frequency/late acquisition (H/L), low frequency/early acquisition (L/E), and low frequency/late acquisition (L/L).

We also performed a mixed model ANOVA on the errors, but these analyses revealed no significant main effect for Group [F (1, 45) = 0.55, MSE = 5.8, p = 0.46] as well as no Group x Frequency [F (1, 45) = 0.44, MSE = 0.25, p = 0.51] or Group x AoA [F (1, 45) = 0.15, MSE = 0.25, p = 0.7] interactions. In summary, word reading performance for individuals with DP was similar to those of the controls when varying age of acquisition and word frequency.

Reading Aloud: N Confusability

Figure 8 displays the results for the reading task where N and confusability were varied. Response times were subjected to a mixed model ANOVA with N (high vs. low) and Confusability (High vs. Low) as within subject factors and with Group (controls vs. DP) as a between subject factor. No significant main effect for Group was found [F (1, 44) = 0.007, MSE = 247, p = 0.94], nor any significant Group x Confusability [F (1, 44) = 0.13, MSE = 73, p = 0.72] or Group x N [F (1, 44) = 2.34, MSE = 877, p = 0.13] interactions. A similar ANOVA performed on the errors revealed no significant main effect for Group [F (1, 44) = 0.024, MSE = 0.15, p = 0.88]. The Group x Confusability [F (1, 44) = 0.027, MSE = 0.019, p = 0.87] and Group x N [F (1, 44) = 0.029, MSE = 0.041, p = 0.87] interactions were not significant either. As in the case of previous tasks, we failed to find evidence of any deficits in DP cases’ abilities to read words that were varied by N and Confusability.

Figure 8
Figure 8

Results for Reading Aloud: N Confusability. Data are shown in left panel (response time) and right panel (number of errors) for control and DP groups. Error bars indicate ±SEM. The four conditions are words of high n/high confusability (HN/HC), high n/low confusability (HN/LC), low n/high confusability (LN/HC), and low n/low confusability (LN/LC).

Global Analyses of Behavioural Tasks

We were concerned that there might have been a global deficit in word recognition in DP that was not identified by our analyses on any individual task. To this end, we collapsed each task’s word conditions together to give us a mean response time for word processing on any given task. These values were then converted to z-scores to standardise measures across experiments, and subjected to a 2 × 2 × 7 mixed model ANOVA with a between subject factor of Group (controls vs. DP) and within subject factors of Experiment (1, 2, 3, 4, 5, 6 and 7) and Measure (Response Times and Errors). These analyses revealed no significant effect for Group [F (1, 43) = 1.08, MSE = 4.35, p = 0.31] nor any Group x Experiment [F(6, 258) = 1.04, MSE = 0.4, p = 0.4], Group x Measure [F(1, 43) = 0.02, MSE = 0.03, p = 0.9], or Group x Experiment x Measure [F(6, 258) = 0.43, MSE = 0.13, p = 0.86] interactions. In conclusion, those with DP exhibited no evidence of global word processing deficits.

Performance of Individual DP Cases Impaired on a Single Condition

The many-to-many model predicts that as a group, prosopagnosia cases should exhibit word processing deficits. We found no evidence to support this suggestion anywhere in our group analyses. To better examine whether there were any consistent impairments in our individual DP cases across tasks, we present the average t-values of each DP case that was impaired in any single condition for each task in Fig. 9. DP8 was the only case that exhibited clear impairment, with abnormally slower response times apparent on all reading aloud tasks, although they were spared in their error rates. Remarkably, DP8 appeared superior to the control mean response times when making lexical judgments in our first two tasks. This suggests that reading and lexical decision making performance can be dissociated.

Figure 9
Figure 9

T-values of individual DP cases that were impaired on a single condition. The columns on the left show response time performance with the columns on the right showing the error rates. L1-L4 relates to the four length tasks in order as presented in our Results section. AxF LDT = Age of Acquisition × Frequency Lexical Decision task, AxF Name = Age of Acquisition × Frequency Naming task, N/Con = N × Confusability task. Asterisks highlight lexical decision tasks and Xs denote tasks that were not completed by the participant.

Discussion

The many-to-many model1,2,3,4 proposes that specialised regions in the brain that recognise faces, will also contribute functionally to word recognition. If this were the case, then one would expect that individuals with DP, who show lifelong impairments of face recognition, to also exhibit deficits in word recognition. In our present study, we administered two lexical decision tasks and five reading aloud tasks to test some of the many-to-many model’s predictions. As a group, our DP participants did not exhibit any deficits in word recognition, thus complementing similar findings in developmental24, 25 and acquired cases of prosopagnosia8, 10, 11. We therefore propose that face recognition abilities are reliant upon dissociable neural substrates from those involved in reading or lexical decision making; a domain-specific account of face recognition fits best with the data from our present study.

Impact of Perceptual Information

Our first tasks were designed to test perceptual processing of words by varying word length while controlling for a variety of linguistic factors. Whereas alexia cases will exhibit impairment when speaking longer words aloud due to increasing confusability26, our DP cases appear spared in this respect. Similarly, our DP cases exhibited no evidence of impairment when confusability was controlled for on our other tasks. Our results neatly complement similar findings in developmental24, 25 and acquired3 prosopagnosia cases. The obvious conclusion that seems apparent from these results is that the many-to-many model is fundamentally wrong in its assumption that prosopagnosia cases must share common perceptual deficits in word processing. Instead, a dissociation account for words and faces is more compatible with our current findings.

It should be noted that our DP cases did exhibit elevated error related WLEs when reading words that were matched for average letter confusability. While elevated WLEs are a classic symptom of alexia, these abnormalities should also be apparent in a reading aloud condition uncontrolled for confusability; something we did not find here. Our DP cases only produced abnormal WLEs because they were making fewer errors on the lower length words, with the number of errors made to the longest words in each condition comparable to the controls. This would indicate that DP cases are not actually impaired in reading as their error rates might level out at words over 6-letters long.

Impact of Linguistic Information

Our second set of tasks examined linguistic processing of words in DP. As in the case of word processing under various perceptual demands, the DP cases as a group seemed unimpaired. Such a finding indicates that DP is not associated with deficits related to their visual experience with words, as indexed by word frequency. This finding is in contrast to such obvious deficits they have in recognizing highly familiar faces. Similarly, despite N being related to processing in the right hemisphere, our DP cases were again generally spared. When the present findings are taken in combination with recent work in developmental24, 25 and acquired3 cases, one must conclude that the perceptual and linguistic processing of words during reading and lexical decisions is not reliant upon face processing regions. Instead, such processes must rely upon distinct regions that are not involved in face recognition.

Dissociation Between Reading Aloud and Lexical Decision Making

Despite those with DP appearing spared in their word processing abilities, we do find that one of our cases exhibited a severe reading impairment with spared lexical decision making. This raises the possibility that a small number of those with DP might also have difficulties when processing words, and that these deficits can dissociate judgments of lexicality and reading ability. Prior work on neuropsychological populations has indicated that word naming and lexical decisions can under certain conditions be doubly dissociated51, 52, although this is the first time we believe such a dissociation has been shown in DP. The fact that this case is impaired regardless of confusability suggests that their difficulties are distinctly different from those observed in alexia.

Why does this case show reading deficits? Nothing from their neuropsychological profile (Fig. 1) appears to mark them out as being particularly unique from our other DP participants. One explanation might be that they suffer from a global neural abnormality that not only affects their face selective regions, but also their word processing areas’ functioning too. Similarly, it might not be the regions specialized to process words themselves that are abnormal, but rather that there are impairments in these regions’ abilities to route signals to one another. DP cases have been shown to exhibit reduced connectivity from posterior to anterior face selective regions18, 53; it might be possible that this DP case also suffers from reduced connectivity between posterior and anterior word selective regions. A third explanation might be that they are impaired at a stage where face and word processing are reliant upon shared cortical resources, possibly in the areas involved in low level visual processing. This latter suggestion seems unlikely though due to the fact that this case is unimpaired in making lexical decisions, and such a task would surely be reliant upon the same low level processes. Neuroimaging research should help elucidate the neural locus of these deficits in such cases.

Can the Many-to-Many Model be Salvaged?

Our findings are unequivocally incompatible with the many-to-many model’s specific prediction that prosopagnosia cases must be impaired in word processing4. Despite this, there are a remarkable number of ways in which data from any word processing experiment in DP can be reinterpreted so that the many-to-many model appears salvageable; or at least salvageable with regards to face and word processing competing for resources in the word and face perception areas of the cortex. For example, if DP cases exhibited superior performance, then this could be interpreted as being due to the existence of domain general, neuronal populations that are typically shared between words and faces during early perceptual processing. The lack of competition in these areas from faces in DP might lead to enhanced cortical specialization for words instead. This greater availability of cognitive resources that undergo ‘neuronal recycling’54, 55 would enable DP cases to exhibit superior performance when processing words. Similarly, despite it seeming implausible, unimpaired word processing may only arise due to neuronal recycling counterbalancing similar levels of impairment caused by underlying neural abnormalities. Conversely, any signs of deficits can be interpreted as confirming the many-to-many model’s predictions of prosopagnosia cases being abnormal in word processing due to associative face processing deficits.

All of these interpretations, however, only subserve to highlight an inherent weakness in the many-to-many model’s construction, whereby every single research outcome from DP cases can be interpreted as supporting it. Arguably, one of the most important requirements for a new model is to allow researchers to test very specific hypotheses derived from its predictions. Unfortunately, the many-to-many model in its current form does not allow any such clear predictions to be made. The many-to-many model is exceptional in its simplicity, but such simplicity fails to explain not only why face and word recognition appear dissociable, but also why many aspects of visual word recognition itself can be dissociated into distinct streams that do not overlap. For example, as mentioned earlier damage to the VWFA spares lexical decision making30 and also reading under a variety of conditions26. Such results are clearly incompatible with the many-to-many model’s prediction that the visual recognition system is functionally integrated. Similarly, the lack of deficits in our DP cases’ response times or error rates suggest that it is unlikely that DP cases to also suffer from subtle abnormalities in their VWFA. Instead, the simple suggestion that word and face processing are dissociable seems to fit best with our findings and recent work24, 25.

One possibility that can reconcile the many-to-many model with our data is that face recognition deficits in DP might not be due to impaired processing in perceptual areas, such as the FFA. Instead, these deficits may be caused by regions that process or store face identity information downstream from visual perception. Support for this suggestion comes from DP cases exhibiting atypical memory related electrophysiological responses, which are thought to be driven by activity in the hippocampal area45, and abnormalities in their anterior temporal lobe56. However, there appears sufficient neuroimaging evidence to suggest that abnormalities in the occipital and fusiform face areas are at least in part causing face recognition problems in DP15, 18, 19, 57. Similarly, even our DP cases with severe deficits in perceptual processing, as indexed by the CFPT, do not seem to exhibit any greater difficulty processing words than our non-apperceptive cases; the one DP case who was impaired when reading, also displayed no related perceptual deficits in face processing. Taken together, it would seem that perceptual difficulties in processing faces do not produce associative impairments when processing words.

In conclusion, we have found no evidence that individuals with DP are impaired when recognizing words. Our findings are incompatible with the suggestion that deficits in face recognition must also lead to word processing impairment4. While a number of arguments have been made against the suggestion that face recognition is not distinct from other forms of visual recognition1, 58, 59, we again find compelling evidence that such an ability must rely upon dissociable processes from that of word recognition. A modular view of face processing is supported by a wealth of animal60, infant61,62,63, behavioural64, 65, genetic66, neuroimaging67,68,69 and neuropsychological8, 10, 70 research. The present study contributes to this already ample body of work that face recognition in adulthood is domain-specific and does not overlap with reading or lexical decision making.

Additional Information

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

  1. 1.

    Behrmann, M. & Plaut, D. C. Distributed circuits, not circumscribed centers, mediate visual recognition. Trends in cognitive sciences 17, 210–219, doi:10.1016/j.tics.2013.03.007 (2013).

  2. 2.

    Plaut, D. C. & Behrmann, M. Complementary neural representations for faces and words: A computational exploration. Cognitive neuropsychology 28, 251–275, doi:10.1080/02643294.2011.609812 (2011).

  3. 3.

    Behrmann, M. & Plaut, D. C. Bilateral hemispheric processing of words and faces: evidence from word impairments in prosopagnosia and face impairments in pure alexia. Cerebral Cortex bhs390 (2014).

  4. 4.

    Plaut, D. C. & Behrmann, M. Response to Susilo and Duchaine: beyond neuropsychological dissociations in understanding face and word representations. Trends in cognitive sciences 17, 546, doi:10.1016/j.tics.2013.09.010 (2013).

  5. 5.

    Roberts, D. J. et al. Processing deficits for familiar and novel faces in patients with left posterior fusiform lesions. Cortex 72, 79–96, doi:10.1016/j.cortex.2015.02.003 (2015).

  6. 6.

    Coltheart, M. Cognitive neuropsychology. Stevens’ handbook of experimental psychology (2002).

  7. 7.

    Campbell, R., Landis, T. & Regard, M. Face recognition and lipreading. Brain 109, 509–521, doi:10.1093/brain/109.3.509 (1986).

  8. 8.

    Farah, M. J. Cognitive neuropsychology: Patterns of co-occurrence among the associative agnosias: Implications for visual object representation. Cognitive Neuropsychology 8, 1–19, doi:10.1080/02643299108253364 (1991).

  9. 9.

    Moscovitch, M., Winocur, G. & Behrmann, M. What is special about face recognition? Nineteen experiments on a person with visual object agnosia and dyslexia but normal face recognition. Cognitive Neuroscience, Journal of 9, 555–604, doi:10.1162/jocn.1997.9.5.555 (1997).

  10. 10.

    Susilo, T., Wright, V., Tree, J. J. & Duchaine, B. Acquired prosopagnosia without word recognition deficits. Cognitive Neuropsychology 32, 321–339, doi:10.1080/02643294.2015.1081882 (2015).

  11. 11.

    Hills, C. S., Pancaroglu, R., Duchaine, B. & Barton, J. J. Word and text processing in acquired prosopagnosia. Annals of neurology 78, 258–271, doi:10.1002/ana.24437 (2015).

  12. 12.

    De Haan, E. H. A familial factor in the development of face recognition deficits. Journal of Clinical and Experimental Neuropsychology 21, 312–315, doi:10.1076/jcen.21.3.312.917 (1999).

  13. 13.

    Duchaine, B. C. Developmental prosopagnosia with normal configural processing. Neuroreport 11, 79–83, doi:10.1097/00001756-200001170-00016 (2000).

  14. 14.

    Duchaine, B. C., Parker, H. & Nakayama, K. Normal recognition of emotion in a prosopagnosic. Perception 32, 827–838, doi:10.1068/p5067 (2003).

  15. 15.

    Garrido, L. et al. Voxel-based morphometry reveals reduced grey matter volume in the temporal cortex of developmental prosopagnosics. Brain 132, 3443–3455, doi:10.1093/brain/awp271 (2009).

  16. 16.

    Lueschow, A. et al. The 170 ms response to faces as measured by MEG (M170) is consistently altered in congenital prosopagnosia. PloS one 10, e0137624, doi:10.1371/journal.pone.0137624 (2015).

  17. 17.

    Rivolta, D. et al. Multi-voxel pattern analysis (MVPA) reveals abnormal fMRI activity in both the “core” and “extended” face network in congenital prosopagnosia. Frontiers in human neuroscience 8, doi:10.3389/fnhum.2014.00925 (2014).

  18. 18.

    Thomas, C. et al. Reduced structural connectivity in ventral visual cortex in congenital prosopagnosia (2008).

  19. 19.

    Zhang, J., Liu, J. & Xu, Y. Neural Decoding Reveals Impaired Face Configural Processing in the Right Fusiform Face Area of Individuals with Developmental Prosopagnosia. The Journal of Neuroscience 35, 1539–1548, doi:10.1523/JNEUROSCI.2646-14.2015 (2015).

  20. 20.

    Warrington, E. K. Recognition Memory Test: Rmt.(Words). Test Booklet 1. (NFER-Nelson Publishing Company, 1984).

  21. 21.

    Steede, L. L., Tree, J. J. & Hole, G. J. I can’t recognize your face but I can recognize its movement. Cognitive Neuropsychology 24, 451–466, doi:10.1080/02643290701381879 (2007).

  22. 22.

    Longmore, C. A. & Tree, J. J. Motion as a cue to face recognition: evidence from congenital prosopagnosia. Neuropsychologia 51, 864–875, doi:10.1016/j.neuropsychologia.2013.01.022 (2013).

  23. 23.

    Liu, R. R., Corrow, S. L., Pancaroglu, R., Duchaine, B. & Barton, J. J. The processing of voice identity in developmental prosopagnosia. Cortex 71, 390–397, doi:10.1016/j.cortex.2015.07.030 (2015).

  24. 24.

    Starrfelt, R., Klargaard, S. K., Petersen, A. & Gerlach, C. Are reading and face processing related? An investigation of reading in developmental prosopagnosia. bioRxiv, 039065 (2016).

  25. 25.

    Rubino, C., Corrow, S. L., Corrow, J. C., Duchaine, B. & Barton, J. J. S. Word and text processing in developmental prosopagnosia. Cognitive Neuropsychology 33, 315–328, doi:10.1080/02643294.2016.1204281 (2016).

  26. 26.

    Fiset, D., Arguin, M., Bub, D., Humphreys, G. W. & Riddoch, M. J. How to make the word-length effect disappear in letter-by-letter dyslexia implications for an account of the disorder. Psychological Science 16, 535–541, doi:10.1111/j.0956-7976.2005.01571.x (2005).

  27. 27.

    Bub, D. N., Black, S. & Howell, J. Word recognition and orthographic context effects in a letter-by-letter reader. Brain and Language 36, 357–376, doi:10.1016/0093-934X(89)90073-4 (1989).

  28. 28.

    Mueller, S. T. & Weidemann, C. T. Alphabetic letter identification: Effects of perceivability, similarity, and bias. Acta Psychologica 139, 19–37, doi:10.1016/j.actpsy.2011.09.014 (2012).

  29. 29.

    Coltheart, M., Rastle, K., Perry, C., Langdon, R. & Ziegler, J. DRC: a dual route cascaded model of visual word recognition and reading aloud. Psychological review 108, 204–56, doi:10.1037/0033-295X.108.1.204 (2001).

  30. 30.

    Bub, D. N. & Arguin, M. Visual Word Activation in Pure Alexia. Brain and Language 49, 77–103, doi:10.1006/brln.1995.1022 (1995).

  31. 31.

    Fiebach, C. J., Friederici, A. D., Müller, K. & Von Cramon, D. Y. fMRI evidence for dual routes to the mental lexicon in visual word recognition. Journal of cognitive neuroscience 14, 11–23, doi:10.1162/089892902317205285 (2002).

  32. 32.

    Paulesu, E. et al. Dyslexia: cultural diversity and biological unity. Science 291, 2165–2167, doi:10.1126/science.1057179 (2001).

  33. 33.

    Katz, L. et al. Behavioral and neurobiological effects of printed word repetition in lexical decision and naming. Neuropsychologia 43, 2068–2083, doi:10.1016/j.neuropsychologia.2005.03.022 (2005).

  34. 34.

    Tulving, E. Episodic and semantic memory: Where should we go from here? Behavioral and Brain Sciences 9, 573–577, doi:10.1017/S0140525X00047257 (1986).

  35. 35.

    Mandler, G. Recognizing: The judgment of previous occurrence. Psychological review 87, 252–271, doi:10.1037/0033-295X.87.3.252 (1980).

  36. 36.

    Eimer, M., Gosling, A., Nicholas, S. & Kiss, M. The N170 component and its links to configural face processing: a rapid neural adaptation study. Brain research 1376, 76–87, doi:10.1016/j.brainres.2010.12.046 (2011).

  37. 37.

    Morrison, C. M. & Ellis, A. W. Roles of word frequency and age of acquisition in word naming and lexical decision. Journal of Experimental Psychology: Learning, Memory, and Cognition 21, 116–133, doi:10.1037/0278-7393.21.1.116 (1995).

  38. 38.

    Andrews, S. Frequency and neighborhood effects on lexical access: Lexical similarity or orthographic redundancy? Journal of Experimental Psychology: Learning, Memory, and Cognition 18, 234–254, doi:10.1037/0278-7393.18.2.234 (1992).

  39. 39.

    Lavidor, M. & Ellis, A. W. Orthographic neighborhood effects in the right but not in the left cerebral hemisphere. Brain and Language 80, 63–76, doi:10.1006/brln.2001.2570 (2002).

  40. 40.

    Duchaine, B. & Nakayama, K. Dissociations of face and object recognition in developmental prosopagnosia. Journal of Cognitive Neuroscience 17, 249–261, doi:10.1162/0898929053124857 (2005).

  41. 41.

    Duchaine, B. & Nakayama, K. The Cambridge Face Memory Test: Results for neurologically intact individuals and an investigation of its validity using inverted face stimuli and prosopagnosic participants. Neuropsychologia 44, 576–585, doi:10.1016/j.neuropsychologia.2005.07.001 (2006).

  42. 42.

    Duchaine, B., Yovel, G. & Nakayama, K. No global processing deficit in the Navon task in 14 developmental prosopagnosics. Social cognitive and affective neuroscience 2, 104–113, doi:10.1093/scan/nsm003 (2007).

  43. 43.

    Bate, S., Haslam, C., Tree, J. J. & Hodgson, T. L. Evidence of an eye movement-based memory effect in congenital prosopagnosia. Cortex 44, 806–819, doi:10.1016/j.cortex.2007.02.004 (2008).

  44. 44.

    Bate, S. et al. Intranasal inhalation of oxytocin improves face processing in developmental prosopagnosia. Cortex 50, 55–63, doi:10.1016/j.cortex.2013.08.006 (2014).

  45. 45.

    Burns, E. J., Tree, J. J. & Weidemann, C. T. Recognition memory in developmental prosopagnosia: electrophysiological evidence for abnormal routes to face recognition. Frontiers in human neuroscience 8, 10.3389/fnhum.2014.00622 (2014).

  46. 46.

    Crawford, J. R. & Howell, D. C. Comparing an individual’s test score against norms derived from small samples. The Clinical Neuropsychologist 12, 482–486, doi:10.1076/clin.12.4.482.7241 (1998).

  47. 47.

    Rastle, K., Harrington, J. & Coltheart, M. 358,534 nonwords: The ARC nonword database. The Quarterly Journal of Experimental Psychology: Section A 55, 1339–1362, doi:10.1080/02724980244000099 (2002).

  48. 48.

    Stadthagen-Gonzalez, H. & Davis, C. J. The Bristol norms for age of acquisition, imageability, and familiarity. Behavior research methods 38, 598–605, doi:10.3758/BF03193891 (2006).

  49. 49.

    Baayen, R. H., Piepenbrock, R. & van H., R. The {CELEX} lexical data base on {CD-ROM}. (1993).

  50. 50.

    Arguin, M. & Bub, D. Parallel processing blocked by letter similarity in letter by letter dyslexia: A replication. Cognitive Neuropsychology 22, 589–602, doi:10.1080/02643290442000202 (2005).

  51. 51.

    Arduino, L. S., Burani, C. & Vallar, G. Reading aloud and lexical decision in neglect dyslexia patients: a dissociation. Neuropsychologia 41, 877–885, doi:10.1016/S0028-3932(03)00015-0 (2003).

  52. 52.

    Boukadi, M. et al. Lexical decision with pseudohomophones and reading in the semantic variant of primary progressive aphasia: A double dissociation. Neuropsychologia 86, 45–56, doi:10.1016/j.neuropsychologia.2016.04.014 (2016).

  53. 53.

    Avidan, G. et al. Selective Dissociation Between Core and Extended Regions of the Face Processing Network in Congenital Prosopagnosia. Cerebral Cortex 24, 1565–1578, doi:10.1093/cercor/bht007 (2014).

  54. 54.

    Dehaene, S. & Cohen, L. Cultural recycling of cortical maps. Neuron 56, 384–398, doi:10.1016/j.neuron.2007.10.004 (2007).

  55. 55.

    Dehaene, S. Evolution of human cortical circuits for reading and arithmetic: The “neuronal recycling” hypothesis. From monkey brain to human brain ed. S. Dehaene, J.-R. Duhamel, MD Hauser & G. Rizolatti, 133–157 (2005).

  56. 56.

    Behrmann, M., Avidan, G., Gao, F. & Black, S. Structural imaging reveals anatomical alterations in inferotemporal cortex in congenital prosopagnosia. Cerebral Cortex 17, 2354–2363, doi:10.1093/cercor/bhl144 (2007).

  57. 57.

    Towler, J., Parketny, J. & Eimer, M. Perceptual face processing in developmental prosopagnosia is not sensitive to the canonical location of face parts. Cortex 74, 53–66, doi:10.1016/j.cortex.2015.10.018 (2016).

  58. 58.

    Bukach, C. M., Gauthier, I. & Tarr, M. J. Beyond faces and modularity: the power of an expertise framework. Trends in cognitive sciences 10, 159–166, doi:10.1016/j.tics.2006.02.004 (2006).

  59. 59.

    Diamond, R. & Carey, S. Why faces are and are not special: an effect of expertise. Journal of Experimental Psychology: General 115, 107–17, doi:10.1037/0096-3445.115.2.107 (1986).

  60. 60.

    Tsao, D. Y., Freiwald, W. A., Tootell, R. B. & Livingstone, M. S. A cortical region consisting entirely of face-selective cells. Science 311, 670–674, doi:10.1126/science.1119983 (2006).

  61. 61.

    Bushneil, I., Sai, F. & Mullin, J. Neonatal recognition of the mother’s face. British Journal of Developmental Psychology 7, 3–15, doi:10.1111/bjdp.1989.7.issue-1 (1989).

  62. 62.

    Turati, C., Bulf, H. & Simion, F. Newborns’ face recognition over changes in viewpoint. Cognition 106, 1300–1321, doi:10.1016/j.cognition.2007.06.005 (2008).

  63. 63.

    Turati, C., Macchi Cassia, V., Simion, F. & Leo, I. Newborns’ face recognition: Role of inner and outer facial features. Child development 77, 297–311, doi:10.1111/j.1467-8624.2006.00871.x (2006).

  64. 64.

    Yin, R. K. Looking at upside-down faces. Journal of experimental psychology 81, 141–145, doi:10.1037/h0027474 (1969).

  65. 65.

    Robbins, R. & McKone, E. No face-like processing for objects-of-expertise in three behavioural tasks. Cognition 103, 34–79, doi:10.1016/j.cognition.2006.02.008 (2007).

  66. 66.

    Shakeshaft, N. G. & Plomin, R. Genetic specificity of face recognition. Proceedings of the National Academy of Sciences 112, 12887–12892, doi:10.1073/pnas.1421881112 (2015).

  67. 67.

    MacKenzie, G. & Donaldson, D. I. Dissociating recollection from familiarity: electrophysiological evidence that familiarity for faces is associated with a posterior old/new effect. Neuroimage 36, 454–463, doi:10.1016/j.neuroimage.2006.12.005 (2007).

  68. 68.

    MacKenzie, G. & Donaldson, D. I. Examining the neural basis of episodic memory: ERP evidence that faces are recollected differently from names. Neuropsychologia 47, 2756–2765, doi:10.1016/j.neuropsychologia.2009.05.025 (2009).

  69. 69.

    Kanwisher, N. Functional specificity in the human brain: a window into the functional architecture of the mind. Proceedings of the National Academy of Sciences 107, 11163–11170, doi:10.1073/pnas.1005062107 (2010).

  70. 70.

    Smith, C. N. et al. When recognition memory is independent of hippocampal function. Proceedings of the National Academy of Sciences 111, 9935–9940, doi:10.1073/pnas.1409878111 (2014).

Download references

Acknowledgements

We would like to thank Tirta Susilo and Mahalia Prater-Fahey for helping collect some of the data. E.B. is supported by a School of Humanities and Social Sciences Postdoctoral Fellowship from Nanyang Technological University.

Author information

Affiliations

  1. Nanyang Technological University, 50 Nanyang Ave, Singapore, 639798, Singapore

    • Edwin J. Burns
  2. Bournemouth University, Fern Barrow, Poole, BH12 5BB, UK

    • Rachel J. Bennetts
    •  & Sarah Bate
  3. Aberystwyth University, Aberystwyth, UK

    • Victoria C. Wright
  4. Swansea University, Swansea University to Singleton Park, Sketty, Swansea, SA2 8PP, UK

    • Christoph T. Weidemann
    •  & Jeremy J. Tree
  5. University of Pennsylvania, Philadelphia, Pennsylvania, 19104, USA

    • Christoph T. Weidemann

Authors

  1. Search for Edwin J. Burns in:

  2. Search for Rachel J. Bennetts in:

  3. Search for Sarah Bate in:

  4. Search for Victoria C. Wright in:

  5. Search for Christoph T. Weidemann in:

  6. Search for Jeremy J. Tree in:

Contributions

J.T. and V.W. designed the experiments. R.B. and V.W. collected the data. V.W. transcribed the response data. E.B. analysed the data and wrote the manuscript. All other authors were responsible for manuscript review and comments.

Competing Interests

The authors declare that they have no competing interests.

Corresponding author

Correspondence to Edwin J. Burns.

Electronic supplementary material

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Creative Commons BY

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.