Neural correlates of dialect perception in 5 month old infants

Neural correlates of dialect

perception in 5 month old

infants

A Near Infrared Spectroscopy study

Master Thesis

The University of Groningen

Research Master Linguistics

Specialization: Neurolinguistics and Models of Grammar

Natalia Egorova

S1809733

6/2, Kalinina apt. 51

Komsomolsk-na-Amure

Russia 681018

+31(0) 6 624 38 52

nataffka@gmail.com

Supervisors:

Dr. Laurie A. Stowe,

Prof. Emmanuel Dupoux,

2 TABLE OF CONTENTS

1. INTRODUCTION ... 4

2. BACKGROUND ... 7

2.1.LANGUAGE DISCRIMINATION ... 7

2.1.1. Lateralization before 4 months of age and familiarity ... 8

Language specificity from birth ... 8

Acoustic characteristics bias ... 8

Familiarity bias ... 9

Predictions for language processing in younger infants ... 11

2.1.2. Lateralization after 6 months of age and language specificity ... 11

Language-specificity through development ... 11

Learning bias... 12

Predictions for language processing in older infants ... 13

2.1.3. Language lateralization between 4 and 6 months. Familiarity or Language-specificity? ... 13

2.2.DIALECTAL DISCRIMINATION ... 14

2.2.1. Development of dialectal perception ... 14

2.2.2. Dialectal perception between 4 and 6 months ... 16

2.3. THE PRESENT STUDY ... 17

2.3.1. Rationale for the present study ... 17

2.3.2. Hypotheses and predictions ... 19

Dialectal discrimination in alternation ... 19

Dialectal preference ... 20

Hypotheses and predictions summary ... 21

3. METHOD ... 22 3.1. PARTICIPANTS ... 22 3.2. MATERIALS ... 22 3.2.1. Stimuli recording ... 22 3.2.2. Talker selection ... 23 3.2.3. Final stimuli ... 26 3.2.4. Stimuli presentation ... 27

3.3. PROCEDURES AND RECORDINGS ... 28

4. RESULTS ... 30

4.1. LOOKING TIME DATA ... 30

4.2. NIRS DATA ... 31

4.2.1. Pure vs Mixed comparison. Discrimination. ... 34

4.2.2. Parisian vs Quebecois comparison. Preference. ... 41

3

4.4. INDIVIDUAL VARIATION IN LT AND NIRS ... 48

5. DISCUSSION ... 51

5.1. LOOKING TIME ... 51

5.2. NIRS ACTIVATION ... 52

5.2.1. Discrimination: Pure vs Mixed ... 52

5.2.2. Preference: Parisian vs Quebecois ... 53

5.2.3. The relation between discrimination and preference ... 55

5.2.4. Comparative topography of activation to language and dialect ... 57

6. CONCLUSIONS ... 60

7. REFERENCES ... 61

8. APPENDICES ... 69

APPENDIX 1.STIMULI ORIGINAL SCRIPT ... 69

APPENDIX 2. FINAL VERSION OF THE SCRIPT WITH TALKERS’ MODIFICATIONS ... 71

4

1. Introduction

Infants’ ability to discriminate between native and non-native languages has been investigated in a number of studies (Bosch & Sebastian-Gallès, 1997; Minagawa-Kawai et al., 2010). The results of these studies showed that infants start out as more or less universal perceivers and converge on the vowels, consonants, rhythm, intonation, and stress categories of the language they are exposed to. More specifically, in the course of development, infants improve their native language skills and deteriorate on or completely lose their ability to perceive linguistic contrasts that are not functional in their language (e.g., Kuhl et al. 2006; but see Best, McRoberts, & Stihole, 1988). Infants’ increasing specialization in their ambient language usually translates, at the neurological level, into more brain activation to the native language and increased left lateralization, as language specific networks seated in the left-hemisphere are recruited to a larger extent (Minagawa-Kawai et al., 2007; Minagawa-Kawai et al., 2010). Language perception studies suggested that left lateralization to native contrasts occurs as a result of native language familiarity and linguistic specialization. However, since the comparison of native vs. non-native language perception conflates familiarity and linguistic relevance, the documented patterns of neural specialization could reflect the emergence of language-specific categories, or they could be attributed to the greater familiarity with the native stimuli.

In contrast to the wealth of data documenting the shift from universal to language-specific perception through the comparison of native and non-native languages (or contrasts) (see, e.g., Kuhl et al., 2008; Gervain & Werker, 2008, for recent reviews on language development), there is very little behavioral evidence about how this shift affects perception of familiar and unfamiliar dialects and absolutely no work on the neural bases of observed behaviors. At the same time investigating whether the same neural networks are involved (to the same extent) when processing native and non-native dialects could greatly advance our knowledge of what biases underlie neural specialization of language processing. Familiar and unfamiliar dialects represent the same language but differ in familiarity. Therefore, if only linguistic relevance is at play in the left-lateralization and greater activation reported for the native language, then no such differences should be found for native dialect perception, which is indexically1 but not linguistically contrastive. If both linguistic relevance and familiarity play a role, then a difference between the two dialects based on familiarity should be observed.

1

5 When addressing the issue of the bases of language and dialectal discrimination, two aspects, age and method, are of relevance, as follows.

The degree to which infants rely on familiarity or on linguistic specificity for discrimination is hypothesized to change from birth to 6 months, with younger infants mainly responding to familiarity and prosody in the input (Nazzi et al., 2000) and older infants additionally having access to language-specific segmental categories, which have been shown to emerge only by 6 months of age (Kuhl et al., 1992). Little is known, however, how familiarity and linguistic specificity interact in the transition period between 4 and 6 months between language-general and language-specific processing, and what neural bases are employed. The developmental timeline for dialects shows similarities to the language development. At 3-5 months infants discriminate between native and non-native languages (Nazzi et al., 2000, Exp. 1; Minagawa-Kawai et al., 2010) as well as familiar and unfamiliar dialects (Kitamura et al., 2006; Nazzi et al., 2000, Exp 5) and therefore are comparable. However, there is a crucial difference in that native and non-native languages should be grouped separately, as two different languages, while both familiar and unfamiliar dialects should be grouped together, into a single native language class (Best et al., 2009). This has an effect on the perception of familiar and unfamiliar dialects as compared to languages. While infants older than 6 months continue to discriminate their native language from non-native (Minagawa-Kawai et al., 2007), they no longer perceive the difference between the dialects. Indeed, at around 6-8 months infants’ dialect discrimination ability starts to decline and they perceive 2 dialects as 1 native language (Kitamura et al., 2006; Diehl et al., 2006; Phan & Houston et al., 2006) by not showing preferences for the native dialect. No information is available on how infants switch from discrimination to ignoring the linguistically irrelevant differences between the dialects and what neural processes accompany this shift that happens between 4 and 6 months of age.

6 The goal of the study was to address 3 gaps in the existing literature: 1) to disentangle familiarity and linguistic relevance in language development; 2) to identify the neural bases of dialect discrimination during the transition between language-general and language-specific processing; and 3) to provide complete information on infants’ dialect discrimination in alternation and preferences tested in one study by using neuroimaging.

In this study a group of 5 month old infants was tested on the contrast of native and

non-native dialect. The brain activity during non-native and non-non-native regional accent processing was

7

2. Background

2.1. Language discrimination

Language perception in young infants has been a topic vigorously researched over the last 30 years. A great deal of data has been obtained in various behavioral and neuroimaging studies. Many behavioral studies testing language discrimination by comparing preferences of the infants to various stimuli found that universal perception, when infants are able to differentiate between almost any linguistic contrast in the world (Eimas et al., 1971; Streeter, 1976; Best & McRoberts, 2003), changes into language-specific perception, when infants get better at discrimination of their native-language vowels and consonants, and get considerably worse at discrimination of non-native contrasts (Kuhl et al., 1992; Werker & Tees, 1984).

However, neurobiological explanations of these findings have diverged. For instance, some researchers suggest that there are no qualitative differences in processing between birth and 6 months, because already at birth, and even in utero, adult-like areas of the brain, namely left superior temporal gyrus, encompassing Heschl’s gyrus, superior temporal sulcus and the temporal pole, as well as the planum temporale, are engaged for language perception and language-discrimination (e.g. Dehaene-Lambertz et al., 2006) or because acoustic characteristics in the signal influence lateralization in both infants and adults alike, when short temporal changes elicit left-lateralized response and longer spectral characteristics require the involvement of the right hemisphere (e.g. Telkemeyer et al., 2009). In contrast, others claim that there is a shift from early processing that employs general auditory mechanisms and bilateral hemispheric activation, to more mature language-specific processing involving specialized native-language networks in the left-hemisphere (Best et al., 1982; Minagawa-Kawai et al., 2007) due to learning and increased familiarity with the native language.

8 neuroimaging studies which look at the neural development of young infants under 4 months, and older infants above the age of 6 months, but no information is available on the neurological state at 5 months.

The studies reviewed below suggest more specific reasons for hemispheric lateralization during discrimination in infants with the emphasis on the age at which linguistic specialization and familiarity are most relevant in order to make predictions of what the neural organization in the period of transition from language-general to language-specific processing is like.

2.1.1. Lateralization before 4 months of age and familiarity

As mentioned earlier there are many views on how brain lateralization develops in the first months. Three major biases suggested to describe language processing at this age are domain-specificity bias, acoustic characteristics bias, and familiarity bias.

Language specificity from birth

The domain-specificity theory suggests that language is innate, i.e. it develops in the absence of experience, and is subserved by a special ‘language’ module in the left hemisphere of the brain (Fodor, 1985). This language module is specific to humans and is not present in animals; it is genetically predefined through specific anatomical differences between left and right hemisphere favoring the development of language networks in the left temporal area of the brain. Since it is genetically preconditioned, the development of the language networks is only limited to refinement of the networks (precision of the areas involved), and reducing the activation levels (efficiency of processing) to ensure optimal treatment of linguistic stimuli. The studies showing left lateralization from birth by testing newborns and young infants under 3 months mainly contrasted language input and non-linguistic input (e.g. language vs music (Dehaene-Lambertz et al., 2009; Best et al., 1982), forward and backward speech (Pena et al., 2003)) rather than comparing neural correlates of discrimination between two languages. Therefore, although the domain-specific bias predicts left-hemisphere activation for any linguistic input at any age, no definite conclusions can be made regarding native and non-native language processing on the basis of this bias.

Acoustic characteristics bias

9 are relevant for discriminating phonemes, whereas long spectral changes are important for pitch discrimination. Since temporal coding of phonemes is crucial for language perception, predominantly left lateralization is observed for speech processing. This importance of temporal changes was shown in an adult study contrasting CV syllables and tones or vowels in isolation, with both speech and non-speech stimuli. The results showed greater involvement of the left planum temporale for speech and non-speech stimulus involving a consonant (Jancke et al., 2002).

In infant studies similar results were found. In EEG and NIRS studies (Telkemeyer et al., 2009, Minagawa-Kawai et al., 2009) as well as non-nutritive sucking and dichotic listening studies (DeCasper & Prescott, 2009) on early auditory perception, newborn infants were shown to be sensitive to the differences between spectral and temporal characteristics in signal processing treating fast temporal variations in the left-hemisphere and longer-lasting spectral variations in the right-hemisphere.

Similar conclusions were drawn by other studies. For example fetal brain response to brief rapid variations such as 58 ms tones is left-lateralized (Jardri et al., 2008), and left-lateralization is found in discrimination of the processing of the fast transitions 40-45 ms of CV syllables (Alexandra et al., 2007). Several ERP studies demonstrated right-hemisphere activation during discrimination of complex and simple tones (Cheour et al., 2002), maternal and non-maternal voices (deRegnier et al., 2002), CV vowels lasting 250 ms (Alexandra et al., 2007). Thus, the conclusion is that functional asymmetry arises depending on the quality of the stimulus. This bias suggests bilateral activation to complex speech stimuli, and does not have specific predictions for the distinction between familiar and unfamiliar stimuli since the signal will contain fast and slow modulations in both cases.

Familiarity bias

The bias that did make predictions about the processing of native and non-native stimuli in young infants is the familiarity bias. Familiarity is often mentioned as a factor influencing language acquisition; however it is one of the most ambiguous terms. Sometimes, by familiarity the authors mean experience with and exposure to the language, sometimes it is emotional feeling of attachment evoked by a familiar thing, sometime it is ‘normality’ of the stimulus (e.g. forward vs backward speech; varied vs flat prosody).

10 discriminate their mother’s voice from a stranger’s voice, even if the phrases are different (Kisilevsky et al., 2003; Moon et al., 1993, deRegnier et al., 2009). They can also discriminate between familiar passages (the ones that the mother had often recited) from unfamiliar ones (DeCasper et al., 1994, Krueger et al., 2004, Cooper & Aslin, 1989, DeCasper & Spence, 1986), familiar languages and voices from unfamiliar ones (Kisilevsky et al., 2008). At 2 months infants also prefer their native language to their foreign language and their mother’s voice over others (Dehaene-Lambertz & Houston, 1998)

As for familiarity-induced lateralization, there is no clear pattern. For example, dichotic listening studies in newborns (e.g. DeCasper & Prescott, 2009) have suggested that familiar stimuli (mother’s voice and language) are processed preferentially by the right hemisphere as a result of prenatal auditory experience. By contrast, an fMRI study on 3 month old infants revealed a number of areas activated in response to the mother’s voice in a familiar language, namely the areas of emotional processing (amygdala, orbito-frontal cortex) and importantly the left posterior temporal lobe and concluded that familiar stimuli play a role in shaping posterior language areas (Dehaene-Lambertz et al., 2009)

In fact, familiarity is often related to positive emotion, which has its own neural correlates. The effect of familiar speech of both the mother and the nurse at a neonatal intensive care unit in premature infants was explored by Saito and colleagues using NIRS. The findings show that the mother’s and nurse’s speech, which were both familiar for the infants, activated frontal and left areas of the brain, which suggests that both of the voices are associated with positive emotions. the nurse’s voice elicited right hemisphere activation which the authors interpret as a result of a negative association with stress (Saito et al., 2009).

11 The familiarity bias is very complex and non-homogenous, but has implications for interpretations for the results elicited by the native language. As for the hypotheses for language perception, preference for the native language is expected. Bilateral or right-hemisphere activation is predicted.

Predictions for language processing in younger infants

Several distinct predictions can be made on the basis of these biases. First, left-lateralized processing is predicted if domain-specific bias is at play. There cannot be anything said about the difference between the native and non-native distinction, because this bias only contrasts linguistic and non-linguistic stimuli. Second, bilateral processing is predicted if discrimination is based on both, acoustic characteristics bias, and the familiarity bias. However, bilateral

processing with no difference between native and non-native stimulus can be predicted by the

acoustic bias because in both cases the segmental and suprasegmental information is present. The bilateral processing with a preference for the native stimulus over the non-native stimulus is predicted by the familiarity bias.

2.1.2. Lateralization after 6 months of age and language specificity

While the biases discussed above applied mainly to younger infants, in older infants the left-lateralization has been explained by language-specific processing.

Language-specificity through development

The developmental account of language-specific processing suggests that left-lateralization is acquired through experience and brain development, specifically under the influence of the ambient input. The evidence in favor of the developmental account is the fact that language discrimination abilities of infants change. By 6 months infants can discriminate vowels from their native language but not non-native vowel contrasts (Kuhl el al., 2006), and by 11 months they develop language-specialized perception of consonantal contrasts as well (Werker & Tees, 1984). For example, native speakers of English improve the perception of r-l sounds from 6-8 to 10-12 months, whereas Japanese infants for whom this contrast is not native do not (Kuhl et al., 2006).

12 traces are formed on the basis of experience with the native language, and those traces are responsible for discrimination later on.

One of the theories that suggested native-language attunement as the mechanism for lateralization is Kuhl’s native language neural magnet theory (NLNM) and its expanded version postulating neural commitment to the native language. This theory indicated 3 stages of development: 1) universal differentiation of all sounds of human speech deriving from the general auditory processing mechanisms (Kuhl, 1991); 2) formation of phonetic representations under the influence of the distributional modes in the input (Kuhl, 1993, Iverson et al., 2003); and 3) transforming frequently activated phonetic representations into perceptual magnets accompanied by a distortion of perception towards the native-language contrasts. This theory postulates that domain-general abilities are responsible for the language discrimination in the initial stages of development, which with experience develop into language-specific abilities (Kuhl et al., 2008).

Another bias that explained language-specific processing in older infants is the learning bias.

Learning bias

13 For the language discrimination in infants, the learning bias suggests a more pronounced activation of the ‘learned’ native language in the left hemisphere, only for older infants.

Predictions for language processing in older infants

Both the developmental language-specificity and the learning bias suggest that around the age of 6 months infants start to process linguistic stimuli in the perisylvian areas of the left

hemisphere. This processing is adult-like and is based on the language-specific neural networks

formed as a result of the acquisition of the native language-specific segmental information. In fact, there is no behavioral evidence of language-specific processing before the age of 6 months. Phonemic learning based on statistical computation and categorization reinforces left-lateral processing as learning continues predicting that with more exposure a larger difference between

familiar and unfamiliar languages should be observed.

It should be noted that although learning and familiarity biases are somewhat similar in that they stress the role of exposure, the familiarity bias is more concerned with the social aspects of language acquisition whereas the learning bias is focused more on the language-specific processing of the ambient input.

2.1.3. Language lateralization between 4 and 6 months. Familiarity or Language-specificity?

The studies reviewed above have concentrated on the ages at which infants were found to show either universal perception and acoustic general response or already native language-specific perception. There are no neuroimaging studies that would focus on the transition period between 4 and 6 months, although they could be very informative for the identification of the biases guiding the organization of language neural networks. Therefore, no evidence on lateralization and native and non-native language perception in 5 month old infants is available. One could hypothesize that infants of this age will behave either as young infants or as older infants, as reviewed above (see Table 1). In the case that 5 month-old infants show neural organization previously shown in younger infants, bilateral response with no or a slight native language preference can be predicted on the basis of the acoustic characteristics bias and the familiarity bias. In the case that they already behave as older infants, they will show a preference and a more left-lateralized response to the native language based on the language-specificity and the learning biases.

14

Table 1 Summary of lateralization biases

Bias Lateralization Native Language

Preference

Age

Language-specificity from birth Left - Younger

Acoustic characteristics Bilateral - Younger

Familiarity Bilateral/Right + Younger

Language-specificity through development

Left + Older

Learning bias Left + Older

2.2. Dialectal discrimination

Dialects are regional accents within one language that arise from peculiarities in the use of pronunciation (Labov, 1972). Often, differences in vocabulary and the use of idioms, syntax and phrase complexity are observed, in addition to those differences in pronunciation. Dialects have an important social role, as they are frequently associated with certain social characteristics (Giles, 1970; Preston, 1989) and can be used for group categorization (Clopper & Pisoni, 2007).

Dialectal perception in adults is characterized by increased processing difficulty for the unfamiliar dialect (Mitterer & Blomert, 2003; Floccia et al., 2006); facilitation of processing through exposure (Scott & Cutler, 1984; Bowie, 2002; Clopper & Pisoni, 2004); and the principal role of early childhood experience with the dialect for subsequent dialectal perception abilities (Sumner & Samuel, 2009).

Although dialectal perception in infants under 12 months has not been studied extensively, the few existing studies focusing on dialectal processing showed that there is a difference in dialectal perception depending on the age of infants. It was tested by two major methods: visual habituation (tapping discrimination of 2 dialects in alternation) and visual fixation (tapping preference).

2.2.1. Development of dialectal perception

15 dialect, but only if they heard their native dialect first, and Australian infants did not show a differential response to the 2 dialects at all. At 8 month American infants in Diehl et al., 2006 study did not show any preference for either of the dialects, suggesting that from 3 to 8 months dialectal discrimination undergoes a decline.

The difference of the dialectal discrimination from the language discrimination is that dialects represent one native language and from birth to 6 months infants need to learn to group both dialects in one category, ignoring non-relevant differences within the native language. Therefore, instead of improving the discrimination between familiar and unfamiliar dialects, infants have to ignore the differences between them.

Kitamura and colleagues interpret the native dialect preference in 3 month olds as the use of auditory general mechanisms for discrimination. They further explain that at 6 months, American infants treat Australian as a variety of their native language, so when they hear Australian first and American second, they do not make a difference between the two. However, when they hear American first, their native dialect serves as a benchmark for further dialects and thus Australian dialect is perceived as different. Australian 6 month olds in Kitamura’s study and American 8 month olds in Diehl’s study have supposedly grouped dialects in the native-language group. Performing a discrimination task they either map an unfamiliar dialect on a familiar dialect representation, or due to a lack of cognitive resources attend only to linguistically relevant contrasts, which makes the distinction between dialects even if perceptually salient not relevant.

Similarly to the studies on language perception, the interpretation of these developmental differences in dialectal perception is based on the distinction between auditory general familiarity-based discrimination in younger infants and language specific processing in older infants. More evidence on the shift from familiarity-based processing to language-specific processing comes from the direction of infants’ preferences and the cues used for discrimination. Namely, younger infants (3 mo) show a familiarity preference and older infants (6 mo and up) show a novelty preference (Kitamura et al., 2006; Diehl et al., 2006). Besides, younger infants have been found to rely more on prosody, which is salient and emotionally charged (acoustic general processing and familiarity), whereas older infants have been shown to employ segmental information (language-specific phonemic learning), as follows.

16 influencing very young infants’ perception of different dialects. Indeed, prosody is very salient to infants (Bosch & Sebastian-Galles, 1997) and suprasegmental attunement happens at an early age (Mehler & Christophe, 1995). Previous low-pass filtered speech experiment in 4 month old infants testing discrimination of prosodically close languages, Spanish and Catalan showed that younger infants relied on suprasegmental information (Bosch & Sebastian-Galles, 1997). As for older infants, they already rely on segmental rather than on suprasegmental information for dialectal discrimination. For example, 6 month old American infants showing discrimination when tested with sentences pronounced in American and Australian female infant-directed speech (IDS), showed no difference in looking times and in heart beat when the authors low-pass filtered the IDS stimuli at 400 Hz thus removing the lexical content while preserving overall pitch, duration, and pitch contours. (Diehl et al., 2006)

2.2.2. Dialectal perception between 4 and 6 months

Little is known about dialectal processing during the transition stage: from discrimination to non-discrimination, from familiar to unfamiliar preference and the loss of preference, from the attention to prosodic cues to the use of segmental cues. The existing evidence about language and dialectal processing in 5 month old infants comes from a series of experiments manipulating familiarity with the rhythmic class run by Nazzi and colleagues. The authors included one experiment (Nazzi et al., 2000, Exp. 5) where sentences in 2 dialects of one language, British and American English, were presented in a visual habituation procedure. They found that infants discriminate between the two dialects and explained the results in terms of familiarity with the familiar dialect prosody.

This explanation is plausible, although a number of other factors could have had potential influence on the discrimination. For example, infants could have attended to the segmental information. At the segmental level, there was a rhotic “r” typical of the American dialect; there were differences in the distribution of vowels in F1/F2 space although the number and type of vowels were similar (Labov, Ash & Boberg, 2006). The infants could have detected some or all of the segmental and suprasegmental differences. Besides, the visual habituation task presupposes presenting familiar and unfamiliar dialects in alternation, emphasizing the contrast between the dialects. No information about discrimination in isolation or the infants’ preferences for either of the dialects is available.

17 English and Dutch, i.e. they discriminated an unfamiliar native dialect from a non-native language. It should be noted that they were not able to discriminate 2 rhythmically similar unfamiliar languages, like German and Dutch (Exp. 6), which suggests that American infants are able to recognize their native language even in an unfamiliar dialect.

Together these studies show that at 5 months infants like 3 month olds can discriminate between dialects based on familiarity with the native language prosody. They also can recognize the unfamiliar dialect as the native language, showing mature dialectal processing like 6 month olds. It leaves the question of whether 5 month old infants rely on familiarity or linguistic relevance for discrimination open. Nazzi did not manipulate the stimuli to test discrimination in the absence of segmental information and the lack of a preference study does not allow concluding that infants do not show preferences for either of the dialects, which would suggest language specific processing rather than familiarity based preference.

2.3. The present study

The purpose of this study was to explore cerebral bases of dialectal discrimination and preference and to compare neural networks found to be involved in language and dialectal discrimination in order to evaluate the role of familiarity and language-specificity in 5 month old infants.

2.3.1. Rationale for the present study

Method

Near-infrared spectroscopy (NIRS) was used for the study. This technique is non-invasive and is well suited for testing infants. Unlike ERP and fMRI it tolerates more movement and is noiseless. The basis of NIRS is red and infrared light that is emitted from Source optoids on a pad. The light penetrates the skull and the cortical tissue, and is then reflected to the surface where it is measured by Detector optoids. The absorption of the light, which differs for oxygenated and deoxygenated blood, is thus said to reflect the level of activation of the brain area studied (for the description of the technique, see the review of Lloyd-Fox et al., 2009).

18 Discrimination and preferences do not map directly on the neural activation observed in NIRS but can be compared based on the amount of activation and the lateralization of the response. Two comparisons (Pure/Mixed and Parisian/Quebecois) were used to tap respectively discrimination and preference. Discrimination in the Pure/Mixed condition was defined as the difference in the amount of activation in response to 1 dialect within block and to the alternation of 2 dialects within block. If infants are able to detect the change of the 2 dialects, it will show up in increased processing cost resulting from alternation in the Mixed condition. If infants do not notice the change between the dialects, no difference between Pure and Mixed trials will be present. Preference in the Parisian/Quebecois condition was measured by the direct comparison of the amount of activation to the two dialects presented in isolation. At the neural level it reflects the efficiency of processing and arousal of attention rather than the preference observed behaviorally. Therefore, the location of the source of activation is relevant to determine the nature of the observed preference: a more left-lateralized response for the familiar dialect would indicate language specificity-based preference, and a more right-lateralized or bilateral response would suggest familiarity-based preference.

Stimuli

There are important differences in the general characteristics of the stimuli used in the present study, as compared to those involved in native vs. non-native language comparisons: connected, infant-directed, audio-visual speech was used.

First of all, complex speech stimuli were used in this experiment to match the stimuli used in previous behavioral studies on dialects. Sentences with infant directed intonation were used as stimuli for the present study (Nazzi et al., 2000; Diehl et al., 2006, Kitamura et al., 2007). Often infant-directed highly intonated positively charged speech elicits activation in the right hemisphere and more specifically in its frontal areas (Saito et al., 2007; Homae et al., 2007). Infants could react to the dialects based on their emotional value, but could not be biased to prefer one over the other unless they had a familiarity preference, because both dialects were rated as equally interesting and engaging by the adult native-speakers and are therefore matched on this dimension.

19 accompanied the presentation of the stimuli. Audio-visual stimuli often elicit bilateral (Lloyd-Fox et al., 2009) or slightly more right-lateralized activation in the posterior temporal sulcus (Grossman et al., 2010). However, for both dialects the mode of presentation was the same, and therefore, this factor should not influence localization of the activation for any particular dialect. All three factors can influence the involvement of the brain areas in dialectal processing. The predominantly right activation associated with emotional familiarity is usually observed in frontal areas, and since probes in the present study are placed over temporal areas, this factor will not have a large impact. On the other hand, the exaggerated prosody of connected infant-directed speech (IDS) and the evidently social nature of the stimuli can involve temporal areas, and may contribute to a less pronounced left lateralization.

2.3.2. Hypotheses and predictions

Two sets of research questions are posed in this study: first, dialectal discrimination in alternation at 5 months is discussed and acoustic general vs language-specific processing are compared; and second, dialectal preference is reviewed and familiarity vs language-specificity are contrasted.

Dialectal discrimination in alternation

The 2 research questions related to dialectal discrimination are as follows:

a) Can dialect discrimination found in Nazzi et al., 2000 be replicated and reflected in neural activity? Activation level should be compared in pure and mixed conditions. The hypothesis regarding this research question is that if infants indeed are able to discriminate 2 dialects of French, behavioral and brain responses to Mixed condition should be bigger than to the Pure condition due to the processing cost resulting from alternation.

b) Is discrimination found in 5 month old infants based on acoustic general processing or on language-specific processing? Lateralization of the activation in the mixed condition should indicate what the basis of discrimination is.

As discussed in Section 2.1., younger infants show bilateral activation, and older infants show left-lateralized activation. Given the age of infants in this experiment, it is likely that although some attention to segmental cues is probably already present, infants will largely rely on prosodic cues and acoustic saliency and show bilateral response.

20 discrimination of the two dialects. If infants are already able to process segmental information, and discrimination of the dialects is based on the analysis of the differences in vowels and consonants, more left-hemispheric activation is expected. This could be explained by the switch from acoustic general processing and the existence of language-specific networks in the left hemisphere, which store information about the native phonemes.

Dialectal preference

c) Do infants have a preference for the familiar or unfamiliar dialects at 5 months? A preference for one of the dialects will indicate that they are perceived as 2 languages; no difference in the amount of activation should suggest that both dialects are treated as 1 language.

Previous behavioral studies show that 3-month olds prefer their own dialect (Kitamura et al., 2006), while some 6-month olds prefer a non-native dialect to the native one (American infants in Diehl et al., 2006). Other results indicate that infants treat dialectal differences as linguistically irrelevant (Best et al., 2009, Phan & Houston, 2006, Schmale & Seidl, 2009) and so not make a difference between the 2 dialects. Although most of the evidence in favor of this hypothesis comes from older infants (e.g., Schmale: 13 months; Best: 17 months), the only result on the age tested here suggests that, at some level, familiar and unfamiliar dialects may be represented similarly (Nazzi et al., 2000 in Exp. 4) Therefore, in the present study, a preference for the familiar dialect is hypothesized, but should be minor.

d) Is the dialectal preference based on familiarity or language-specificity? The

lateralization of the activation to the Parisian and Quebecois conditions should reveal

the biases responsible for the preference.

21

Hypotheses and predictions summary

22

3. Method

3.1. Participants

Thirty 5 month-old infants (M = 5.0, range 4,6-5,4; 13 females) were recruited through letters sent to parents of eligible infants. Infant information had been provided by the City Council to the Laboratory of Cognitive Science and Psycholinguistics. All of the participants were French monolinguals, with no hearing or developmental problems according to parental report. All caregivers gave a written informed consent before participating in the NIRS experiment, and their child received a certificate of participation. When necessary, parents were reimbursed for travel expenses.

Twelve infants did not complete 4 out of the 5 sequences of 4 blocks and were therefore excluded in order to have a homogeneous sample, leaving data from 18 infants. These 18 infants finished 4-5 trials for each of the Pure-Parisian and Pure-Quebecois comparisons and 8-10 trials per Mixed vs Pure comparison, for a total of 12-15 minutes testing time. Infants who provided fewer than 10 datapoints per comparison were further excluded from the analysis. For Pure vs Mixed comparison 13 (16 in the Temporal region comparison) infants contributed data; and for Parisian vs Quebecois 6 (13 in the Temporal region comparison) infants’ data were analyzed.

3.2. Materials

3.2.1. Stimuli recording

Equipment

All the recordings were made in a soundproof cabin using an AKG microphone and Sony digital camera DCR-HC96E, streaming into a Dell laptop computer (Inspiron 14) using Adobe Premier Elements 2.0. All the videos were processed with Virtual Dub-1 software. The audio tracks were extracted with Pazera Free Audio Extractor.

Procedure

23 changes they introduced, all the modifications were very consistent within dialects, and the number of syllables in the final version did not differ significantly for the two dialects (p=0,96). The modifications that Canadian speakers introduced included, for example: inserting the pronoun ‘tu’ (e.g., 'tu veux tu une banane écrasée?'), substituting phrases (e.g., “il faut que je m’en aille, byebye” instead of “il faut que je parte, au revoir”), and words (e.g., ‘mimi’ by ‘cute’), simplifying sentences. (See Appendix 2 for the adaptations to the script made by the 4 selected talkers) Immediately before recording, speakers watched a video in order to help them

tune into their respective dialects (a cooking program in Paris

http://www.youtube.com/watch?v=ypgVC0Fo3vs&NR=1 or in Quebec

http://www.youtube.com/watch?v=5lB-AX8_lSA). Throughout the recording, speakers saw a picture of a 5 month-old.

3.2.2. Talker selection

Thirteen female native speakers of French were recorded (5 Quebecois; 8 French). The Canadians were recruited through an email to the members of the group of Quebecers living in

Paris (http://www.quebecfrance.info); the French women were recruited through personal

contact. Out of the 13 speakers, 3 were discarded due to differences in the recording instructions (i.e., they had not been allowed to make changes to the script, and as a result sounded relatively unnatural). Of the remaining 10, 2 talkers of each dialect were selected to maximally reduce low-level differences between the dialects. Specifically, these talkers were the best match on the basis of three criteria: (a) prosodic parameters associated with infant-directed speech, namely average pitch, pitch excursions, and rate of speech (e.g., Fernald et al., 1989); (b) visual features (e.g., colour and size of the hair); and (c) clarity of speech and how engaging to a child they were, as judged by 5 naive French raters. Details on this selection process ensue.

Prosodic measurements. The audio tracks from the videos of the remaining 10 speakers (4

24

Table 2. Prosodic characteristics initial sample, 10 speakers. Each cell shows the F-value for the factor given on the row on the dependent measure given in the column, *** stands for p<.001, ** p<.01, * p<.05

Factor Mean Max Range Duration

Dialect 12.73*** 10.51** 4.10* 7.81**

Talker 6.84*** 2.84** 2.23* 3.93***

Figure 1 Prosodic characteristics, initial sample, 10 speakers.

Nonetheless, in the final sample, there was no main effect of dialect for any of the parameters, since four speakers with overlapping prosodic characteristics were chosen, as shown in Table 3 and Figure 2.

Table 3 Prosodic characteristics, final sample,4 speakers. Each cell shows the F-value for the factor given on the row on the dependent measure given in the column, *** stands for p<.001, ** p<.01, * p<.05

Factor Mean Max Range Duration

Dialect 0.50 1.02 0.6 1.25

25

Figure 2 Prosodic characteristics, final sample,4 speakers.

Visual features. During recording, a blue scarf was used to mask differences in the

speakers’ clothes, somewhat visible in the shoulder-length shot used. To further avoid differences in appearance, speakers were asked to take off their jewellery and glasses.

The speakers were also matched on their appearance (there were one woman with fair curly hair and one with dark straight hair per dialect) and age (range 25-29 years), see Figure 3).

Figure 3 Speakers’ appearance.

P1: P2:

26

Ratings. Finally, a rating questionnaire was constructed and administered to 5 adult

subjects (3 female, 2 male) to ensure that the 4 selected talkers were equally engaging and sounded equally interesting. The questionnaire was also meant to evaluate the strength of the talker’s accent. The rating questionnaire contained 7 questions about the speakers. A 7-point Lickert scale was used with construct-specific anchors at score 1 and score 7. (See Appendix 3 for the questionnaire). The results of this rating (see Table 4) showed that the talkers did not differ when rated on emotional characteristics and potential interest for the infant but differed significantly in the perception of their dialect (but not between talkers within 1 dialect).

Table 4. Emotional characteristics and accent perception, final sample, 4 speakers. Each cell shows the F-value for the factor given on the row on the dependent measure given in the column, *** stands for p<.001, ** p<.01, * p<.05

Factor How clear is the speech of the speaker? Is the speaker interested in conversation? Will the baby pay attention to the speaker? How affectionate is the speaker How happy is the speaker? Does the speaker have an accent? Parisian or quebecois accent? Dialect 2.13 1.50 0.10 0.76 0.40 114.37*** 92.16*** Talker 1.39 3.20 1.50 0.19 0.22 0.86 0.26 3.2.3. Final stimuli

3.2.4. Stimuli presentation

The 20 blocks were divided into 5 sequences of 4

mixed blocks and 2 pure blocks (1 Parisian and 1 Quebecois), with the order of talkers within blocks and order of presentation of type of blocks counterbalanced across infants. Each sequence of 4 blocks lasted for about 3.5 minutes. At the end of each sequence, infants were presented with paired pictures of 2 of the talkers shown during the sequences, in order to gather speaker and dialect-dependent visual preferences. For as long as the infant remained engaged, ad

sequences of 4 blocks followed by paired comparisons were presented, up to a maximum of 5 sequences (maximum duration 13 minutes 47 seconds). Figure

presentation procedure.

Figure 4 Presentation procedure.

A. B. saliency preference 20 s 1s sequence 112 s 5 Sequences Sequence2 Stimuli presentation

The 20 blocks were divided into 5 sequences of 4 blocks so that each sequence contained 2 mixed blocks and 2 pure blocks (1 Parisian and 1 Quebecois), with the order of talkers within blocks and order of presentation of type of blocks counterbalanced across infants. Each sequence about 3.5 minutes. At the end of each sequence, infants were presented with paired pictures of 2 of the talkers shown during the sequences, in order to gather speaker

dependent visual preferences. For as long as the infant remained engaged, ad

sequences of 4 blocks followed by paired comparisons were presented, up to a maximum of 5 sequences (maximum duration 13 minutes 47 seconds). Figure 4 demonstrates the stimuli

Presentation procedure. Panel A. Sequence presentation. B. Sentence structure. C.

C. preferential looking1 20 s 15s sequence212 s ... 14s 4 sentences each 4 blocks each block1 Sentence1 Sentence2 Sentence3 Sentence4 block2 block3 block4 27 blocks so that each sequence contained 2 mixed blocks and 2 pure blocks (1 Parisian and 1 Quebecois), with the order of talkers within blocks and order of presentation of type of blocks counterbalanced across infants. Each sequence about 3.5 minutes. At the end of each sequence, infants were presented with paired pictures of 2 of the talkers shown during the sequences, in order to gather speaker- dependent visual preferences. For as long as the infant remained engaged, additional sequences of 4 blocks followed by paired comparisons were presented, up to a maximum of 5 demonstrates the stimuli

. B. Sentence structure. C. Sentence structure

28

3.3. Procedures and recordings

In the soundproof testing cabin a Dell LCD monitor (177FPc, 17'' was positioned on a table at 100 cm distance from the infants’ face. Audio speakers (Logitech speakers S-0264B) were placed under the monitor at the same distance. The sound intensity was at 70dB as measured with a sound-level meter. A digital video camera (Canon Legria FS200) was placed behind the monitor. The light in the room was kept constant.

The testing area was separated from the area where an experimenter controlled the NIRS system. The imaging device UCL optical topography system NTS2 (Everdell et al., 2005) was used for the experiment. It produced light at 670 and 850 nm and acquired images of functional activation in the brain at 10 frames per second. 16 laser diode sources (8 per wavelength) were illuminated simultaneously, and each of 8 avalanche photodiode detectors recorded light from several sources at the same time. The contribution from each source was demultiplexed in software using fast Fourier transforms. Frequency multiplexed system was used.

Optic fiber cables extended from the NIRS system to the testing booth and into a custom silicon cap on the infant's head. The cables were braided and fixed on the wall over the caregivers' shoulders.

A custom computer program recorded and analyzed the signal. Changes in oxygenated haemoglobin (oxy-HB), deoxygenated haemoglobin (deoxy-Hb) and total haemoglobin (total-Hb) in blood from temporal and front-temporal area of the brain were detected using the double channels of the NIRS system. The oxygen saturation level as well as the changes in the concentration of oxy-Hb, deoxy-Hb and total-Hb in real time were measured.

Procedure

29 infants' hands to prevent them from reaching for the NIRS cap or the glass fibers. During the experiment the caregivers were listening to masked music through head-phones.

The experimenter then placed the NIRS cap on the infants head following the 10-20 system (Sharbrough et al., 1991). The optical probes were adjusted over the left and right temporal areas, centered over the T3 and T4 positions on the left and right respectively, the cap position was fixed and tightened with velcro-straps and a chin-strap. (See Figure 5)

Figure 5 Optical probe positions.

30

4. Results

The behavioral (looking times) and neuroimaging data (NIRS activation) will be first analyzed separately, and then in relation to each other. In accordance with the goals of the study, 2 major comparisons are made: a) a comparison in the amount of activation to Pure and Mixed conditions (reflecting discrimination of dialects); b) a comparison in the amount of activation to Parisian and Quebecois conditions (reflecting dialectal preference).

4.1.

Looking time data

The time infants spent looking at the screen vs looking away during experimental trials was coded using Supercoder (Hollich, 2005). Looking times were calculated for the duration of the stimulation of each trial, that is, excluding the baseline periods. The measure of interest was

proportion looking time, calculated as the total looking time for a given trial divided by the

length of that trial. This measure was used because the trial length differed considerably throughout the experiment, therefore a proportion, rather than raw, looking time was a more accurate evaluation of infants’ behavior in response to the experimental stimuli.

The average proportion of looking times for Parisian was 0,873 (SE=0,021) and for

Quebecois 0,851 (SE=0,022). This difference was not significant. Mixed and Pure blocks also did not differ significantly in the average proportion looking time [Pure 0,862 (SE=0,012); Mixed 0,833 (SE=0,015)].

Figure 6 Looking times: Panel A: Overall Looking times by Condition (PAR, MIXED, QUE); Panel B: Looking times by Condition by Sequence

A. B.

31 6 Panel B, there was no clear preference in any sequence throughout the experiment, and large error bars suggest much individual variability. These results were confirmed through a between-subject ANOVA with Condition (Parisian, Mixed, Quebecois) as a fixed variable and Subject as a random variable. There was no effect of Condition (F(2,34)=1,421, p=0,255), but an effect of Subject F(17,37) =8,476, p=0,000). The amount and the direction of preferences differ greatly from infant to infant, see the Figure 7 below.

Figure 7 Mean looking time by condition by subject

4.2.

NIRS data

32 comparisons of interest. First of all Pure trials vs Mixed trials were compared. Then Parisian vs Quebecois, which comprised the Pure trials, were compared. Both sets of conditions were analyzed in the same way.

Combined information about oxyHb and deoxyHb gives a better understanding of the hemodynamic response and is important to report (Lloyd-Fox et al., 2009). In the present study both oxygenated hemoglobin (oxyHb) and deoxygenated hemoglobin (deoxyHb) measures were used for the analysis because both showed a good signal to noise ratio (Saito et al., 2007).

33

Figure 8 Grand average curves Pure vs Mixed conditions. Channels plotted according to the layout in Figure 5B. The x-axis shows time in seconds; the y-axis shows concentration in mmol/mm. The rectangle along the x-axis represents the time of stimulation. The continuous red and blue lines are oxyHb and deoxyHb concentrations in response to the Pure condition; the dashed magenta and cyan lines are oxyHb and deoxyHb concentrations in response to the Mixed condition.

Figure 9 Grand average curves Par vs Que condition. . Channels plotted according to the layout in Figure 5B. The x-axis shows time in seconds; the y-x-axis shows concentration in mmol/mm. The rectangle along the x-x-axis represents the time of stimulation. The continuous red and blue lines are oxyHb and deoxyHb concentrations in response to the Quebecois condition; the dashed magenta and cyan lines are oxyHb and deoxyHb concentrations in response to the Parisian condition.

Front Back Front

34

4.2.1. Pure vs Mixed comparison. Discrimination. Activation Pure and Mixed

For the Pure condition one-sample test of the average concentration in oxyHb revealed a significant increase from the baseline in channels 5, 8 on the left lateral pad, and channels 14,15,16,18 on the right lateral pad. For the Mixed condition the same channels 5,8,14,15,16,18 were activated and additionally 2,4,6,9 on the left and 11,12,17 on the right pad showed significant activation. Figure 10 shows t-maps (p<0,05, uncorrected) of the one-sample and paired t-tests on oxyHb data. Thus, overall the Mixed condition activated more channels in both hemispheres, and slightly more channels were activated on the right than on the left. Table 5 shows mean activation and standard error for oxyHb by condition by hemisphere. A comparison of Pure and Mixed conditions in a paired t-test showed that the activation to the two conditions was different in channels 2,8,9 on the left, and 14,17 on the right, see Table 6 and Figure 10. All of these channels were more activated to the Mixed condition than to the Pure condition.

Figure 10 Activation t-map oxyHb, Pure and Mixed

35

Table 5 OxyHb means and standard errors by condition by hemisphere

Condition Hemisphere Mean Std. Error of Mean

Mixed L ,071990453 ,0092980155 R ,077738000 ,0089601995 Total ,074940715 ,0064424005 Pure L ,041470397 ,0081203894 R ,041611779 ,0080553861 Total ,041542184 ,0057078881

Similar analyses were performed on deoxyHb activation concentrations. For the Pure condition one-sample test of the maximum increase in deoxyHb revealed a significant increase from the baseline in channels 2, 8 on the left lateral pad, and channels 20,15 on the right lateral pad. For the Mixed condition channels 4,5,6,8,9 on the left and 12,13,14,15,16,17,18,19 on the right showed significant activation. Figure 11 shows t-maps of the uncorrected values of the one-sample and paired t-tests on deoxyHb data. Thus, overall the Mixed condition in deoxyHb activated more channels in both hemispheres, and slightly more channels were activated on the right than on the left. Table 6 shows mean activation and standard error for deoxyHb by condition by hemisphere.

Figure 11 Activation t-map deoxyHb, Pure and Mixed

36

Table 6 DeoxyHb means and standard errors by condition by hemisphere

Condition Hemisphere Mean N Std. Error of Mean

Mixed L -,034792969 128 ,0050662424 R -,048786741 135 ,0049168222 Total -,041976084 263 ,0035480506 Pure L -,016157874 127 ,0056497403 R -,022210603 131 ,0052678783 Total -,019231159 258 ,0038556961

A comparison of Pure and Mixed conditions for deoxyHb in a paired t-test showed that the activation to the two conditions was different in channels 14,16,17,19 on the right lateral pad. All of these channels were more activated to the Mixed condition than to the Pure condition (see Figure 11).

Lateralization Pure and Mixed

Although the one way t-tests reported above suggest that there are specific areas of each hemisphere involved in processing Pure and Mixed conditions, this claim needs further statistical support. Lateralization to the conditions was assessed with a repeated measures ANOVA Condition*Hemisphere*Region.

Since one of the principle goals of the study was to explore neural bases of dialectal processing and lateralization, the next step of the analysis was to define broad regions of interest (ROI) and test activation in these areas in response to the experimental stimuli. For this purpose, 20 channels were grouped according to hemisphere (left and right) and ROI (anterior, temporal, posterior).

37

Figure 12 Channels grouped by regions (anterior, temporal, posterior)

A repeated measures ANOVA was performed with Condition (Pure; Mixed) x Hemisphere (Left; Right) and Region (Anterior; Temporal; Posterior) as factors and oxyHb concentrations as a dependent variable. The number of subjects included in this analysis was N=13, infants’ data containing missing values for the relevant contrast were excluded. The analysis showed a significant main effect of Condition (F (1,12) = 5,761, p=0,034). A pairwise comparison of the Pure and Mixed conditions showed that there was significantly more activation to the Mixed (M=0,072; SE=0,016) than to the Pure (M=0,043; SE=0,014) condition, mean difference between the conditions M=0,029; SE=0,012; p=0,034.

The main effect of region was also significant F(2,12)=12,196, p=0,001. A pairwise comparison of the 3 regions showed that there was significantly more activation in the Temporal area (M=0,095; SE=0,017) than in the Anterior (M=0,032; SE=0,014) and in the Posterior (M=0,045; SE=0,017) areas, mean difference between the Anterior and Temporal areas M=0, 063, SE=0,016, p=0,034, between Posterior and Temporal areas M=0,05, SE=0,01, p=0,001, between Anterior and Posterior areas M=0,013,SE=0,014, p>0,05. (Figure 13)

38 The main effect of region was marginally significant F(2,12)=3,616, p=0,062. A pairwise comparison of the 3 regions showed that there was significantly more activation in the Temporal area (M=-0,046; SE=0,011) than in the Anterior (M=-0,018; SE=0,006) and no difference from the Posterior (M=-0,025; SE=0,006) areas, mean difference between the Anterior and Temporal areas M=0,028, SE=0,01, p=0,05, between Posterior and Temporal areas M=0,021, SE=0,01, p=0,169, between Anterior and Posterior areas M=0,007,SE=0,008, p=1.

There was no main effect of Hemisphere and no interactions with the factor Hemisphere.

Figure 13 Activation to Pure and Mixed by Hemisphere by Region, oxyHb and deoxyHb.

A=anterior; T=temporal; P=posterior.

Summary Pure vs Mixed analysis

41 4.2.2. Parisian vs Quebecois comparison. Preference.

Activation Parisian and Quebecois

For the Parisian condition one-sample test of the average concentration in oxyHb revealed a significant increase from the baseline in channels 5, 9 on the left lateral pad, and channels 17,18 on the right lateral pad. For the Quebecois condition the channels 1,3,6,8 on the left and 13,14,15 on the right showed significant activation. Both hemispheres were active to both conditions. Table 8 shows activation to conditions by hemisphere.

Table 8 OxyHb means and standard errors by condition by hemisphere

Condition Hemisphere Mean Std. Error of Mean

Par L ,039706583 ,0105841779 R ,042607292 ,0098374555 Total ,041249514 ,0071874390 Que L ,047550011 ,0119356313 R ,058633478 ,0112260735 Total ,053414279 ,0081656046

A comparison of Parisian and Quebecois conditions in a paired t-test did not show any channels activated more to any of the two conditions. Figure 14 shows t-maps (uncorrected) of the one-sample t-tests.

For the Parisian condition one-sample test of the maximum decrease in deoxyHb revealed a significant decrease from the baseline in channel 8 on the left lateral pad, and channels 14,15,18 on the right lateral pad. For the Quebecois condition the channels 8,9 on the left and 15 on the right showed significant activation. Both hemispheres were active to both conditions although the activation was rather weak.

Table 9 DeoxyHb means and standard errors by condition by hemisphere

Condition Hemisphere Mean Std. Error of Mean

42 A comparison of Parisian and Quebecois conditions in a paired t-test with deoxyHb did not show any channels activated more to any of the two conditions.

Figure 14 shows t-maps (uncorrected) of the one-sample t-tests. Table 10 shows uncorrected, FDR corrected and Bonferroni corrected activated channels.

Figure 14 Activation t-map OxyHb and DeoxyHb

Lateralization Parisian vs Quebecois

As in the comparison between Pure and Mixed, lateralization was relevant to the question, to what extent familiarity and language-specificity are at play. A repeated measures ANOVA was performed with Condition (Parisian; Quebecois) x Hemisphere (Left; Right) and Region (Anterior; Temporal; Posterior) as factors and oxyHb concentrations as a dependent variable.

(F(1,5)=0,013, p=0,912), but a significant main

pairwise comparison of the 3 regions (Bonferroni adjusted) showed that there was significantly more activation in the Temporal area (M=0,080; SE=0,029) than in the Anterior (M=0,029; SE=0,025) and in the Posterior (M=0,026; SE=0,032) areas, mean difference between the Anterior and Temporal areas M=0,051, SE=0,013, p=0,035, between Posterior and Temporal areas M=0,053, SE=0,009, p=0,005, between Anterior and Posterior areas M=0,003,SE=0,009, p=1.

Similar results have been found in the deoxyHb data, the number of subjects without missing values for the relevant comparison being N=6. The analysis show

Condition (F(1,5)=0,128, p=0,735) n

significant main effect of Region (F(2,4)=5,599, p=0,023). A pairwise comparison of the 3 regions (Bonferroni adjusted) showed that there was significantly more activation in the Temporal area (M=-0,044; SE=0,009) than in the Anterior (M=

Posterior (M=-0,019; SE=0,011) areas, mean difference between the Anterior and Temporal areas M=0,023, SE=0,007, p=0,069, between Posterior and Temporal areas M=0,026, SE=0,008, p=0,067, between Anterior and Posterior areas M=0,003,SE=0,01, p=

Figure 15 Activation to Par and Que by Hemisphere by Region, oxyHb and deoxyHb

A=anterior; T=temporal; P=posterior.

In the repeated measures ANOVA analysis reported above the number of subjects with valid datapoints for Region*Hemisphere*Condition in all regions was only 6, however, for the comparison of Condition and Hemisphere only in the temporal area there were 13 s

contributing the data. The temporal area in the previous analyses was most activated

experimental conditions, and is most relevant to linguistic processing. Therefore, in order to (F(1,5)=0,013, p=0,912), but a significant main effect of Region (F(2,4)=16,297, p=0,001). A pairwise comparison of the 3 regions (Bonferroni adjusted) showed that there was significantly more activation in the Temporal area (M=0,080; SE=0,029) than in the Anterior (M=0,029; rior (M=0,026; SE=0,032) areas, mean difference between the Anterior and Temporal areas M=0,051, SE=0,013, p=0,035, between Posterior and Temporal areas M=0,053, SE=0,009, p=0,005, between Anterior and Posterior areas M=0,003,SE=0,009,

have been found in the deoxyHb data, the number of subjects without missing values for the relevant comparison being N=6. The analysis showed

dition (F(1,5)=0,128, p=0,735) nor an effect of Hemisphere (F(1,5)=0,505, p=0,509), but significant main effect of Region (F(2,4)=5,599, p=0,023). A pairwise comparison of the 3 regions (Bonferroni adjusted) showed that there was significantly more activation in the

0,044; SE=0,009) than in the Anterior (M=-0,021; SE=0,007

0,019; SE=0,011) areas, mean difference between the Anterior and Temporal areas M=0,023, SE=0,007, p=0,069, between Posterior and Temporal areas M=0,026, SE=0,008, p=0,067, between Anterior and Posterior areas M=0,003,SE=0,01, p=1.

vation to Par and Que by Hemisphere by Region, oxyHb and deoxyHb

P=posterior.

In the repeated measures ANOVA analysis reported above the number of subjects with valid datapoints for Region*Hemisphere*Condition in all regions was only 6, however, for the comparison of Condition and Hemisphere only in the temporal area there were 13 s

The temporal area in the previous analyses was most activated

experimental conditions, and is most relevant to linguistic processing. Therefore, in order to 43 effect of Region (F(2,4)=16,297, p=0,001). A pairwise comparison of the 3 regions (Bonferroni adjusted) showed that there was significantly more activation in the Temporal area (M=0,080; SE=0,029) than in the Anterior (M=0,029; rior (M=0,026; SE=0,032) areas, mean difference between the Anterior and Temporal areas M=0,051, SE=0,013, p=0,035, between Posterior and Temporal areas M=0,053, SE=0,009, p=0,005, between Anterior and Posterior areas M=0,003,SE=0,009,

have been found in the deoxyHb data, the number of subjects without ed neither an effect of an effect of Hemisphere (F(1,5)=0,505, p=0,509), but a significant main effect of Region (F(2,4)=5,599, p=0,023). A pairwise comparison of the 3 regions (Bonferroni adjusted) showed that there was significantly more activation in the 0,021; SE=0,007) and in the 0,019; SE=0,011) areas, mean difference between the Anterior and Temporal areas M=0,023, SE=0,007, p=0,069, between Posterior and Temporal areas M=0,026, SE=0,008,

vation to Par and Que by Hemisphere by Region, oxyHb and deoxyHb.

44 explore the activation and lateralization for the Parisian Quebecois comparison, another repeated measures ANOVA (Condition*Hemisphere) was performed only with the data from the temporal region. The results confirmed that there were no differences between the conditions, Condition (F(1,12)=0,007, p=0,943), but revealed a trend for a difference in Hemisphere (F(1,12)=3,611, p=0,082). The analysis of deoxyHb data, however, did not show a similar trend, Hemisphere (F(1,12)=0,165, p=0,692). The graphs illustrating activation by condition Figure 15 by hemisphere by region show that in oxyHb data this dominance of the right hemisphere is present in the temporal area, and in deoxyHb data, it is present in the Anterior and Posterior areas, but not in the temporal area.

Summary Parisian vs Quebecois analysis