TESTING PHONOLOGICAL REPRESENTATIONS THROUGH  
BEHAVIORAL AND ELECTROPHYSIOLOGICAL METHODS 

By 

Andrew Trotter 

 

 

 

 

 

 

 

 

A DISSERTATION 

 

Submitted to 

Michigan State University 

in partial fulfillment of the requirements 

for the degree of 

 

Linguistics—Doctor of Philosophy 

 

2019 

 

 

 

 

 

ABSTRACT 

TESTING PHONOLOGICAL REPRESENTATIONS THROUGH  
BEHAVIORAL AND ELECTROPHYSIOLOGICAL METHODS 

By 

Andrew Trotter 

Languages differ in the sounds that make up their phonemic inventories. These 

sounds, or phonemes, are represented abstractly in the mind of a speaker, making up 

the underlying representations of lexical items. Additional sounds may exist in a given 

language, surfacing via a derivative process of phonological rule application, where 

they are called allophones. There are also many speech sounds that are unrepresented 

entirely, as separate categories, within a particular language. In either case, learners of a 

second language have difficulty perceiving and producing the sounds that do not exist 

or have differing phonological status in their native language. English and Spanish 

present us with an ideal case study of these differences. Both languages contain the [d], 

[ð], and [ɾ] sounds, but they differ in how they organize them. While /d/ is a phoneme in 

both languages, in Spanish [ð] is an allophone of /d/, while /ɾ/ is a separate phoneme. 

Conversely, [ɾ] is an allophone of /d/ in English while /ð/ is a separate phoneme. There 

is a large amount of literature showing that sounds that contrast in one's language are 

more perceptible than those that do not. Boomershine et al., (2008) showed this to be 

true of the [d], [ð], and [ɾ] sounds in native Spanish and English speakers, where 

Spanish speakers more easily perceived the differences between [d]/[ɾ] than [d]/[ð]. 

 

 

English speakers, on the other hand, had difficulty distinguishing [d] from [ɾ], but no 

issue with [d] and [ð]. The studies in this dissertation extend this work by first 

replicating the results of Boomershine et al., (2008) with a group of monolingual English 

speakers as well as a group of native Spanish speakers on a forced choice perception 

task. Additionally, I add a group of native English-speaking advanced learners of 

Spanish and show that their perception of the relevant sounds is more like that of the 

native Spanish speakers on a number of behavioral metrics. In a second study, I test the 

same three speaker groups in an Electroencephalography study (EEG) using the 

mismatch negativity protocol (MMN) which has been previously shown to probe 

auditory categorical perception (Näätänen et al., 1978). If the MMN is sensitive to 

phonemic contrasts, as has been claimed, the expectation is that, similarly to the 

behavioral perception results, speakers should show larger mismatch responses to 

phonemic contrasts than to allophonic contrasts. I also explore the possibility of using 

the MMN to probe category formation in the learner group. However, while there are 

subtle differences in the EEG data for each speaker group, the results are contra 

predictions. Instead, all three language groups pattern similarly on each sound in the 

EEG study, with larger MMNs being elicited by the [ɾ]/[d] comparison than the [ð]/[d] 

comparison. This mismatch response suggests that the MMN is not probing 

phonological status but is sensitive to phonetic category. A discussion of methodology 

and the validity of using the MMN in phonology research is include

 

 

For Ozzie

 

iv 

ACKNOWLEDGEMENTS  

It has been a wonderful journey. I’ve been lucky enough to pursue my own 

research interests, I’ve had access to wonderful academic resources, I’ve sat in lectures 

by the preeminent linguists and cognitive scientists of the day, and more often than not, 

there was free pizza afterward. But graduate school takes something out of you. It 

vaporizes time, crushes personal economics, and stunts social growth. I cannot say for 

sure it was worth it. Ask me in twenty years. I don’t know anyone who’s been able to 

grit their way to a PhD. At least, whenever I read one of these things, it’s the people 

surrounding the candidate that apparently did all the heavy lifting. It has been no 

different for me. With that in mind I’d like to acknowledge those who kept me going 

and ultimately pushed me over the finish line. First and foremost, my family; my 

Mother Vicki Trotter, Father Jim Trotter, and Stepmother Janet Trotter. My cousin 

Nicholas Oas and his wife Carrie Waggoner, both fellow Spartans who kept me well fed 

and socialized throughout the years. My family’s love and encouragement were always 

felt. My best friends Stephen Wright and Nate Graham, who let me keep one foot 

outside of the academy. My lovely partner Alba, who turned out to be the most 

surprising plot twist of all. She kept me tethered to planet Earth in the final weeks of 

frantic writing. I’d like to especially recognize the members of my cohort, Joe Jalbert 

and Ho-Hsin Huang, who due in no small part to our collective misery became 

invaluable friends. Ours, I think, was a special cohort; small in number and tightly knit. 

 

v 

Joe and Ho-Hsin got hitched halfway through this journey, and I was honored to stand 

up with them on that day. They now have a beautiful little girl, and while all of our 

lives have changed dramatically from the start of graduate school, I’m thrilled to have 

had such great friends accompany me to the end. My committee members, Yen-Hwei 

Lin whose thoughtful comments on all of my work were so appreciated, Suzanne 

Wagner who should know her words of encouragement throughout my graduate career 

were always timely, Bill VanPatten whose humor always lightened the air in the room, 

especially when reverence was expected. Finally, my co-advisors Karthik Durvasula 

and Alan Beretta. Karthik, who against his own better judgement became a friend first 

and advisor second. He always knew exactly when to give praise and encouragement, 

and when to lay on the pressure. I owe him much for this. Alan has been a mentor and 

friend since I was an undergraduate. He endured missed deadlines and endless stress 

all while pulling every string he could find to support me. There is simply no way this 

dissertation could have been completed without his unwavering support, relentless 

positivity, and advocation on my behalf. I’m most proud about what we accomplished 

in the neurolinguistics lab. The lab started out as Alan and I presiding over a bunch of 

boxes filled with equipment of foreign origin. Our collective experience with EEG close 

to zero, we decided to have a go at it anyway. Now, three generations of students have 

been brought up in the lab, measuring craniums, ruining perfectly coiffed hair, and 

producing some great science along the way. We’ve been part of a wonderful 

 

vi 

international collaboration with the University of Iceland. Most importantly, the lab has 

been a welcoming place for students from all backgrounds. We’ve had students in 

syntax, semantics, phonetics, phonology, and sociolinguistics participate and produce 

by incorporating EEG into their work. All are invited to join and free to tinker as much 

or as little as they see fit. Alan has masterfully engineered the lab meetings to be a 

supportive and collaborative space where curious undergrads up to veteran PhDs 

engage one another fully. I am humbled to have been a part of it. 

 

 

 

vii 

TABLE OF CONTENTS 

LIST OF TABLES ......................................................................................................................... xi 

LIST OF FIGURES ...................................................................................................................... xii 

Chapter 1: Introduction ............................................................................................................... 1 
1.1 Description .......................................................................................................................... 3 
1.2 Outline of Dissertation ...................................................................................................... 5 

Chapter 2: Motivation and Prior Work ..................................................................................... 7 
2.1 Introduction ........................................................................................................................ 7 
2.1.1 The Status of d, ð, and ɾ in English .......................................................................... 9 
2.1.2 The Status of d, ð, and ɾ in Spanish ......................................................................... 9 
2.2 Spanish Acquisition in English Learners ..................................................................... 10 
2.2.1 Acquisition of Spanish /ɾ/ for English Speakers .................................................. 10 
2.2.2 English-speakers acquisition of Spanish spirantization ..................................... 11 
2.3 Theories of Second Language Acquisition .................................................................. 12 
2.3.1 Speech Learning Model ........................................................................................... 14 
2.3.2 Perceptual Assimilation Model .............................................................................. 15 
2.4 Mismatch Negativity....................................................................................................... 16 
2.5 Previous MMN Work ..................................................................................................... 20 
2.5.1 Cross-linguistic ......................................................................................................... 20 
2.5.2 Work on L2 Learners ............................................................................................... 23 

Chapter 3: Behavioral Task ....................................................................................................... 28 
3.1 Introduction ...................................................................................................................... 28 
3.2 Methods ............................................................................................................................ 29 
3.2.1 Participants ............................................................................................................... 31 
3.2.2 Stimuli ........................................................................................................................ 32 
3.2.3 Procedure .................................................................................................................. 42 
3.3 Results & Analysis ........................................................................................................... 43 
3.3.1 Within-language group results .............................................................................. 43 
3.3.1.1 Behavioral results for monolingual English speakers .................................. 44 
3.3.1.2 Behavioral results for native Spanish speakers ............................................. 47 
3.3.1.3 Behavioral results for advanced Spanish Learners ....................................... 49 
3.3.1.4 Summary of within language behavioral results .......................................... 50 
3.3.2 Across-language group comparisons.................................................................... 51 
3.3.2.1 Monolingual English Speakers and Native Spanish Speakers .................... 51 

 

viii 

3.3.2.2 Advanced Spanish Learners and Monolingual English Speakers .............. 52 
3.3.2.3 Advanced Spanish Learners and Native Spanish Speakers ........................ 54 
3.3.2.4 Summary of cross language results ................................................................. 55 

Chapter 4: Electrophysiological task ....................................................................................... 56 
4.1 Introduction ...................................................................................................................... 56 
4.2 Methods ............................................................................................................................ 56 
4.2.1 Participants ............................................................................................................... 56 
4.2.2 Stimuli ........................................................................................................................ 57 
4.2.3 Procedure .................................................................................................................. 57 
4.3 Results & Analysis ........................................................................................................... 59 
4.3.1 Within-language group results .............................................................................. 59 
4.3.1.1 EEG results for monolingual English speakers ............................................. 60 
4.3.1.1.1 Results for [ɾ] ............................................................................................... 61 
4.3.1.1.2 Results for [ð] .............................................................................................. 62 
4.3.1.2 EEG results for native Spanish speakers ........................................................ 63 
4.3.1.2.1 Results for [ɾ] ............................................................................................... 64 
4.3.1.2.2 Results for [ð] .............................................................................................. 65 
4.3.1.3 EEG results for advanced Spanish Learners .................................................. 66 
4.3.1.3.1 Results for [ɾ] ............................................................................................... 67 
4.3.1.3.2 Results for [ð] .............................................................................................. 67 
4.3.1.4 Summary of within language EEG results ..................................................... 68 
4.3.2 Peak latency & amplitude by segment ................................................................. 69 
4.3.2.1 Monolingual English Speakers ........................................................................ 70 
4.3.2.2 Native Spanish speakers ................................................................................... 71 
4.3.2.3 Advanced Spanish learners .............................................................................. 72 
4.3.2.4 Summary of peak latency & amplitude results ............................................. 73 
4.3.3 Across-language group results .............................................................................. 75 
4.3.3.1 English compared to Spanish ........................................................................... 77 
4.3.3.2 English compared to Learners ......................................................................... 79 
4.3.3.3 Spanish compared to Learners ......................................................................... 80 
4.3.3.4 Summary of across-language group results................................................... 82 

Chapter 5: Discussion ................................................................................................................ 84 
5.1 Interpretation of the Behavioral results ....................................................................... 84 
5.2 Interpretation of the Electrophysiological results ...................................................... 86 
5.3 Conclusion ........................................................................................................................ 97 

APPENDICES ........................................................................................................................... 100 
APPENDIX A Electrode Cluster Locations ..................................................................... 101 

 

ix 

APPENDIX B Participant Survey ...................................................................................... 103 

REFERENCES ........................................................................................................................... 106 

 

 

 

 

x 

LIST OF TABLES 

Table 1: Behavioral Results for all Language Groups and Sound Comparisons, * denotes 
significant differences between sound comparisons. ........................................................... 51 

Table 2: Within language group mismatch mean amplitudes, * denotes significant 
difference from standard. The results show a significant mismatch in all comparisons 
for all language groups, except for the [ð] in the iMMN condition for the native Spanish 
speakers. ...................................................................................................................................... 69 

Table 3: Within Language Group Peak Amplitude and Latency Results, * denotes 
significant differences between sound comparisons ............................................................ 73 

 

 

 

 

xi 

LIST OF FIGURES 

Figure 1: Underlying and surface forms for [d], [ð], and [ɾ] sounds in English and 
Spanish. ......................................................................................................................................... 8 

Figure 2: Single token and variable token MMN paradigms. The variable standard 
varies the tokens within the same phonetic category, minimizing the effect that 
acoustics plays in building the standard. ............................................................................... 19 

Figure 3: Screenshot of AXB Experiment in Praat ................................................................. 30 

Figure 4: Spectrogram of [d] Stimulus, Version 1, Female Speaker ................................... 33 

Figure 5: Spectrogram of [d] Stimulus, Version 2, Female Speaker ................................... 34 

Figure 6:Spectrogram of [d] Stimulus, Version 3, Male Speaker ........................................ 35 

Figure 7: Spectrogram of [r] Stimulus, Version 1, Female Speaker .................................... 36 

Figure 8: Spectrogram of [r] Stimulus, Version 2, Female Speaker .................................... 37 

Figure 9: Spectrogram of [r] Stimulus, Version 3, Male Speaker ........................................ 38 

Figure 10: Spectrogram of [th] Stimulus, Version 1, Female Speaker ................................ 39 

Figure 11: Spectrogram of [th] Stimulus, Version 2, Female Speaker ................................ 40 

Figure 12: Spectrogram of [th] Stimulus, Version 3, Male Speaker .................................... 41 

Figure 13: Mean % Accurate by Speaker Group and Sound Comparison ........................ 45 

Figure 14: Mean Reaction Times by Speaker Group and Sound Comparison ................. 46 

Figure 15: Percent A’ by Speaker Group and Sound Comparison ..................................... 47 

Figure 16: Averaged deviant [ɾ] and standard [d] ERP for the monolingual English 
speakers. Showing a large MMN from 300-400ms ............................................................... 62 

Figure 17: Averaged deviant [ɾ] and standard [ɾ] ERP for the monolingual English 
speakers. Showing a large iMMN from 300-400ms .............................................................. 62 

Figure 18: Averaged deviant [ð] and standard [d] ERP for the monolingual English 
speakers. Showing a MMN from 300-400ms ......................................................................... 63 

 

xii 

Figure 19: Averaged deviant [ð] and standard [ð] ERP for the monolingual English 
speakers. Showing a iMMN from 300-400ms ........................................................................ 63 

Figure 20: Averaged deviant [ɾ] and standard [d] ERP for the native Spanish speakers. 
Showing a large MMN from 300-400ms ................................................................................. 64 

Figure 21: Averaged deviant [ɾ] and standard [ɾ] ERP for the native Spanish speakers. 
Showing a large iMMN from 300-400ms ................................................................................ 65 

Figure 22: Averaged deviant [ð] and standard [d] ERP for the native Spanish speakers. 
Showing a MMN from 300-400ms ........................................................................................... 66 

Figure 23: Averaged deviant [ð] and standard [ð] ERP for the native Spanish speakers. 
Showing an iMMN from 300-400ms ....................................................................................... 66 

Figure 24: Averaged deviant [ɾ] and standard [d] ERP for the advanced Spanish 
learners. Showing a large MMN from 300-400ms ................................................................. 67 

Figure 25: Averaged deviant [ɾ] and standard [ɾ] ERP for the advanced Spanish learners. 
Showing a large iMMN from 300-400ms ................................................................................ 67 

Figure 26: Averaged deviant [ð] and standard [d] ERP for the advanced Spanish 
learners. Showing a MMN from 300-400ms ........................................................................... 68 

Figure 27: Averaged deviant [ð] and standard [ð] ERP for the advanced Spanish 
learners. Showing an iMMN from 300-400ms ....................................................................... 68 

Figure 28: Mean amplitudes for all speaker groups by comparison and configuration. A 
discussion of comparisons across language groups is available in section 4.3.3 .............. 69 

Figure 29: Peak amplitudes for each speaker group and sound comparison in both the 
classic MMN and iMMN conditions. In both conditions, the only significant differences 
in peak amplitude between the flap and fricative was for the monolingual English 
speakers. ...................................................................................................................................... 74 

Figure 30: Peak latencies for each speaker group and sound comparison in both the 
classic MMN and iMMN conditions. While the results show consistently faster peak 
latencies for the flap in all groups and in both conditions, none of the differences reach 
statistical significance. ............................................................................................................... 75 

Figure 31: Classic MMN difference waves for all speaker groups. Deviant [r] subtracted 
from standard [d]. ...................................................................................................................... 76 

 

xiii 

Figure 32: Identity MMN difference waves for all speaker groups. Deviant [r] subtracted 
from standard [r]. ....................................................................................................................... 76 

Figure 33: Classic MMN difference waves for all speaker groups. Deviant [ð] subtracted 
from standard [d]. ...................................................................................................................... 77 

Figure 34: Identity MMN difference waves for all speaker groups. Deviant [ð] 
subtracted from standard [ð]. ................................................................................................... 77 

Figure 35: Acoustic distance calculations for each sound comparison showing that 
identical sounds have relatively low acoustic distance while [ɾ] is further from [d] than 
[ð] is from [d]. ............................................................................................................................. 92 

Figure 36: EEG Channel Locations for MMN Cluster Analysis ........................................ 102 

 

 

 

 

xiv 

 

Chapter 1: Introduction 

Sounds that differ from one another in a language can be categorized as either 

contrastive, sounds that native speakers of said language can readily distinguish from 

one another and easily produce, or non-contrastive, sounds that native speakers have 

difficulty distinguishing from one another as well as producing. These contrastive 

sounds, or phonemes, constitute the phonemic inventory of the language and are 

represented in the mind of the speaker. The world’s languages can differ in the number 

and type of sounds they make use of in their phonemic inventory such that one 

language may contain phonemes that another language may lack. This property of 

language is a major contributor to the difficulty learners encounter when acquiring a 

second language (Flege, 1995) . The non-contrastive sounds can either surface within a 

particular language via a process of phonological rule application, where they are called 

allophones, or they may be unrepresented entirely within the language. In either case, 

learners have difficulty perceiving and producing the sounds that do not exist or have 

differing phonemic status in their native language. These phonemes and allophones are 

found at different levels of representation. In the generative view, there exist at least 

two levels, the underlying level and surface level. There may be multiple derivative 

iterations before arriving at the surface form, but we can think of phonemes being at the 

underlying level and allophones being surface representations (Kenstowicz, 1994; 

 

1 

McCarthy, 2010). The differences between contrastive and non-contrastive sounds in 

different languages is readily illustrated when looking at English and Spanish. There 

are slight differences in place of articulation of [d] between the languages. The [d] is 

alveolar in English and surfaces as dental in Spanish (Martínez-Celdrán et al., 2003). 

Throughout the dissertation, I’ll be referring to both simply as [d]. In English both the 

voiced alveolar stop sound /d/ as in doze, and voiced interdental fricative /ð/ as in those 

are contrastive. However, [ð] is an allophone of /d/ in Spanish, where it surfaces only 

after a spirantization rule applies (Harris, 1969). Conversely, in Spanish, the sounds /ɾ/ 

and /d/ are separate phonemes (though there are no minimal pairs, or lexical items 

whose pronunciations are distinguished solely by the two sounds), while in English [ɾ] 

is an allophone of both /d/ and /t/ (Ladefoged & Johnson, 2011). An English-speaking 

learner of Spanish must learn to actively contrast /ɾ/ and /d/, while a Spanish-speaking 

learner of English must do the same with /d/ and /ð/, a task both students often do with 

great difficulty. In this first chapter I lay out the facts concerning the sounds of interest 

in both English and Spanish and how those facts motivate the present study. I give an 

overview of the project and the experiments within, and I outline the content of the 

remaining chapters for the reader. I’ll begin with the basics of the phoneme and 

allophone distinction on which the dissertation relies. 

 

 

2 

1.1  Description 

This dissertation is concerned with how a learner’s perception of both native and 

non-native sounds changes as experience in a second language grows. The studies 

within specifically ask whether a native English-speaking learner of Spanish is able to 

form a phonemic representation for [ɾ] like that of a native Spanish speaker. They also 

ask how learning the allophonic rule for [ð] in Spanish alters the perception of [ð] in by 

the native English speaker. There is some debate on how a second language is manifest 

in the brain, particularly whether the sounds from the native language (L1) and the 

sounds from the second language (L2) are discrete systems or if they share some or all 

storage and retrieval properties (Paradis, 1985). If the sound systems in L1 and L2 

interact, then we can expect to find bi-directional influence from both languages within 

a speaker (Cook, 1992; Selinker, 1974). Furthermore, this influence should change based 

on the proficiency level in the L2.  A key question in second language learning is 

whether or not learners create new phonemic categories for L2 contrasts, or if they 

expand the phonetic space of their L1 via some memory mechanism to include the 

relevant L2 sounds, while mapping them phonologically to some formerly acquired L1 

phoneme. The experiments in this dissertation are designed to target the phonological 

level of representation in the mind of the learner, and to look for differences in 

categorical perception between L1 and L2 contrasts. In short, the goal is to observe the 

 

3 

formation of new categories of sounds in a learner and how this acquisition may affect 

the L1 representations. 

The experiments herein utilize the mismatch negativity response (MMN) which is 

an automatic response to categorical distinctions measured using 

electroencephalography (EEG).  The MMN is a negative deflection in the difference 

wave when subtracting the average ERP waveform of a “standard” stimuli from the 

average ERP waveform of a “deviant” stimuli. A typical MMN paradigm consists of 

repeated “standard” stimulus and an oddball “deviant” stimulus in a 7:1 ratio (Kilner et 

al., 2009; Näätänen & Alho, 1995, 1997; Näätänen, et al.,2004). These stimuli are usually 

categorically different in some way (e.g. different phonemes). When the “standard” is 

repeated, it builds a trace for that category in memory. Once this memory has been built 

up in the auditory cortex, it is possible to interrupt the stream of “standards” with a 

deviant, resulting in the elicitation of an MMN. The “deviant” sound is compared to the 

memory trace of the “standard” and if found to be different, will elicit a characteristic 

ERP waveform on the fronto-central electrodes about 150ms post stimulus. The ability 

to reliably detect categorical distinction in speech perception allows us to track changes 

in categorical perception of speech sounds in a learner. A detailed account of the MMN 

component, and its history in speech perception work is provided in Chapter 2. With 

that in mind, the big picture questions that this dissertation attempts to answer are: 

 

4 

•  Do we see evidence of L2 category formation in second language learners? I 

argue that we do see this in the behavioral study outlined in Chapter 3, but 

that the electrophysiological results are less informative here. 

•  Do we see evidence that perception is modulated by level of representation? 

Again, the behavioral evidence points to this being true. 

•  Are we able to use the Mismatch Negativity to probe phonemic (underlying) 

representations? The results of the electrophysiological study in Chapter 4 

suggest that the MMN probes phonetic category but is not sensitive to 

phonemic category.  

1.2 Outline of Dissertation 

Chapter 2 motivates the need for the studies in the dissertation, introduces the 

relevant sounds in each language group, and details the past work on which my studies 

are built upon. I’ll start with a descriptive account of the [d, ð, and ɾ] sounds in both 

English and Spanish. Following that, review the literature on acquisition of the Spanish 

/ɾ/ phoneme and the allophonic rule for [ð] by native English speakers, and discuss 

theories of second language acquisition that may guide our predictions. Both behavioral 

and electrophysiological data are reviewed. Both studies in this dissertation use three 

groups of speakers. There is a group of monolingual English speakers, a group of native 

Spanish speakers, and a group of native English-speaking advanced learners of 

 

5 

Spanish. Chapter 3 is dedicated to a behavioral study where these groups are subject to 

a timed discrimination task, which while not particularly novel sets a baseline of 

performance to help interpret the results of the second study, detailed in Chapter 4. The 

electroencephalography study in Chapter 4 uses the same speakers as the behavioral 

study in a passive auditory mismatch task, allowing us to see any automatic or pre-

attentive responses to the sounds of interest. The tasks, their methodologies, and results 

and statistical analyses are detailed in their respective chapters. Finally, Chapter 5 

provides the conclusion, discussion, and directions for future work. It contains an 

interpretation of the behavioral results that clearly show the advanced Spanish learners 

becoming more like the native Spanish speakers in their perception. I also discuss the 

EEG results which do not necessarily align with the behavioral results. I find an 

unexpected mismatch response to the allophonic [d] - [ɾ] comparison among the 

monolingual English speakers, along with a stronger mismatch response on both the 

allophonic and phonemic comparisons in the advanced Spanish learners. In fact, both 

comparisons pattern similarly for all three language groups in the EEG experiment, 

while exhibiting clear differences in the behavioral study. Comparisons to similar EEG 

work and possible explanations are discussed. 

 

 

 

6 

Chapter 2: Motivation and Prior Work 

In this chapter, I will first discuss the status of the three sounds focused on in this 

dissertation in both English and Spanish ([d], [ɾ], and [ð]). There is asymmetry in how 

these languages classify the sounds in terms of phonemes or allophones, allowing us to 

study the effect allophony has on perception. I discuss the difficulties English speakers 

encounter when learning to correctly utilize both the flap and fricative in Spanish, and I 

then turn to discussing two current models of second language acquisition and what 

predictions each makes for the current study. Finally, I discuss how the mismatch 

negativity experimental paradigm can potentially bring further clarity to this research 

by manipulating the level of representation that a listener taps into during an 

experiment and highlight previous electrophysiological work on the acquisition of L2 

sounds.  

2.1 Introduction 

For native speakers of English, learning Spanish involves altering their 

representations of two sounds in their L1 (these are not the only sounds that differ 

between English and Spanish, but are the ones studied in this dissertation). The studies 

herein are focused on the discriminability of the [d], [ð], and [ɾ] sounds in English and 

Spanish. There are slight differences in place of articulation of /d/ between the 

 

7 

languages. The /d/ is alveolar in English and dental in Spanish (Martínez-Celdrán et al., 

2003). Spanish contrasts the voiced alveolar stop /d/ and the alveolar flap /ɾ/ 

phonemically while in English, [ɾ] is an allophone of both /d/ and /t/ (Harris, 1969; 

Ladefoged & Johnson, 2011). Figure 1 illustrates the underlying and surface levels of 

representation for the [d], [ð], and [ɾ] sounds in English and Spanish. 

 

Figure 1: Underlying and surface forms for [d], [ð], and [ɾ] sounds in English 
and Spanish. 

English speakers learning L2 Spanish must learn to contrast /ɾ/ with /d/. A key 

question that this dissertation attempts to address is whether a learner can form a new 

category for /ɾ/ or if proper /ɾ/ perception by an L2 Spanish speaker does not rely on the 

creation of a phonological representation of /ɾ/. English speakers also must learn the 

spirantization rule in Spanish, whereby stops become fricatives after sonorants. In 

Spanish, /d/ surfaces as the voiced dental fricative [ð] in the same context that /d/ 

surfaces as [ɾ] in English further complicating acquiring this allophonic rule for Spanish 

learners. /ð/ already exists as a phoneme in English so learners of L2 Spanish must learn 

to use the allophonic representation simultaneously with the phonemic representation.  

 

8 

2.1.1 The Status of d, ð, and ɾ in English 

In English, /d/ and /ð/ are different phonemes while [ɾ] is an allophone of /d/ and 

/t/. The voiced dental fricative contrasts with the voiced alveolar stop as in those [ðoz] 

and doze [doz]. However, the alveolar flap [ɾ], only surfaces as an allophone of /d/ and 

/t/ post sonorant and before unstressed vowels as in writer or reader (Kahn, 1976; Zue & 

Laferriere, 2005).This flapping rule is extremely productive in American English, being 

realized over 90% of the time the environment allows it (Herd et al., 2010; Patterson & 

Connine, 2001). 

2.1.2 The Status of d, ð, and ɾ in Spanish 

In Spanish, unlike in English, /d/ and /ɾ/ contrast phonemically. There are no 

minimal pairs because /d/ is spirantized to [ð] after sonorants. As with flapping in 

English, Spanish spirantization, where voiced obstruents /b,d,g/ are spirantized to 

[ß,ð,ɣ], is a highly productive phonological rule in Spanish, with spirantization of /d/ 

occurring 99% of the time (Waltmunson, 2005). This dissertation is concerned with the 

contrast of the [d] and [ð] sounds, which are allophonic in Spanish. /d/ undergoes 

spirantization to [ð] unless it follows a liquid or nasal as in (1a)-(1b). 

(1) 

 

a.  /hada/  

->  

[aða]    

‘fairy’ 

b.  /falda/  

->  

[falda]  

‘skirt’ 

9 

Since /ð/ contrasts with /d/ in English and since Spanish spirantization occurs in 

a similar environment as English flapping, this difference in how [ð, d, ɾ] are 

phonemically categorized in the two languages may cause difficulties for English 

learners of Spanish. The following portion of the chapter details the difficulties reported 

for English speakers acquiring the /d-ɾ/ contrast in Spanish. 

2.2 Spanish Acquisition in English Learners 

This section details English-speaking Spanish learners’ acquisition of /ɾ/ as a 

meaningful contrast as well as the spirantization rule for post-sonorant Spanish voiced 

stops. 

 2.2.1 Acquisition of Spanish /ɾ/ for English Speakers 

English speakers produce [ɾ] as an allophone of /t/ and /d/ after sonorants and 

before unstressed syllables, but must learn to contrast it with /d/ in Spanish. Face (2006) 

showed that at least in terms of production, this contrast is very difficult for beginner 

and even intermediate native English-speaking learners of Spanish. In a reading task, 41 

native speakers of English were placed into two groups: one group of intermediate 

Spanish learners who had taken four semesters of college-level Spanish, and one 

advanced learner group who were Spanish majors and enrolled in upper-level Spanish 

courses. Their production was compared to a group of native Spanish speakers. The 

native Spanish speakers produced the [ɾ] correctly 92% of the time, while the 

 

10 

intermediate Spanish learner group did so only 49% of the time. The advanced Spanish 

learners approached the native Spanish group with 79% correct production of [ɾ]. When 

errors happened, both learner groups tended to replace the [ɾ] with the American 

English retroflex [ɹ]. 

2.2.2 English-speakers acquisition of Spanish spirantization  

English-speaking L2 learners of Spanish have to acquire a number of phonetic 

and phonemic differences in order to sound like a native Spanish speaker. One of these 

differences is Spanish spirantization of the voiced obstruents /b, d, g/. Previous studies 

on English-speaking novice and intermediate learners have shown that production of 

the spirtantized forms is rarely achieved. Zampini (1994) found that voiced obstruents 

were spirantized only 32% of the time. Of the three /b, d, g/, the /d/ to [ð] spirantization 

was the least likely to be achieved at less than 10% of the expected instances. In her 

study, when failing to produce the fricative, the voiced stop was always produced 

instead. Zampini claimed that the existence of the phoneme /ð/ in English is what 

causes the /d/ to [ð] spirantization to lag behind /b/ to [ß] and /g/ to [ɣ] in learner’s 

production. Herd (2011), however, offered that it is possible that the learners were 

producing the alveolar flap [ɾ] intervocalically instead of [ð] since that is the 

environment in which it is already produced in English. While Zampini’s studies are 

informative for new learners of Spanish, they do not include advanced learners or 

 

11 

bilinguals in their experiments. A study by Face & Menke (2009) looked at the Spanish 

production of 53 English speaking L2 Spanish learners broken into three groups based 

on Spanish instruction level. While they found similar results to Zampini among the 

less experienced groups, their advanced learner group showed significant development 

toward native-like rates of spirantization. The most advanced group had an average of 

thirteen years of Spanish instruction and fifteen months of time in a Spanish-speaking 

country. These learners produced spirants in 80% of the expected contexts. This 

suggests that this spirantization rule can be learned in adulthood with enough training.  

2.3 Theories of Second Language Acquisition 

This difficulty in acquiring L2 contrasts and correctly applying phonological 

rules is of course not limited to English speakers learning Spanish. It is well established 

that listeners perceive differences in sounds that are present in their native language 

more easily than those that are not (Best, et al., 1988; Dupoux et al., 1997; Strange, 1995; 

Strange & Jenkins, 1978, amongst others). Errors in pronunciation of non-native words 

by non-native speakers of a language are the hallmark of this difficulty. For many L2 

learners, these errors persist throughout life, even when they are conscious of them, and 

continuously corrected. One of the key debates in the second language literature, and 

an important question asked in this dissertation, is whether or not a L2 learner can form 

non-native phonemic categories late in life. The Critical Period Hypothesis (Birdsong & 

 

12 

Molis, 2001; Johnson & Newport, 1989; Lenneberg, 1967) states that a learner must be 

exposed to language stimuli before a critical age (sometime before the completion of 

puberty) in order to attain native-like performance in that language. The strong version 

of the hypothesis would claim that L2 learners, beyond the critical period, have no hope 

of becoming native-like in their second language, at best approximating a native 

speaker. Advocates of this view point to the strong correlation between foreign-

accented (FA) speech and either age of arrival (AOA) or first exposure to the L2, 

something Flege et al., (1995) showed can account for 94% of the variance in FA rating 

among L2 learners. However, in their view, AOA correlates with a plethora of other 

variables on L2 input and usage that confound the data and the claim that failure for the 

majority of L2 learners to achieve native-like speech production and perception is due 

to neurological development. Most theories of second language acquisition reject a 

strong critical period hypothesis, and there is evidence that a decline in language 

learning ability may be tied to factors other than the onset of puberty. In a massive 

study of native and L2 English speakers, Hartshorne et al. (2018), had over half a 

million subjects take an English grammar quiz online along with a demographic 

questionnaire. Test results along with demographic information and linguistic 

background information were used to create means for each subject’s native language 

and dialect. This data was used to train a computational model for second language 

learning. The results suggest that L2 acquisition ability remains intact at least into early 

 

13 

adulthood before beginning to decline around age 18. They claim that the existence of a 

critical period for second language acquisition, if it exists, cannot be due to changes in 

neuronal development or hormones in childhood or adolescents. Instead, it may be due 

to changing social pressures in adulthood, neuronal changes later in life, or increased 

interference from the L1. 

 2.3.1 Speech Learning Model  

Flege’s Speech Learning Model (SLM) (Flege, 1991, 1995, 2007, 2011) leaves open 

the possibility for learners to create new categories. The ease of forming L2 categories 

rests on dissimilarity of the L2 phoneme from existing L1 phonemes. Flege’s Speech 

Learning Model rejects the critical period hypothesis. While “earlier is better” is 

uncontroversial in any model of L2 learning, the SLM claims that there is no neuro-

biological mechanism that closes off language learning, and that the same acquisition 

mechanism children use to learn their L1 is preserved across the lifespan (Flege, 2007). 

For the SLM the type, frequency, and quality of input can also explain the bulk of the 

variance in FA speech outcomes. When it comes to the acquisition of nonnative 

contrasts, the SLM makes predictions about the ease of learning those contrasts based 

on how similar (rather dissimilar) the L2 sounds is from an existing L1 sounds. The 

SLM predicts that initially, an L2 learner will substitute the unfamiliar L2 sound with 

the closest (in terms of phonetic similarity) L1 sound. As experience in the L2 grows, 

 

14 

these sounds will gradually dissimilate from each other until a new category is formed. 

SLM additionally predicts that there may be change in the original L1 category as the 

creation of the new L2 category crowds the phonetic space, causing a dispersive shift 

that might perceptibly alter the L1 category, making it markedly different from that of a 

monolingual. As it relates to the sounds studied in this dissertation, the SLM would 

predict that learners would initially replace [ɾ] with a perceived similar L1 sound (/ɹ/), 

but that advanced learners would eventually be able to create a category for /ɾ/. 

 2.3.2 Perceptual Assimilation Model 

Like the Speech Learning Model, the Perceptual Assimilation Model (Best, 1994, 

1995; Best et al., 2001; Best & Tyler, 2007) relates the ease with which a learner will 

acquire an L2 sound by how readily the learner is able to assimilate that sound to an 

existing L1 sound. Under this framework, non-native sounds are perceived in one of 

three ways: They can be assimilated to an existing L1 category as a good exemplar, 

perceived as a speech sound but not a good exemplar of an existing L1 sound, or 

perceived as non-speech. If the L2 sound is perceived as a good exemplar of an existing 

L1 sound, the learner will have difficulty creating a new category for that sound and the 

likelihood of learning the L2 sound will be small. However, if the L2 sound is not 

readily perceived as an exemplar of an existing category, the learner should be able to 

perceive the difference and potentially form a new category for that sound. PAM is a 

 

15 

Direct Realist theory (Fowler, 1986, 1989, 2005), which assume that phonological 

primitives are articulatory gestures that have physicality (Browman & Goldstein, 1986, 

1989). This differs from the SLM, which assumes that representations of made up of 

acoustic cues. PAM also makes explicit distinctions between phonetic and phonological 

categories, whereas the SLM does not. For PAM, category creation would require a shift 

in the L1 exemplar in order for the target L2 sound to not assimilate to that L1 category.  

Under both models, the possibility for adults to acquire new categories exists, 

though the underlying mechanism for doing so differs. While the experiments in this 

dissertation are not designed to test specific models of second language acquisition, I 

mention these because the ability for adults to form new L2 categories and their relation 

to the L1 are central to the predictions tested.  

2.4 Mismatch Negativity 

The Mismatch Negativity (MMN) is a component of the auditory event-related brain 

potential (ERP) that responds to change (Näätänen, et al., 1978; Näätänen & Michie, 

1979). Though originally discovered using encephalography (EEG), equivalent 

components have been found using other brain recording methods such as 

magnetoencephalography (MEG) (Csépe et al., 1992; Hari et al., 1984), functional 

magnetic resonance imaging (fMRI) (Celsis et al., 1999; Molholm et al., 2005), and 

positron emission tomography (PET) (Tervaniemi et al., 2000). The “classic” MMN 

 

16 

manifests as a negative deflection in the difference wave obtained by subtracting the 

ERP generated by a frequent “standard” stimulus from that of an odd-ball “deviant” 

stimulus. It is generally seen on the frontocentral electrodes, peaking at approximately 

150ms post onset of the deviant stimulus. The MMN component overlaps with the so-

called N1-P2 complex, which is a negative deflection in the ERP occurring 100ms post 

stimulus followed by a positive-going deflection around 200ms post stimulus ( 

Näätänen et al., 1989; Sams et al., 1985; Tiitinen et al., 1994). These components are so-

called “obligatories”, regularly occurring with any stimulus, but can be differentiated 

from the MMN by looking for reversed polarity of the MMN on the mastoid or 

temporal electrodes (Deacon et al., 2000). Generation of the MMN depends upon the 

creation of a “standard” in short-term memory. This memory trace is ephemeral, so if 

the “standard” stimuli are presented in too low a number or with too long an 

interstimulus interval (ISI), it cannot be maintained, and no MMN will be generated 

upon presentation of the “deviant” stimulus (Mäntysalo & Näätänen, 1987). The 

presentation of the “standard” stimuli can also affect the MMN response. The more 

consecutive “standard” stimuli that are presented, the larger the response to the 

“deviant” will be. Likewise, the shorter the ISI between standards, the larger the MMN 

amplitude (Sabri & Campbell, 2001). Thus, the ratio of standards to deviants in the 

experiment presentation matters. A typical MMN paradigm consists of repeated 

standard stimulus and an oddball deviant stimulus in a 7:1 ratio (Kilner et al., 2009; 

 

17 

Näätänen et al., 2007, for reviews). While the MMN is strongest on frontocentral 

electrodes in EEG, the cortical source of neuronal activity cannot be straightforwardly 

inferred from scalp-distributions due to interference from the skull. Other recording 

methods have been informative here, with intracranial, fMRI, and PET recordings 

suggesting that the MMN has its origin in the auditory cortex, specifically Broadman 

areas 22, 41, and 42. (Baudena et al., 1995; Halgren et al., 1995; Halgren et al., 1998). 

There are a couple of ways to set up an MMN experiment but in general, the simplest 

way to do this is with two tokens. A single standard and deviant. A variation that has 

been used increasingly in language related studies is to use a variable standard (Eulitz 

& Lahiri, 2004; Phillips et al., 2000). That is to use multiple similar tokens (so a few 

different versions of [sa]) and then a few different versions of [za]. You still present the 

standard and the deviants in the same 7:1 ratio, but because the tokens aren’t identical, 

the standard is less reliant on acoustic similarity and hopefully more on phonetic 

category. There are also two ways you measure the response. I use both in my 

dissertation. First is the classic MMN, where you simply subtract the voltage of the 

deviant from the voltage of the standard. The second way, sometimes called the identity 

MMN, is based on presentation context. You subtract the voltage of on stimulus when it 

is presented as a deviant, from the voltage of that same stimulus when presented as a 

standard. The idea is that the iMMN controls for the mismatch response generated 

simply from the differences between the two stimuli (Pulvermüller & Shtyrov, 2006; 

 

18 

Tavabi et al., 2009). Figure 2 shows the single and variable standard paradigms as well 

as illustrating the difference between the classic and identity MMN. In the first two 

blocks of stimuli, there is only a single deviant along with a single standard token 

presented seven times the number of deviant presentations. In the second block, the 

variable standard, multiple different tokens of both the standard and deviant sounds 

are presented in the same 7:1 ratio of standard to deviant presentations. In the “Classic” 

MMN measurement, you simply subtract the mean voltage of the deviants in a block 

from the mean voltages of the standards in a block. Alternatively, with the “iMMN” 

you subtract the mean voltage of a sound when it is presented as a deviant in one block, 

from the mean voltage of that same sound when it is presented as a standard in another 

block.  

Figure 2: Single token and variable token MMN paradigms. The variable 
standard varies the tokens within the same phonetic category, minimizing the 
effect that acoustics plays in building the standard. 

 

 

19 

2.5 Previous MMN Work 

2.5.1 Cross-linguistic 

 Näätänen et al., (1997) showed that responses to linguistic stimuli in a mismatch 

paradigm were both distinct from non-linguistic acoustic stimuli and language specific. 

In an experiment run on both Finnish and Estonian subjects, they found that in the 

frequent presence of a prototypical Finnish phoneme /e/ ), an infrequent stimulus 

(MMN deviant) of either another prototypical Finnish phoneme /ö/ or the Estonian 

phoneme not found in Finnish /õ/, had different MMN responses. In the Finns, the 

MMN response to the Finnish deviant /ö/ was enhanced in amplitude when compared 

to the response to the Estonian phoneme. Crucially, when Estonian speakers were 

subjected to the same experiment, they did not show a difference in amplitude for 

deviant /ö/ when compared to deviant /õ/. They claim that this language-dependent 

MMN effect shows the existence of language-specific neural traces to phoneme 

representations. They also found that, the MMN elicited in the Finns by the vowel 

existing in Finnish, was localized to the left posterior auditory cortex while the MMN 

generated by the Estonian vowel were bilaterally generated. This suggests 

subcomponents of the MMN that are sensitive to the differences between purely 

acoustic and phonological information. 

 

20 

Phillips et al. (2000)  in a magneto-encephelography (MEG) study, showed that 

the MMN can be used to show evidence for abstract phonological representations. The 

MMN has a magnetic analog that can be elicited using MEG as opposed to EEG. While 

this is sometimes still called the MMN in the literature, the MEG specific component is 

commonly called the MMF for Mismatch Field (Hari et al., 1984). Where previous 

mismatch studies tended to use a single exemplar token for their standard and deviant 

sound, (Phillips et al., 2000) varied their standard tokens such that the construction of a 

standard would be dependent on recognizing the phonological features of the segment. 

In this case, they presented English speaking subjects with multiple tokens of /d/ and /t/ 

where the VOT was varied within each category. Because the standard is no longer 

acoustically the same stimulus in this design, any memory trace must be based on the 

phonological category /d/ or /t/ rather than simple acoustics (Aulanko et al., 1993). They 

observed an MMN in this varied standards presentation. In a second phonetic 

experiment, they increased the VOT of all their stimuli by 20ms, causing half of the 

previous standards to fall just above the categorical perception threshold for voicing 

and into the /t/ category. So, while the stimuli were still presented in a ‘many to one’ 

ratio in terms of relative VOT length, the number of /t/ and /d/ stimuli were equal. No 

MMN was observed in the second experiment, suggesting that no standard was able to 

be created in the listener based on VOT length, and that the MMN is sensitive to 

phonological category. This varied standard technique has since been utilized to probe 

 

21 

even more granular phonological information by others. Eulitz & Lahiri, (2004) and 

Hestvik & Durvasula, (2016) have used this paradigm to test predictions for 

underspecification theory. 

Kazanina et al., (2006) expanded on this work to show that not only are there 

language-specific responses to speech sounds, but that those responses can differ based 

on the functionality of the sounds and contrasts within the language. Russian speakers 

contrast [d] and [t] phonemically across a voice-onset time (VOT) continuum. While 

Korean speakers also produce [t] and [d] across a similar VOT continuum, they do not 

contrast these sounds phonemically. In Korean, [d] and [t] map to a single phoneme 

realized as [d] between voiced segments and [t] elsewhere. While it had been well 

established in behavioral studies that discrimination of allophonic contrasts is poorer 

than in phonemic ones, and thus functional status matters in speech perception, these 

behavioral differences could have arisen via higher-level conscious processes. 

Establishing electrophysical results via the MMN would show that these functional 

differences in speech sounds with a language were available preattentively.  

Miglietta et al., (2013) ran an MMN study on phonemic perception versus 

allophonic perception in Italian speakers. In their study, they found distinct MMNs 

were elicited for both phonemic and allophonic contrasts. They looked at the allophonic 

variation [ε-e] and phonemic contrast [e-i]. They found no amplitude differences 

between the MMN responses to these contrasts, but did find significant differences in 

 

22 

peak latency. The phonemic contrasts had earlier peak latencies than did the allophonic 

alternations, suggesting easier processing of the phonemic contrasts.  

2.5.2 Work on L2 Learners 

Comparatively little work has been done utilizing the MMN in second language 

acquisition research. Of the MMN work that has been done, the majority focuses on 

bilingual populations or change detection in children within different language 

environments (Cheour et al., 1998; Cheour, et al., 2001; Cheour, et al, 2002; Peltola et al.,  

2005; Shestakova et al., 2003). Relatively little has been done in the way of adult second 

language learning, though there is evidence that the MMN can probe category 

formation in adult second language learners.  

Winkler et al., (1999) was one of the earliest studies to provide evidence for 

category formation in second language learners. In their study, native speakers of 

Finnish, native speakers of Hungarian, and native Hungarians who were second 

language learners of Finnish were tested on the /e/-/ä/ contrast. These sounds are 

allophones in Hungarian but distinct in Finnish. The naïve Hungarian speakers showed 

no MMN for this contrast and similarly could not distinguish the sounds behaviorally. 

Native Finns showed a distinct MMN for this contrast and easily distinguished the 

sounds in a behavioral task. Importantly, the fluent L2 Finnish speakers showed both 

MMN and behavioral results similar to the Native Finns. 

 

23 

Peltola et al. (2003) achieved slightly different results, showing that the MMN 

amplitude does vary within second language learners as fluency changes. Their study 

looked at vowel discrimination in native Finnish learners of English. Native English 

speakers, Native Finnish speakers, and Native Finns with advanced English training 

were all played blocks of Finnish /i/ - /e/, English /ɪ/ - /e/, English /i/ - /ɪ/, and English /e/ 

- /ɪ/ in an MMN paradigm. There is no phonemic /ɪ/ vowel (as in “bit” in English), in 

Finnish. The results as shown in the difference waves suggest two things. First, it does 

not appear that the “advanced” learners of English group perform like native English 

speakers on English contrasts. The authors assume this to mean that phonological 

category formation has not yet taken place in these learners. Second, these learners 

display a significantly lower amplitude MMN for the Finnish vowel contrast than do 

the native Finnish speakers with no English training. The claim here is that perhaps the 

learning process has somehow degraded their perception of L1 constants. Others have 

reported bidirectional influence of L2 learning on the L1 in production experiments 

(Mackay & Flege, 2004; Yeni-Komshian et al., 2000) but to my knowledge this is the first 

instance of neurophysiological evidence of that influence on perception. 

Herd (2011) investigated whether English speaking learners of Spanish could be 

trained to both perceive and produce the [d], [ɾ], and [r] contrasts in Spanish. She tested 

participants on each contrast before and after a number of training protocols. While 

primarily concerned with understanding which types of training produce the most 

 

24 

effective results, she used both traditional behavioral identification and production 

tasks along with electroencephalography. This ERP data can show whether there is 

automatic processing of the contrasts taking place, and if the training protocols are 

enough to show evidence for phonemic category formation in the learners. She ran an 

MMN task on two groups of speakers. One native Spanish speaking group and another 

group of English speakers who had undergone perception training on the Spanish 

contrasts in question.  For the English trainees, the MMN task was administered twice, 

both pre and post training. As one would predict, the native Spanish speakers showed a 

clear MMN response to the [d]-[ɾ] comparison. Surprisingly, the English trainees also 

showed a significant MMN to the [d]-[ɾ] comparison in both the pre and post training 

environment. While the mismatch response to [ɾ] in the pre-training English speakers is 

not predicted, it is possible that the learners had already created that category. My EEG 

study finds similar results, with a large MMN response to /ɾ/ among Spanish learners. 

Additionally, and more surprisingly, the monolingual English speakers in my study 

(not tested by Herd), also showed a large MMN response to the [d]-[ɾ] comparison. This 

makes it harder to interpret the results of Herd and my learners as having created a new 

phonemic category for /ɾ/. 

Barrios (2013) and (Barrios, et al., 2016) ran a very similar study to the second 

study in this dissertation. In their study, they looked at L1-Spanish speakers who were 

advanced L2-English learners as opposed to the advanced L2-Spanish learners here. 

 

25 

Their study also utilized magnetoencephalography (MEG) rather than 

electroencephalography (EEG) as in the current study. Due to these similarities, the 

predictions in their study and mine are more or less aligned and thus comparing results 

is informative.  They found no statistical significance in the differences in groups, but 

noted that the Mismatch Field (MMF, the MMN correlate generated by MEG, Hari et al., 

1984) amplitudes elicited were contra expectations. In their study, English speakers 

both showed similar MMF responses for both the [d]-[ɾ] comparison and the [d]-[ð]  

comparison, while Spanish speakers had similar MMF responses to English speakers on 

the [d]-[ɾ] as well as an unexpected contrastive MMF for [d]-[ð]. These MEG responses 

largely mirror the results of my EEG study detailed in Chapter 4. 

In this chapter I discussed the sounds of interest and talked about how they are 

represented differently in both English and Spanish. I summarized the acquisition work 

on English-speaking learners of Spanish for the flap and spirantization rule and 

discussed the major theoretical framework in second language acquisition. I introduced 

the mismatch negativity paradigm and how it may be used to probe categorical 

perception, specifically phonetic and phonemic categories. I reviewed the literature on 

the MMN as it pertains to linguistic contrasts as well as the MMN literature on second 

language learners and bilinguals. In summary, there is strong evidence that within a 

language, the MMN is sensitive to phonological status in speech sounds. In some cases, 

MMN responses tend to be stronger for sounds that are contrastive in the language than 

 

26 

for those that are not. Allophony seems to play a part in the mismatch response. In 

some cases, it results in attenuation of the amplitude of the MMN, where in others it 

seems to affect the peak latency of the response. This dissertation contributes to the 

growing body of work on advanced L2 learners and bilinguals and their 

neurophysiological responses to native and non-native contrasts. Here, the literature 

has been considerably less convergent, highlighting the need for additional work in this 

area. In the next chapter, I will discuss a psycholinguistic study testing the 

discrimination of [d], [ɾ], and [ð] among three groups of speakers. Naïve English 

speakers, Native Spanish speakers, and English-speaking advanced learners of Spanish. 

In Chapter 4, I will present a mismatch negativity study on those same subjects to see if 

any insights on their speech perception can be gained during the pre-attentive processes 

that the behavioral study in Chapter 3 is unable to tease out. 

 

 

 

27 

Chapter 3: Behavioral Task 

3.1 Introduction 

The following chapter discusses in detail the behavioral task given to each of the 

subjects in my three language groups. This chapter will establish the perceptual facts as 

they fall out from the AXB discrimination task. This will create a set of comparison 

results to check against the electrophysiological results presented in Chapter 4. One of 

the reasons to pursue neurolinguistic methods such as the MMN with tools like EEG or 

MEG lies in the belief that these tools and methods can reveal insights about language 

processing that the relatively high latency of behavioral studies cannot. Second, 

replicating prior results on the perceptual sensitivities between the relevant sounds for 

the three language groups is important as both a sanity check and to ensure that our 

participant populations are reasonably categorized. There are robust results on the 

perceptibility of contrastive versus non-contrastive sounds within speakers (Beddor & 

Strange, 2005; Boomershine, et al., 2008), but here I claim that the simple presence or 

absence of a sound from a speaker’s phonemic inventory is not sufficient to predict 

discriminability. Like Boomershine et al. (2008), I claim that the more nuanced 

phonological status of the sounds is what determines perceptibility. That is, both 

allophony and contrast need to be taken into account. To that end, replication of the 

 

28 

Boomershine results (with a slightly different discrimination task) will give needed 

weight to this claim. I present the within-language results for each language group 

(monolingual English speakers, native Spanish speakers, and advanced Spanish 

learners) in turn. I then look at the cross-language comparisons.  I will present the 

behavioral results for each group in three different ways. First, a simple accuracy rating. 

This is straightforwardly the percentage of trials in which each subject correctly 

identified the target sound in the AXB task. Second, a sensitivity measurement, in this 

case A’, which will control for any response bias in the subjects. I use A’ instead of the 

more widely used d’ (Macmillan & Creelman, 2004) as an estimate of sensitivity because 

the A’ calculation, while not truly non-parametric, is less reliant on assumptions about 

the distribution of the signal and noise curves (Donaldson, 1993; Zhang & Mueller, 

2005; Stanislaw & Todorov, 1999). Finally, I will also present an analysis of the reaction 

times from the offset of the last stimulus in the AXB task.  

3.2 Methods  

Every participant took part in a behavioral task immediately prior to the EEG 

experiment. This task was designed to get a baseline perceptual discriminability of the 

three crucial sounds in the dissertation for each language group. The goal here was to 

both replicate the findings of past perception studies to ensure the ‘goodness’ of my 

stimuli as well as to collect accuracy and reaction time data for the three groups to use 

 

29 

as comparison with the EEG results. An AXB task was used to force attention to the 

sounds. An AXB task is a common discrimination task in which three stimuli are played 

sequentially. Subjects are instructed to attend to the second stimuli, X, and then indicate 

which of the first (A) or third (B) stimulus is most similar to the target X. There is some 

evidence to suggest that AXB tasks produce more reliable discrimination results when 

compared to the more standard ABX paradigms, as ABX tests tend to show strong 

baises toward “B = X” (Gerrits & Schouten, 2004; Schouten, et al., 2003). Subjects were 

told they were being timed and to make their choices quickly. All behavioral 

experimentation took place in the Michigan State University Psycholinguistics lab in 

East Lansing, MI. An image of the AXB task given to participants in Praat is below in 

Figure 3. 

Figure 3: Screenshot of AXB Experiment in Praat 

30 

 

 

3.2.1 Participants  

All subjects were Michigan State University students and were placed into one of 

three groups. One group of monolingual English speakers who had no formal 

instruction in Spanish beyond high school (N=13, 6 men, 7 women, average age=23 

years) and had never spent time in a Spanish speaking country. One of the 13 

monolingual English subjects is omitted from the behavioral results due to investigator 

error. A second group of native Spanish speakers (N=11, 4 men, 7 women, average 

age=26 years) who were L2 speakers of English. One speaker was from Spain, one was 

Argentinian, and the rest spoke Mexican or Caribbean Spanish. All speakers’ native 

dialects follow the patterns described in Chapter 1 for all sounds of interest. The 

language background survey given to participants is available in APPENDIX B. A final 

group of native English speakers with advanced training in Spanish (N=14, 6 men, 8 

women, average age=24 years) were included. These speakers had a minimum of 6 

consecutive semesters of college Spanish instruction including courses on Spanish 

phonology. Subjects in this group also had spent a minimum of 3 months in a Spanish 

speaking country. Additionally, subjects in this group all stated intentions of using 

Spanish in their daily lives either at work or abroad. Most were training to become 

Spanish teachers. These speakers self-reported their average Spanish ability as 8 on a 0-

10 scale, with 0 meaning no ability and 10 being ‘native-like’. In all groups, no subject 

 

31 

had any history of hearing or language disorders. All participants were right handed. 

All participants had normal or corrected-to-normal vision.  

3.2.2 Stimuli  

Materials consisted of three tokens each of the following VCV sequences: [ada], 

[aɾa], [aða]. All tokens were recorded by native Greek speakers, two female tokens and 

one male token were used. These tokens were graciously supplied by Amanda 

Boomershine and her colleagues and are a subset of the stimuli used in their 2008 paper 

which compared discriminability in English and Spanish speakers in two different tasks 

designed to emphasize either phonetic or phonemic contrasts (Boomershine et al., 2008). 

The stimuli used here come from their 3rd and 4th experiments, where they tested 

subjects using tokens produced by native Greek speakers. The Greek language contrasts 

all three sounds of interest phonemically and, most importantly, has a phonemic 

contrast between all three sounds intervocalically. Using the Greek tokens guarantees 

that no influence from English or Spanish exists in the stimuli that could potentially 

bias listeners. It also makes for natural sounding stimuli as English speakers would 

have difficulty producing [aɾa] naturally in isolation, while Spanish speakers would 

have similar problems producing natural tokens of [ada] because of the spirantization 

process in the language. The nine total tokens (three each of [ada], [aɾa], and [aða] were 

normalized in Praat to be a comfortable and consistent volume of 70dBA. To control the 

 

32 

amplitude across tokens and speakers, the peak amplitude was equated for each of the 

tokens. To further control for any coarticulatory or cueing effects, the stimuli were 

further modified in Praat to have equal vowel length (150 +/- 10ms) on either side of the 

target consonant. To maintain a natural sounding token without pops or distortion, 

vowel fragments were spliced out peak-to-peak at zero crossings until the vowel 

lengths of all stimuli were roughly equal. The consonants under study are naturally of 

varying length and so consonantal fragments were spliced out to create stimuli of the 

same total length throughout the study (350 +/- 10ms). For detailed spectrograms of 

each stimulus, see the Figures 4-12 below. 

Figure 4: Spectrogram of [d] Stimulus, Version 1, Female Speaker 

 

33 

 

Figure 5: Spectrogram of [d] Stimulus, Version 2, Female Speaker  

 

 

34 

Figure 6:Spectrogram of [d] Stimulus, Version 3, Male Speaker 

 

 

35 

Figure 7: Spectrogram of [r] Stimulus, Version 1, Female Speaker 

 

 

36 

Figure 8: Spectrogram of [r] Stimulus, Version 2, Female Speaker  

 

 

37 

Figure 9: Spectrogram of [r] Stimulus, Version 3, Male Speaker 

 

 

38 

Figure 10: Spectrogram of [th] Stimulus, Version 1, Female Speaker  

 

 

39 

Figure 11: Spectrogram of [th] Stimulus, Version 2, Female Speaker  

 

 

40 

Figure 12: Spectrogram of [th] Stimulus, Version 3, Male Speaker  

 

 

 

 

41 

3.2.3 Procedure 

Subjects performed an AXB discrimination task. Each subject heard three tokens 

in a row and were asked to choose whether the first sound they heard or third sound 

they heard was most similar to the second sound in the sequence. Stimuli were real 

speech tokens of [d] [ð] and [ɾ] produced by Greek speakers and detailed in the above 

section. The middle (2nd) sound was always the male version of the token while the 

peripheral sounds were always a version of the female tokens. For example, a subject 

might hear [ada] followed by [ada] followed by [aɾa]. They would then choose 

(correctly) that the first sound is more similar to the second. “First” and “Third” were 

represented as large text boxes on the screen and participants would indicate their 

choice with a mouse. For each trial, the inter-stimulus interval (ISI) between sounds was 

200ms. Between trials, the inter-trial interval was an additional 200ms. There was no 

time limit for the subject to choose an answer (though reaction times greater than 

3500ms were excluded from the final analysis). The experiment software used was Praat 

(Boersma & Weenink, 2013). Stimuli were provided via closed-back over-ear 

headphones (Sennheiser HD202). Both accuracy and reaction times were recorded. All 

possible combinations of sounds and orders were given randomly to each subject for a 

total of 72 trials each. This took each participant on average about 10 minutes to 

complete.  

 

42 

3.3 Results & Analysis 

This section provides details about the results of the AXB discrimination task for 

each of the three language groups as well as the three different measures taken as 

indices of discriminability. Percentage accuracy, A’ estimates, and mean reaction times 

were calculated for each participant and then compared both within language groups 

and across language groups. A note on the statistical tests used throughout Chapters 3 

and 4; T-tests are one-tailed tests in comparisons that had a priori predictions about the 

direction of the outcome, while two-tailed tests were used otherwise. In practice, this 

means that for the monolingual English group and the native Spanish group, the 

expectation is that the phonemic contrast would be more easily discriminated than the 

allophonic contrast in the respective native language. Within language groups, the tests 

are paired. Across languages groups, because the groups vary in size, they are 

unpaired.  

3.3.1 Within-language group results 

This section gives the results and statistical analyses for the AXB task comparing 

discriminability of different pairs of sounds within a single language group. That is, 

how readily do monolingual English speakers discriminate [d] from [ɾ] compared to 

their ability to distinguish [d] from [ð] We begin with said group. 

 

43 

3.3.1.1 Behavioral results for monolingual English speakers 

It is known that pairs of sounds that contrast phonemically in a speaker’s 

language are perceptually more distinct than sounds which are not phonemically 

contrastive. Given prior results on the differences in discriminability of phonemic and 

allophonic contrasts, we predicted that the monolingual English speakers would 

perform poorly on this task when the target pair was [d]/[ɾ], but would be significantly 

more adept at discriminating the sounds when the target was [d]/[ð].  

The behavioral results for English speakers are robust. As predicted, English 

speakers easily discriminate between [d] and [ð] which contrast phonemically in 

English. In contrast, the allophonic pair [d] and [ɾ] are much more difficult for the 

English speakers to discriminate between. This difference is borne out on all three 

dimensions: percentage accuracy, A’ values, and reaction times. The mean percentage 

accuracy for [d]/[ð] was 94.1% which was significantly higher than the mean percentage 

accuracy for [d]/[ɾ] of 68.1% (t(11)=-3.84, p=0.002). Mean A’ scores for the [d]/[ð] contrast 

were significantly higher (M=0.969) than those for the [d]/[ɾ] contrast (M=.710) (t(11)=-

3.43, p=0.003). In addition to being more sensitive to the [d]/[ð] contrast, English 

speakers also responded significantly faster during the AXB task on the phonemic pair 

(M=2.92s) than the allophonic [d]/[ɾ] pair (M=3.23s) (t(11)=-3.02, p=0.006).  

 

44 

To summarize, the monolingual English speakers performed significantly better 

when discriminating the phonemic [d]/[ð] pair than the allophonic [d]/[ɾ] pair along 

every dimension measured. Figures 13-15 below show the behavioral results for all of 

the language groups and should be referred to in order to visualize the results 

discussed in the following sections.  

Figure 13: Mean % Accurate by Speaker Group and Sound Comparison 

 

 

45 

Figure 14: Mean Reaction Times by Speaker Group and Sound Comparison 

 

 

46 

 

Figure 15: Percent A’ by Speaker Group and Sound Comparison 

 

3.3.1.2 Behavioral results for native Spanish speakers 

For the Spanish speakers, the prediction was the mirror of the English results. 

That is, the phonemic pair of [d]/[ɾ] would be much easier for the Spanish speakers to 

discriminate than the allophonic pair [d]/[ð]. These predictions were largely borne out. 

For the behavioral AXB task the native Spanish speakers had a mean accuracy of 91.2% 

when discriminating between [d] and [ɾ]. This high accuracy was expected given that 

[d] and [ɾ] are phonemically contrastive in Spanish. Although the group performed well 

on this contrast, it is noteworthy that they were slightly less accurate at this phonemic 

 

47 

contrast than the monolingual English speakers were on their phonemic contrast of 

[d]/[ð]. This could be simply due to differences in acoustic similarity between the 

contrasts but could also be due to the fact that unlike [d] and [ð] in English, [d] and [ɾ] 

while phonemes in Spanish, do not appear in the same environment. [d] is spirantized 

between vowels and [ɾ] is never word initial in Spanish. As expected, the native Spanish 

speakers were much less accurate on the allophonic pair [d] and [ð]. On this contrast 

they had a mean accuracy of 71.4% (t(10)=2.58, p=0.02). Despite this large difference in 

accuracy, sensitivity scores were surprisingly not significantly different between the 

contrasts for Spanish speakers. Mean A’ for the phonemic [d]/[ɾ] pair were 0.87, while 

the allophonic pair had a mean A’ score of 0.85 (t(10)=0.22, p=0.415). This group also 

showed reaction time differences between the contrasts. Reaction times to the phonemic 

[d]/[ɾ] pair were faster (M=3.13s) than the allophonic [d]/[ð] pair (M=3.50s). This result 

was marginally significant (t(10)=-1.57, p=0.07).  

These results suggest, like those from the monolingual English speakers, that 

phonemically contrastive sounds are more readily discriminated between than 

allophonic sounds. While both the contrastive [d]/[ɾ] pair and the allophonic [d]/[ð] had 

similar A’ sensitivity estimates, both percentage accuracy and reaction times were 

improved on the contrastive pair. It is important to note that unlike the monolingual 

English speakers, these native Spanish speakers have extensive experience in a second 

language, namely English. So, it is perhaps unsurprising that their discriminability 

 

48 

measurements pattern slightly different from expectations when compared to the 

monolingual English speakers who have no L2 experience at all, let alone any in 

Spanish. Still, the trend toward the contrastive sounds having an advantage in 

discriminability is clear.  

3.3.1.3 Behavioral results for advanced Spanish Learners 

Predictions for the advanced Spanish learners are less straightforward. There are 

a several possibilities to discuss. If the learners have been able to acquire both the 

phonemic representation of [ɾ] and the allophonic variant of [ð] through their 

experience in L2 Spanish, then one would expect their results to look similarly to the 

native Spanish speaking group. However, if they have been unable to acquire ‘native-

like’ representations of the Spanish sounds, then they should pattern more like the 

monolingual English speakers. A third possibility is some type of intermediate 

performance between the two groups, pointing toward acquisition-in-progress (Peltola 

et al., 2005). For accuracy rates, the learners mean for [d] - [ɾ] was 81.2% compared to a 

mean of 80.6% for [d]/[ð] (t(13)=0.2, p=0.84). In terms of sensitivity, the mean A’ score 

for [d]/[ɾ] was 0.874. The mean A’ score for the [d] - [ð] comparison was 0.864. These 

were not significantly different (t(13)=0.148, p=0.442). Reaction times for the two 

contrasts were also not significantly different. The mean reaction time for the [d] - [ɾ] 

comparison was 3.02s while the mean reaction time for the [d] - [ð] comparison was 

3.03s (t(13)=-0.028, p=0.489).  

 

49 

3.3.1.4 Summary of within language behavioral results 

The previous behavioral results suggest that for both the monolingual English 

speaking group and the native Spanish speaking group, the straightforward predictions 

that L1 phonological status prevails in perception. That is, that contrasts that are 

phonemic in L1 are more readily perceived than contrasts that are allophonic in the L1. 

This is shown by statistical differences in three different measures of discriminability; 

percentage correct, A’ sensitivity measures, and reaction time measurements. For the 

advanced learners of Spanish however, behaviorally these contrasts are not significantly 

different on all three of these measures, perhaps indicating some intermediate level of 

acquisition. We will now turn to the between language comparisons which will give us 

more insight into this learning trajectory by the advanced Spanish learners. Table 1 

summarizes these results. An asterisk indicates that the difference between the [d] - [ð] 

and [d] - [ɾ] comparisons was statistically significant.  

 

 

 

50 

[d]/[ɾ] 
68.1* 

[d]/[ɾ] 
3.23* 

[d]/[ð] 
94.1* 

A’  
[d]/[ð] 
.969* 

Mean Reaction Time (s)  % Correct 
[d]/[ð] 
2.92* 

 
 
Monolingual 
English 
Speakers 
Native Spanish 
Speakers 
Advanced 
Spanish 
Learners 
Table 1: Behavioral Results for all Language Groups and Sound Comparisons , * 
denotes significant differences between sound comparisons.  

91.2* 

.850 

.870 

.864 

.874 

[d]/[ɾ] 
.710* 

3.50* 

3.13* 

71.4* 

3.03 

3.02 

81.2 

80.6 

3.3.2 Across-language group comparisons  

In the previous sections, I detailed the results from within each speaker group. 

The following sections will look at how the different language groups compare to one 

another on the measures of mean amplitude, peak amplitude, and peak latency. Refer to 

Figures 11-13 above to visualize these results. 

3.3.2.1 Monolingual English Speakers and Native Spanish Speakers 

I will first compare the monolingual English speakers to the native Spanish 

speakers. There are clear predictions about how these groups should differ 

behaviorally, that is the phonemically contrastive pair in the respective L1 should be 

advantageous for perception. The within language behavioral results discussed above 

give us no reason to believe otherwise. For the [d]/[ɾ] contrast, which is allophonic in 

English but phonemic in Spanish, the English speakers had a mean A’ score of 0.71 

while the Spanish speakers were predictably more sensitive to this contrast with 

A’=0.87. This result was marginally significant (t(21)=-1.450, p=0.08).  The monolingual 

 

51 

English speakers were also much less accurate on this contrast (68%) than the native 

Spanish speakers (91.2%) (t(21)=-2.81, p=.006). The monolingual English speakers had a 

mean reaction time to this contrast of 3.23s, slower than the native Spanish speakers at 

3.13s (t(21)=0.58, p=0.28). Turning to the opposite case, the [d]/[ð] contrast, English 

speakers had a mean A’ score of 0.969 while the Spanish speakers had an A’ score of 

0.852. This result was significant (t(21)=1.711, p=0.05). The monolingual English 

speakers had significantly higher accuracy when discriminating this contrast at 94.1% 

compared to 71.4% for the native Spanish speakers (t(21)=3.96, p=0.0005).  For reaction 

times, the monolingual English speakers had a mean reaction time of 3.29s compared to 

3.50s for the native Spanish speakers. (t(21)=-2.13, p=.02). 

3.3.2.2 Advanced Spanish Learners and Monolingual English Speakers 

The more interesting case is of course the comparisons of our English and 

Spanish groups with the advanced learner group. We’ve seen that the advanced 

Spanish learners showed no statistical differences between the [d]/[ɾ] or [d]/[ð] 

comparisons. At the outset of the chapter, I laid out three possible outcomes for the 

learner group. They could behave like the monolingual English speakers, or they could 

behave like the native Spanish speakers, or something else entirely, presumably an 

intermediate level of perception between the other groups. I will first compare them to 

the monolingual English speakers. For the [d] - [ɾ] comparison, the learners have 

significantly higher A’ scores (M=0.874) than do the English speakers (M=0.710) (t(24)=-

 

52 

1.810, p=0.04). The advanced Spanish learners are also more accurate on this contrast 

(81.2%) than the monolingual English speakers (68%) (t(24)=-1.76, p=004). The advanced 

Spanish learners had a mean reaction time of 3.02s compared to 3.23s for the 

monolingual English speakers. On the [d]/[ð] comparison, for which the learners need 

to acquire an allophonic variant for a sound they already contrast phonemically, A’ 

scores average 0.864 compared to the English speakers 0.969. This is significantly lower 

than their monolingual peers (t(24)=2.722,p=0.006).   In terms of accuracy, the learners 

correctly discriminated [d] from [ð] 80.6% of the time compared with 94.1% for the 

monolingual English speakers. This difference was statistically significant (t(24)=2.96, 

p=0.007). The advanced Spanish speakers had a mean reaction time of 3.03s while the 

monolingual English speakers had a mean reaction time of 3.29s. 

Based on A’ sensitivity estimates and the accuracy hit rates, it looks as though 

the advanced learners of Spanish have improved significantly at discriminating [d] 

from [ɾ] over their monolingual counterparts. This could signal acquisition of a 

phonemic representation of [ɾ] which would make these sounds clearly contrastive for 

them. More puzzling is the advanced learners of Spanish significant decline in 

discriminability between [d] and [ð] when compared to the monolingual English 

speakers. This suggests a level of L2 interference perhaps caused by having to maintain 

multiple representations of the same sound. While unexpected, work on Finnish 

 

53 

bilinguals have shown similar degradation in discriminability of L1 contrasts for vowels 

(Peltola et al., 2003) 

3.3.2.3 Advanced Spanish Learners and Native Spanish Speakers 

When compared to the native Spanish speakers, the results look quite different. 

On the [d]/[ɾ] contrast, which is phonemically contrastive in the target language, the 

learners have a mean A’ score of 0.874, while the native Spanish speakers score 

A’=0.872. These scores are not statistically different (t(23)=0.025, p=0.49). The native 

Spanish speakers were correct 84.7% of the time while the advanced Spanish learners 

had an 81.9% accuracy rate (t(23)=0.36, p=0.36). Mean reaction time for the advanced 

Spanish learners was 3.02s compared to 3.13s for the native Spanish speakers. Similarly, 

the [d]/[ð] comparison, which is allophonic in the target language, the learners have a 

mean A’ score of 0.864. The native Spanish speakers A’ score of 0.852 is not significantly 

different than that of the learners (t(23)=-0.151, p=0.44). The advanced Spanish learners 

had an accuracy of 80.6%, while the native Spanish speakers had an accuracy of 71.4%. 

This difference was not significant (t(23)=1.28, p=0.21).  The advanced Spanish learners 

had a mean reaction time of 3.03s compared to 3.50s for the native Spanish speakers. 

Based on A’ scores and accuracy rates, the performance of the advanced learners 

of Spanish is not clearly distinguishable from that of the native Spanish speakers on 

both contrasts. As mentioned in section 3.3.1.3, part of this could be due to the fact that 

 

54 

the native Spanish speaking group is not monolingual. Their experience in English 

could have increased their ability to discriminate [d] from [ð]. What is clear is that the 

advance Spanish learners are able to reliably discriminate the Spanish phonemic 

contrast between [d] and [ɾ]. 

3.3.2.4 Summary of cross language results 

The cross-language group results from the behavioral discrimination task seem 

to suggest that as expected, the advanced Spanish learners are becoming more ‘native 

Spanish-like’ in the perception of both contrasts. Although I used a slightly different 

discrimination task with my subjects (AXB vs ABX), the results here replicate those of 

the English and Spanish speakers from Boomershine et al., (2008), while adding the 

results of the advance Spanish learners. 

 

 

 

55 

Chapter 4: Electrophysiological task 

4.1 Introduction 

Immediately following the AXB task all subjects went to prepare for the EEG 

experiment. All EEG experiments took place in the MSU neurolinguistics lab. 

4.2 Methods 

4.2.1 Participants 

All subjects from the behavioral experiment in Chapter 3 participated in the 

MMN task. To recall, there was one group of monolingual English speakers who had no 

formal instruction in Spanish beyond high school (N=13, 6 men , 7 women, average 

age=23 years) and had never spent time in a Spanish speaking country. A second group 

of native Spanish speakers (N=11, 4 men, 7 women, average age=26 years) who were L2 

speakers of English. A final group of native English speakers with advanced training in 

Spanish (N=14, 6 men, 8 women, average age=24 years) were included. These speakers 

had a minimum of 6 consecutive semesters of college Spanish instruction including 

courses on Spanish phonology. Subjects in this group also had spent a minimum of 3 

months in a Spanish speaking country. 

 

56 

4.2.2 Stimuli 

Stimuli in the MMN task were identical to those in the behavioral task described 

above in Chapter 3 save for the exclusion of the male voiced tokens. The MMN 

procedure used only the female versions of the token (2 versions of each sound) to 

control for acoustic differences between male and female speakers. 

4.2.3 Procedure 

Subjects are briefly introduced to the EEG equipment and the preparation and 

procedure is explained to them. Subjects begin by having a head measurement taken to 

ensure a properly sized electrode cap is selected. This is so the electrode placement is 

consistent between all subjects regardless of morphological differences. Once a cap is 

fitted, preparation of the individual electrodes begins. A small amount of electrolyte gel 

is used between the electrode and the subjects scalp (onestep clear gel). The skin under 

each electrode is then gently abraded with a wooden applicator to create good 

conductivity between the scalp and electrode via the gel. To ensure the best signal, all 

impedances on each electrode are kept below 5kΩ. Once the subject is prepared for 

recording, they are instructed to remain still for the duration of the experiment. Because 

the MMN is a so-called preattentive process, subjects do not need to directly attend to 

the stimuli being presented (Näätänen et al., 1993; Sussman et al., 2003). To distract 

subjects (and to keep them awake), participants were made to watch a movie while the 

 

57 

MMN series of sounds were presented to them. The movie was on silent with no 

subtitles. Subjects were each played two separate series of sounds through headphones 

(Sennheiser HD203) in the MMN paradigm, where a series of similar stimuli are 

presented in order to build a ‘standard’ representation in the listener’s mind, followed 

by the presentation of a ‘deviant’ stimulus that differs in at least one dimension from 

the standards. One series consisted of standard [ada] and deviant [aða], followed by the 

reverse presentation. The second series consisted of standard [ada] and deviant [aɾa], 

followed by the reverse presentation. The order of these series was alternated for every 

other subject. The electroencephalogram (EEG) was recorded from 32 sintered Ag/AgCl 

electrodes held in place on the scalp by an elastic cap (GND WaveGuard 64 Electrode 

cap; Advanced Neuro Technology BV., Enschede, The Netherlands). It was amplified 

using a Full-band EEG DC Amplifier (Advanced Neuro Technology), with a 256 Hz 

sampling rate (and a bandpass filter of 0.01 to 30 Hz which was applied offline). 

Electrode impedances were kept below 5 kΩ. Averaged mastoids were used as 

reference (Dien, 1998; Luck, 2005). All signals were recorded continuously and captured 

within Advanced Neuro Technology’s Advanced Source Analysis software (Zanow & 

Knösche, 2004). A 100ms pre-stimulus baseline was subtracted from all epochs. Epochs 

were defined as -100ms – 700ms in relation to the stimulus onset. A voltage rejection 

criterion of +/- 75 µV was applied to the data such that any epoch with samples beyond 

this threshold were removed from analysis. This resulted in a loss of approximately 6% 

 

58 

of total trials. Individual subject epochs were then averaged using an 800ms averaging 

window. Grand average difference waves were then constructed from the resulting 

subject averages. All post-processing of the recording data along with measurements 

and ERP plots were done with ERPLAB, an ERP plug-in for the MATLAB toolbox 

ERPLAB (Delorme & Makeig, 2004; Lopez-Calderon & Luck, 2014; The Mathworks Inc, 

2018). 

4.3 Results & Analysis 

4.3.1 Within-language group results 

This section gives the results and statistical analyses for the EEG task comparing 

discriminability of different pairs of sounds within a single language group. For the 

MMN task, I will present the results in two ways. I will first compare the average ERP 

amplitude over the 300-400ms time window of each of the crucial sound pairs against 

one another in the classic MMN configuration. This shows the amplitude of the deviant 

response when compared to the amplitude of the response to the standard stimulus. 

Following that, I will show each of the crucial sounds in turn in the identity or iMMN 

configuration, with the average amplitude of the deviant stimulus being compared to 

the average amplitude of the standard stimulus of the same sound. It is important to 

show both comparisons given that there have been significant differences in how 

MMNs have been interpreted in the speech perception literature depending on the 

 

59 

configuration used in analysis. In the next section, I will present the cross-language 

comparisons, but here I will discuss each language in turn.  

4.3.1.1 EEG results for monolingual English speakers 

In this section I will describe several analyses of the ERP obtained from the 

monolingual English speakers on the MMN task. The MMN literature presents a 

number of ways one may determine the electrodes of interest, but the mismatch 

negativity is well known to present in the fronto-central region of the scalp. For this 

reason, it is unnecessary to perform whole-head analysis by region-of-interest. Indeed, 

it is not uncommon to perform single electrode analysis when doing MMN studies in 

EEG. Here, I choose instead to do a cluster analysis, averaging values from four fronto-

central electrodes on the basis of visual inspection of mastoid inversions, which 

distinguishes the MMN from other fronto-central components (Schrager, 1998) . The 

contiguous electrodes Fz, FC1, FC2, and Cz form on the vertex area of the scalp and 

were the most visually similar in the grand-averaged waveform to the inverted 

mastoids M1 and M2.  

The first analysis I present is an averaged amplitude between two ERP latencies. 

The window of analysis was chosen as 300-400ms post stimulus onset because the VCV 

stimuli used in my experiment contain initial vowels of approximately 150ms in length, 

making 300-400ms post onset 150-250ms post consonant onset. This is both the 

 

60 

traditional location of the MMN (part of the N1-P2 complex), and where clear 

mismatches occurred in my data by visual inspection (Kilner et al., 2009; Näätänen et 

al., 2007). Before looking at the results of individual mismatch contrasts in the EEG task, 

I present the results of the omnibus ANOVA. In the 300-400ms time window there was 

a main effect of condition, standard vs. deviant (F(1,72)=9.54, p=.003). There was also an 

interaction effect of segment ([d], [ɾ], [ð]) and condition (F(2,72)=4.28, p=.02). I will now 

turn to the post-hoc tests for the various contrasts. 

4.3.1.1.1 Results for [ɾ] 

I will first look at the mismatch between the allophonic flap consonant [ɾ] and 

[d]. Based on previous work, I predicted that the mismatch for [ɾ] in English would be 

attenuated due to its non-contractiveness with the phoneme /d/. It was also shown that 

these same subjects displayed poor discriminability for the [d]/[ɾ] contrast in the 

previous behavioral task. This predicted result was not borne out. The monolingual 

English speakers displayed a prominent classic MMN response for [ɾ] upon visual 

inspection (Figure 16). When compared to a standard [d], the deviant [ɾ] was 

significantly more negative in the 300-400ms time period (t(12)=5.23, p=1.1e-4). The 

iMMN analysis yielded similar results (Figure 17). [ɾ] as a deviant was significantly 

more negative in the 300-400ms time window than [ɾ] as a standard (t(12)=5.31, p=9.1e-

5).  

 

61 

Figure 16: Averaged deviant [ɾ] and standard [d] ERP for the monolingual 
English speakers. Showing a large MMN from 300-400ms   

 

 

Figure 17: Averaged deviant [ɾ] and standard [ɾ] ERP for the monolingual 
English speakers. Showing a large iMMN from 300-400ms   

 

4.3.1.1.2 Results for [ð] 

Turning now to the mismatch responses for [ð]. Because [ð] contrasts 

phonemically with [d] in English, and behavioral results showed excellent 

discriminability between the two sounds, I predicted a prominent MMN response from 

the monolingual English subjects. This prediction was borne out. In the 300-400ms time 

window, there was a significant classic MMN for deviant [ð] when compared to 

 

62 

standard [d] (t(12)=4.43, p=.0004) (Figure 18). The iMMN response was even stronger 

when comparing deviant [ð] to standard [ð] (t(12)=3.56, p=.002) (Figure 19). One thing to 

note when comparing the mismatch responses for [ð] and [ɾ] is that the [ɾ] components 

while greater in amplitude are not nearly as sustained as those generated by [ð].  

Figure 18: Averaged deviant [ð] and standard [d] ERP for the monolingual 
English speakers. Showing a MMN from 300-400ms   

 

 

Figure 19: Averaged deviant [ð] and standard [ð] ERP for the monolingual 
English speakers. Showing a iMMN from 300-400ms   

4.3.1.2 EEG results for native Spanish speakers 

In this section, I will detail the electrophysiological results of the native Spanish 

speaking group on the mismatch task. This task was identical to the one given to the 

 

63 

monolingual English speakers, and as in the previous section, I will discuss the classic 

MMN and iMNN results in the 300-400ms time windows for both [ɾ] and [ð]. 

4.3.1.2.1 Results for [ɾ] 

In Spanish, unlike in English, [ɾ] contrasts phonemically with [d] and so a clear 

mismatch response for [ɾ] is expected. In the 300-400ms time window, ANOVA 

revealed a main effect for condition, standard vs. deviant (F(1,60)=10.65, p=.001). Post 

hoc tests revealed a significant classic MMN for deviant [ɾ] when compared to standard 

[d] (t(10)=3.42, p=.003) (Figure 20). There was also a significant iMMN for deviant [ɾ] 

when compared to standard [ɾ] (t(10)=3.87, p=.002) (Figure 21).  

Figure 20: Averaged deviant [ɾ] and standard [d] ERP for the native Spanish 
speakers. Showing a large MMN from 300-400ms   

 

 

64 

 

Figure 21: Averaged deviant [ɾ] and standard [ɾ] ERP for the native Spanish 
speakers. Showing a large iMMN from 300-400ms   

4.3.1.2.2 Results for [ð] 

For [ð], which is an allophone of [d] in Spanish and behavioral results showed 

that native Spanish speakers find the distinction between [ð] and [d] significantly 

harder to perceive than the phonemic [ɾ]/[d] contrast. For these reasons, an attenuated 

mismatch response is expected for [ð] in this group. This result is borne out in the 

MMN analysis. In the 300-400ms time window, the classic MMN for deviant [ð] when 

compared to standard [d] showed a significant mismatch (t(10)=3.78, p=.002) (Figure 

22). However, in the iMMN analysis where deviant [ð] is compared to standard [ð], 

there was only a marginal difference (t(10)= 1.53, p=.08) (Figure 23).  

 

65 

Figure 22: Averaged deviant [ð] and standard [d] ERP for the native Spanish 
speakers. Showing a MMN from 300-400ms   

 

 

Figure 23: Averaged deviant [ð] and standard [ð] ERP for the native Spanish 
speakers. Showing an iMMN from 300-400ms   

 

4.3.1.3 EEG results for advanced Spanish Learners 

In the 300-400ms window the advanced Spanish learners had a main effect for 

condition (F(1,78)=11.03, p=.001) and marginally significant interaction effect of segment 

by condition (F(2,78)=2.41, p=.09).  

 

66 

4.3.1.3.1 Results for [ɾ] 

There was a significant classic MMN for [ɾ] (t(13)=3.73, p=.003) (Figure 24). There 

was also a significant iMMN (t(13)=3.67, p=.003) (Figure 25). 

Figure 24: Averaged deviant [ɾ] and standard [d] ERP for the advanced Spanish 
learners. Showing a large MMN from 300-400ms   

 

 

Figure 25: Averaged deviant [ɾ] and standard [ɾ] ERP for the advanced Spanish 
learners. Showing a large iMMN from 300-400ms   

4.3.1.3.2 Results for [ð] 

There was a significant classic MMN for [ð] (t(13)=6.04, p=4.15e-5) (Figure 26). 

There was also a significant iMMN for [ð] (t(13)=4.29, p=8.7e-4) (Figure 27). 

 

67 

Figure 26: Averaged deviant [ð] and standard [d] ERP for the advanced Spanish 
learners. Showing a MMN from 300-400ms   

 

 

Figure 27: Averaged deviant [ð] and standard [ð] ERP for the advanced Spanish 
learners. Showing an iMMN from 300-400ms   

4.3.1.4 Summary of within language EEG results 

Contra expectations, all language groups showed significant mismatch responses 

for both comparisons. This was true of both the classic MMN and identity iMNN 

conditions. The only exception was the iMMN for the [ð] in the native Spanish speakers. 

This difference was marginally significant. Table 2 shows all the mean amplitude 

mismatch responses for each language group by sound and condition. Additional 

results are visualized in Figure 28. 

 

68 

iMMN [ɾ] 
-2.07 μV * 

MMN [ð] 
-1.26 μV * 

-2.07 μV * 

-2.01 μV * 

MMN [ɾ] 
-1.81 μV * 

Speaker Group 
Monolingual 
English 
Native English 
learner of Spanish 
Native Spanish 
Table 2: Within language group mismatch mean amplitudes, * denotes 
significant difference from standard. The results show a significant mismatch in 
all comparisons for all language groups, except for the [ð] in the iMMN 
condition for the native Spanish speakers. 

iMMN [ð] 
-1.51 μV * 

-1.15 μV * 

-0.99 μV * 

-1.13 μV * 

-1.36 μV 

-1.79 μV * 

-2.02 μV * 

 

Figure 28: Mean amplitudes for all speaker groups by comparison and 
configuration. A discussion of comparisons across language groups is available 
in section 4.3.3 

4.3.2 Peak latency & amplitude by segment 

The standard practice in mismatch negativity studies is to compare the mean 

voltages within your chosen time window between conditions. This is what was done 

 

69 

in the previous sections in both the classic MMN and iMMN comparisons. However, 

there are other ways in which one can measure the mismatch response, and a few have 

been used in prior cross-linguistic work. Peak latency, that is the time post-stimulus that 

the negative deflection of the mismatch response is at its voltage maximum, has been 

argued to be sensitive to allophony (Miglietta et al., 2013). In the following sections I 

will present the peak latency and peak amplitude results for each language group, by 

segment and phonological status.  

4.3.2.1 Monolingual English Speakers 

The ANOVA on peak amplitude revealed main effects for phonological status, 

allophonic vs. phonemic contrast (F(1,96)=5.41 , p=.002) and by condition. (F(3,96)=13.41, 

p=<.001). No interaction between status and condition was observed.  

The ANOVA on peak latency revealed a main effect for phonological status, 

allophonic vs. phonemic contrast (F(1,96)=5.65, p=.02), but no other main effects or 

interaction effects. 

In terms of peak amplitude, in the MMN configuration, the allophonic contrast 

flap (-3.68μV), was significantly more negative than the phonemic contrast fricative (-

2.45μV) (t(12)=-3.40, p=.005). For the iMMN difference wave, the results were similar 

with the flap being more negative (-3.46μV) than the fricative (-2.27μV) (t(12)=-1.80, 

p=.06). 

 

70 

Turning to peak latency, for the classic MMN, while the allophonic contrast flap 

(330.5ms) was on average faster than the phonemic contrast fricative (339.2ms), this 

difference did not reach significance (t(12)=-1.16, p=.27). In the iMMN configuration, 

peak latencies were even less different with flap (336.8ms) and fricative (336.3ms) 

showing no difference (t(12)=0.09, p=.93). 

4.3.2.2 Native Spanish speakers 

The ANOVA on peak amplitude revealed a main effect for condition. 

(F(1,80)=7.12 , p=.0002) but not for status. No interaction between status and condition 

was observed.  

The ANOVA on peak latency revealed a main effect for phonological status, 

allophonic vs. phonemic contrast (F(1,80)=9.53, p=.002), and a main effect for condition 

(F(3,80)=3.75, p=.014). There was no interaction effect between status and condition. 

For peak amplitude, in the classic MMN configuration, the phonetic contrast flap 

was more negative (-3.15 μV) than the allophonic contrast fricative (-2.55 μV), however 

this difference was not significant. (t(10)=-0.96, p=.23). The iMMN was similar, with flap 

(-2.14 μV) not significantly different from fricative (-2.53 μV) (t(10)=-0.03, p=.98). 

For peak latency, in the classic MMN configuration, the phonetic contrast flap 

was faster (331.32ms) than the allophonic contrast fricative (339.48ms), however this 

difference was not significant. (t(10)=-0.81, p=.43). In the iMMN, flap had a peak latency 

 

71 

of 340.55ms and fricative a peak latency of 340.91ms. These were not different (t(10)=-

0.03, p=.98). 

4.3.2.3 Advanced Spanish learners 

The ANOVA on peak amplitude revealed a main effect for condition, 

(F(1,104)=9.84 , p<.0001) but not for status. No interaction between status and condition 

was observed.  

The ANOVA on peak latency revealed a main effect for phonological status, 

allophonic vs. phonemic contrast (F(1,104)=9.15, p=.003), but no main effect for 

condition. There was no interaction effect between status and condition. 

For peak amplitude, in the classic MMN, the allophonic contrast flap was more 

negative (-4.18 μV) than the phonemic contrast fricative (-3.27 μV), however this 

difference was not significant (t(13)=-1.43, p=.17). For the iMMN, flap was again more 

negative (-3.68 μV) than fricative (-3.07 μV), but this was not significant (t(13)=-0.89, 

p=.39). 

For peak latency, the allophonic contrast flap was faster (337.89ms) than the 

phonemic contrast fricative (352.12ms) in the classic MMN, however this difference was 

not significant (t(13)=-1.55, p=.14). In the iMMN configuration, flap again had a slightly 

earlier peak latency (347.01ms) than fricative (350.72ms), but this was not significant 

(t(13)=-0.33, p=.74). 

 

72 

4.3.2.4 Summary of peak latency & amplitude results 

The peak latency and peak amplitude results show a consistent pattern across all 

language groups with regard to segment. The differences in amplitude and latency do 

not seem to be driven by allophony, but rather by segment. While the only difference 

that reached statistical significance was the difference in peak amplitude between the 

flap and fricative in the monolingual English speakers, in most cases, the flap contrast 

consistently patterned more negative in amplitude peak and consistently earlier peak 

latency than the fricative contrast. Table 3 summarizes these results. 

 

Peak Amplitude 
MMN (μV) 
[d]/[ð] 
-2.45* 

[d]/[ɾ] 
-3.68* 

Peak Amplitude 
iMMN (μV) 
[d]/[ð] 
-2.27* 

[d]/[ɾ] 
-3.46* 

Peak Latency 
MMN (ms) 
[d]/[ð] 
339.20 

Peak Latency 
iMMN (ms) 
[d]/[ð] 
336.30 

-2.55 

-3.15 

[d]/[ɾ] 
330.50 

 
Monolingual 
English Speakers 
Native Spanish 
Speakers 
Advanced 
Spanish 
Learners 
Table 3: Within Language Group Peak Amplitude and Latency Results, * denotes 
significant differences between sound comparisons 

[d]/[ɾ] 
336.80 

337.89 

331.32 

-3.27 

-4.18 

340.91 

340.55 

350.72 

347.01 

-2.14 

339.48 

-2.53 

-3.07 

-3.68 

352.12 

 

73 

Figure 29: Peak amplitudes for each speaker group and sound comparison in 
both the classic MMN and iMMN conditions. In both conditions, the only 
significant differences in peak amplitude between the flap and fricative was for 
the monolingual English speakers. 

 

 

74 

 

Figure 30: Peak latencies for each speaker group and sound comparison in both 
the classic MMN and iMMN conditions. While the results show consistently 
faster peak latencies for the flap in all groups and in both conditions, none of 
the differences reach statistical significance.  

4.3.3 Across-language group results 

When comparing the results across the language groups, it is instructive to look 

at the grand-average difference waves for each group on each contrast. This allows us to 

see quite clearly how amplitude and peak latency changes for each group. Again, I 

present the waveforms in both classic MMN (where the deviant response is compared 

directly to the standard [d] response, e.g. deviant [ɾ] compared to standard [d]) and the 

iMMN (where the deviant response is compared to the response where said segment is 

 

75 

presented as a standard, e.g. deviant [ɾ] compared to standard [ɾ]). These difference 

waves are show in Figures 31 - 34 below.  

Figure 31: Classic MMN difference waves for all speaker groups. Deviant [r] 
subtracted from standard [d]. 

 

Figure 32: Identity MMN difference waves for all speaker groups. Deviant [r] 
subtracted from standard [r]. 

 

 

76 

Figure 33: Classic MMN difference waves for all speaker groups. Deviant [ð] 
subtracted from standard [d]. 

 

 

Figure 34: Identity MMN difference waves for all speaker groups. Deviant [ð] 
subtracted from standard [ð].  

4.3.3.1 English compared to Spanish 

In terms of mean amplitude, the iMMN response to the [ɾ]-[d] comparison was 

significantly stronger in the monolingual English speakers (-2.07 μV) than that of the 

native Spanish speakers (-0.98 μV) (t(20)=2.34, p=.03).  However, in the classic MMN 

condition, while the pattern is similar to the iMMN condition, the statistical difference is 

only marginal. Monolingual English speakers had a mean amplitude of -1.81μV 

 

77 

whereas the native Spanish speakers had a mean amplitude of -1.15μV. (t(21.9)=1.36, 

p=.09).  

For the [ð]-[d] contrast, the monolingual English speakers had slightly more 

negative mean amplitudes (-1.51μV) in the iMMN condition than the native Spanish 

speakers (-1.36μV), but this difference was not significant (t(21.6)=0.37, p=0.36). The 

results for the classic MMN condition were similar. With the mean amplitude of the 

monolingual English speakers (-1.26μV) being slightly more negative than that the 

native Spanish speakers (-1.13μV), and no statistical difference in means (t(21.6)=0.31, 

p=0.38). 

Looking only at peak amplitude, we see slightly different results. For the [ɾ]-[d] 

contrast we see a peak amplitude for monolingual English speakers in the iMMN 

condition of -3.46μV compared to -2.15μV for the native Spanish speakers. This result 

was statistically significant (t(19.5)=2.29, p=0.02). The MMN condition did not reach 

statistical significance with the monolingual English speakers having a peak amplitude 

of -3.7μV and the native Spanish speakers having a peak amplitude of -3.15μV 

(t(18.6)=0.86, p=0.2).  

The [ð]-[d] comparison shows the monolingual English speakers with a peak 

amplitude of -2.67μV and the native Spanish speakers with a peak amplitude of -2.53μV 

in the iMMN condition. These were not significantly different (t(18.7)=0.29, p=0.39). The 

 

78 

MMN condition was similar with the monolingual English speakers having a slightly 

lower peak amplitude of -2.45μV than the native Spanish speakers at -2.55μV. This was 

also not significantly different ((t(19.7)=-0.19, p=0.58). 

4.3.3.2 English compared to Learners 

For mean amplitude, the iMMN response to the [ɾ]-[d] comparison was nearly 

identical in the monolingual English speakers (-2.07μV) than that of the advanced 

Spanish learners (-2.01) (t(23)=-0.09, p=.93).  In the classic MMN condition, the advanced 

Spanish learners trended toward a larger mismatch response than the monolingual 

English speakers, but this difference did not reach statistical significance. Monolingual 

English speakers had a mean amplitude of -1.81μV whereas the advanced Spanish 

learners had a mean amplitude of -2.07μV (t(21.5)=0.39, p=.35).  

For the [ð]-[d] comparison, the monolingual English speakers had less negative 

mean amplitudes (-1.51μV) in the iMMN condition than the advanced Spanish learners 

(-2.02μV), but this difference was not significant (t(24.9)=1.31, p=0.2). The results for the 

classic MMN condition were similar. With the mean amplitude of the monolingual 

English speakers (-1.26μV) being slightly less negative than that the advanced Spanish 

learners (-1.79μV), and no statistical difference in means (t(24.9)=1.29, p=0.1). 

Turning to peak amplitude, we see similar results. For the [ɾ]-[d] comparison we 

see a peak amplitude for monolingual English speakers in the iMMN condition of -

 

79 

3.46μV compared to -3.67μV for the advanced Spanish learners. There was no 

statistically significant difference between these groups (t(19.9)=0.31, p=0.76). The MMN 

condition also did not reach statistical significance with the monolingual English 

speakers having a peak amplitude of -3.7μV and the advanced Spanish learners having 

a slightly higher peak amplitude of -4.18μV (t(19.7)=0.67, p=0.51).  

The [ð]-[d] comparison shows the monolingual English speakers with a peak 

amplitude of -2.67μV and the advanced Spanish learners with a peak amplitude of -

3.07μV in the iMMN condition. These were not significantly different (t(24.9)=1.07, 

p=0.29). The MMN condition, however, showed the monolingual English speakers 

having a more positive peak amplitude of -2.45μV than the advanced Spanish learners 

at -3.27μV. This difference was marginally significant (t(24.9)=1.95, p=0.06). 

4.3.3.3 Spanish compared to Learners 

Mean amplitude for the iMMN response to the [ɾ]-[d] comparison was 

significantly stronger in the advanced Spanish learners (-2.01μV) than that of the native 

Spanish speakers (-0.98μV), t(18.1)=1.7, p=.05.  In the MMN condition, advanced 

Spanish learners had a mean amplitude of -2.07μV whereas the native Spanish speakers 

had a mean amplitude of -1.15μV. This difference was marginally significant 

(t(20.7)=1.41, p=.08).  

 

80 

For the [ð]-[d] comparison, the advanced Spanish learners had a significantly 

more negative mean amplitudes (-2.02μV) in the iMMN condition than the native 

Spanish speakers (-1.36μV) (t(22.2)=1.66, p=0.05). The results for the MMN condition 

were similar. With the mean amplitude of the advanced Spanish learners (-1.79μV) 

being slightly more negative than that the native Spanish speakers (-1.13μV), and this 

difference was marginally significant(t(22.5)=1.56, p=0.06). 

In the peak amplitude analysis, we see for the [ɾ]-[d] comparison a peak 

amplitude for advanced Spanish learners in the iMMN condition of -3.68μV compared 

to -2.15μV for the native Spanish speakers. This result was marginally significant 

(t(22.1)=1.97, p=0.06). The MMN condition did not reach statistical significance with the 

advanced Spanish learners having a peak amplitude of -4.18 and the native Spanish 

speakers having a peak amplitude of -3.15 (t(22.5)=1.24, p=0.2).  

The [ð]-[d] comparison shows the advanced Spanish learners with a peak 

amplitude of -3.07μV and the native Spanish speakers with a peak amplitude of -2.53μV 

in the iMMN condition. These were not significantly different (t(18.9)=1.17, p=0.13). The 

MMN condition showed a marginally significant differences with advanced Spanish 

learners having a more negative peak amplitude of -3.27μV than the native Spanish 

speakers at -2.55μV (t(20.1)=1.5, p=0.07). 

 

81 

4.3.3.4 Summary of across-language group results 

Beginning with mean amplitudes over the 300ms-400ms time window, we find 

that the monolingual English speakers have significantly larger mismatch responses to 

the [ɾ] in the iMMN condition than do the native Spanish speakers, though this 

difference is attenuated in the classic MMN condition. For the [ð], the monolingual 

English speakers tended to have larger mismatch responses than the native Spanish 

speakers, but these differences were not significant in either the classic MMN or iMMN 

condition. These results persisted in the advanced Spanish learners. On both 

comparisons they patterned more similarly with the monolingual English speakers than 

the native Spanish speakers. The only significant difference was, like the monolingual 

English speakers, a stronger classic MMN for the [ɾ] than the native Spanish speakers, 

though the iMMN condition was marginally significant.  

Peak amplitudes paint a slightly different picture. The monolingual English 

speakers again show a stronger mismatch response for the [ɾ] comparison than the 

native Spanish speakers in the iMMN condition, but no significant difference on the [ð] 

comparison. The advanced Spanish learners diverge on some measures from both the 

monolingual English and native Spanish groups.  On the [ɾ], the advanced Spanish 

learners pattern more like the monolingual English speakers, showing a marginally 

statistical difference in the iMMN condition than the native Spanish speakers. This did 

not persist in the classic MMN condition. For the [ð], the advanced Spanish learners 

 

82 

showed marginally stronger classic MMN responses when compared to both the 

monolingual English speakers and the native Spanish speakers. However, the iMMN 

condition showed no such difference. 

To summarize, both the monolingual English speakers and the advanced 

Spanish learners show stronger mismatch responses to the [ɾ] comparison than the 

native Spanish speakers, both in mean amplitude and peak amplitude. This is contra 

expectations given that the [ɾ] is a phoneme in Spanish, but not in English. If anything, 

the advanced Spanish learners trend toward stronger mismatch responses than the 

monolingual English speakers. For the [ð], the monolingual English speakers and the 

native Spanish speakers do not differ significantly in mean or peak amplitude, but the 

advanced Spanish learners had marginally stronger mismatches than both the 

monolingual English and native Spanish groups in the classic MMN condition. 

 

 

 

83 

Chapter 5: Discussion 

In this chapter I’ll discuss and interpret the results of the experiments in chapters 

3 and 4, and discuss their contributions to the broader field. I’ll begin with the 

behavioral study in chapter 3. 

5.1 Interpretation of the Behavioral results 

The behavioral study detailed in Chapter 3 was an AXB discrimination task run 

on three speaker groups. Groups of monolingual English speakers, native Spanish 

speakers, and native English speaking advanced learners of Spanish all participated in 

the same AXB task where they were asked to discriminate between multiple tokens of 

[ɾ], [ð], and [d]. The expectation in this study was that these groups would differ in their 

ability to discriminate the consonants from one another based on the phonological 

status of the sounds in their native language. The purpose of testing the advanced 

Spanish learner group was to determine whether that group’s discrimination ability 

would become more ‘native Spanish-like’ when compared to their monolingual English 

counterparts. We predicted that the monolingual English speakers would have 

difficulties discriminating [ɾ] from [d] because [ɾ] is an allophone of [d] in English, while 

at the same time discriminating very well [ð] from [d]. For full results, see section 3.3. In 

summary, the predictions were borne out. In every measure; reaction time, accuracy, 

 

84 

and A’, the language groups more easily discriminated the pair that was phonemically 

contrastive in their respective language. That is, the flap [ɾ] was more easily 

discriminated from the [d] by the native Spanish group and the [ð] was more easily 

discriminated from the [d] but the monolingual English group. These results replicate 

those found in Boomershine et al., (2008). The current study adds an additional novelty 

of a native English-speaking group of advanced Spanish learners. These learners appear 

to perform like the native Spanish speakers when discriminating flap [ɾ] from [d], 

significantly outperforming the monolingual English group. They also appear to show 

some bi-directional influence of their L2 as their discriminability of the fricative [ð] from 

[d] was, while better than the native Spanish speakers, significantly impaired compared 

to the monolingual English group. This decrease in discriminability on a L1 phonemic 

contrast, is apparently due to having to juggle more than one representation of [ð] in 

their minds. While this result was not expected, it is not without precedent. Influence of 

the L2 on the L1 has been shown in previous work, including in EEG work (Caramazza 

et al., 2005; Guion, 2003; Peltola et al., 2003; Sancier & Fowler, 1997)).   

Returning to the initial questions laid out in Chapter 1 of this dissertation; Do we 

see evidence of L2 category formation in second language learners? It would appear 

from the behavioral data that this is a possibility, and that the advanced Spanish 

learners have created a category for the flap [ɾ]. However, reaction times in this type of 

discrimination task are on the order of seconds; an eternity when it comes to the brain. 

 

85 

There could be all kinds of scaffolding being built to get to the correct answer in the 

three seconds or so it takes a participant to provide their answer. This motivated the 

need for the EEG study. 

Do we see evidence that perception is modulated by level or representation? Yes, 

like Boomershine et al. (2008), we clearly see that when sounds are in allophonic 

(surface) contrast in a language they are more difficult to perceive than when they are in 

phonemic contrast. Using multiple standard tokens in the EEG experiment task and 

comparing results from the MMN and iMMN condition is an attempt at sussing out 

these differences experimentally.  

Are we able to use the MMN to probe phonemic representations? The EEG 

experiment in Chapter 4 was designed to probe this question specifically. I will now 

turn to the Electrophysiological study. 

5.2 Interpretation of the Electrophysiological results 

MMN studies, specifically studies involving L2, have provided disparate results 

that can be difficult to reconcile. Generally, these studies have been designed to probe 

phonological information via the MMN, usually assumed to mean phonemic 

representations. Perhaps most clearly in the Kazanina et al., (2006) study, the 

differences in the MMN between Russian and Korean speakers are said to be due to 

differences in the status of the [t] and [d] sounds in those languages. Russian contrasts 

 

86 

[t] and [d] phonemically whereas Korean does not; consequently, the Russian speakers 

had a large MMN to this contrast whereas Korean speakers did not. Results such as 

these tend to mirror behavioral results on perception, where listeners more readily 

perceive phonemic contrasts than allophonic ones. The prediction then has been that the 

MMN should be larger in amplitude when a contrast is very salient to the listener, and 

attenuated (or nonexistent) on contrasts that are difficult to perceive. However, cross-

linguistic MMN results have not been so clear. For example, Winkler, et al., (1999) 

tested Hungarian and Finnish speakers on both in-category and across-category vowel 

contrasts and found that both contrasts resulted in an MMN in both languages with the 

across-category contrast producing a larger MMN. In Finnish, the across-category 

contrast elicited a double peak MMN (presented as distinct early and late responses). In 

Hungarian, the across-category had only a single peak consistent with the early 

response of the Finns. Näätänen’s work on Finnish and Estonian produced similar 

results for contrastive pairs (large MMNs) but with deviants not present in the listener’s 

L1, an attenuated MMN was produced (Naatanen & Alho, 1997).  

In my study, I contrasted [d] with [ɾ] and [ð] in both English and Spanish 

speakers, along with advanced learners of Spanish. These contrasts provide both a 

phonemic and allophonic pair to the speakers. If the EEG results were to mirror the 

behavioral results, we would expect small mismatch effects for the flap-stop [ɾ]-[d] 

comparison in the monolingual English speakers along with large mismatch responses 

 

87 

to the fricative-stop [ð]-[d] comparison. We would expect this pattern to be reversed in 

the native Spanish speakers. For the advanced Spanish learners, we would expect to see 

an increased MMN for the flap-stop [ɾ]-[d] comparison and perhaps a slightly 

attenuated MMN for the fricative-stop [ð]-[d] comparison when compared to the 

monolingual English group. Essentially, they should become more ‘Spanish-like’. This 

was both the predicted and borne out result of the behavioral study. However, I found 

that both the phonemic and allophonic contrasts produce MMNs in all subject groups, 

though there are some differences in amplitude and latency. This is surprising given the 

behavioral results where listeners are poor at discriminating the allophonic contrasts in 

their L1.  

Compared to the results of the EEG study, the results of the discrimination study 

detailed in Chapter 3 are very straightforward. Each of the predictions made is borne 

out and it is a near replication of the Boomershine et al., (2008) study. This reinforces the 

validity of the study and allows us to be confident in the stimuli. It follows then that the 

unexpected results of the EEG study are not due to issues with the stimuli or 

methodology. So how then do we reconcile the unexpected EEG results with the 

comparatively beautiful behavioral results? First, it is important to acknowledge that 

given the past EEG work of this nature, these results are not that surprising. Herd, 

(2011) showed similar results for her English subjects. Her English subjects produced a 

large MMN for the allophonic contrast of [d] and [ɾ]. She offers that this may be due to 

 

88 

[ɾ] being an allophone of both [t] and [d] in English, and that in the context of one of 

those stops, an English listener may be perceiving the other, essentially producing an 

MMN you would expect for a phonemic contrast.  Barrios et al., (2016) ran an MEG 

study with monolingual English speakers, native Spanish speakers, and advanced L2 

English learners on the same contrasts as the ones in this dissertation and found no 

statistical significance in the differences in groups. However, they found similar 

mismatch responses to both the [d]-[ɾ] comparison and the [d]-[ð] comparison in the 

English speakers, while Spanish speakers had similar mismatch responses to English 

speakers on the [d]-[ɾ] as well as an unexpected mismatch response for [d]-[ð]. Barrios 

does ultimately claim there is evidence for the advanced L2-English learners having 

acquired the proper phonemic contrast despite the allophonic status in L1, but this is 

made solely on the grounds that there is a statistically significant mismatch negativity 

for both the phonemic and allophonic contrasts among that language group. However, 

the L1-English and L1-Spanish group do not pattern as expected, nor do they match the 

behavioral data. The L1-English speakers show no significant mismatch for either the 

phonemic or allophonic contrast, and most surprisingly, the L1-Spanish speakers had 

only a significant mismatch for the allophonic contrast. These results are strikingly 

similar to mine, though that makes them no less difficult to interpret. Miglietta et al., 

(2013) also found significant MMNs for allophonic contrasts in Italian, both an 

allophonic and phonemic contrast elicited MMNs but the phonemic pair had a shorter 

 

89 

peak latency.  In my study, I found the opposite, that the allophonic contrasts displayed 

faster peak latencies than the phonemic ones. 

So, the MMN results are not as simple as appealing to phonemic status where 

contrastive equates to an MMN and non-contrastive no MMN. It also seems that 

variations in phonetic detail used by the listener can alter the MMN by way of latency 

or amplitude.  It is important to understand that mismatch negativity is simply a 

marker of change detection. It is not in any way ‘linguistic’ and seems to be stimulus 

agnostic. Although it can be used as evidence of categorical perception, it does not 

index allophony or phonemic categories. Those are simply aspects of the stimuli that 

could result in changes to the ERP waveform. It is predominately found on fronto-

central electrodes, but localization of brain activity cannot be reliably inferred from 

scalp distributions. We must resist the temptation to ascribe meaning to the ERP 

components where there is none.  So, it falls to the researcher to design stimuli and 

experiments in such a way to constrain the type of change that could be detected in the 

signal.  

There are some amplitude differences in mismatch responses to the stimuli that 

vary by speaker group. Both the monolingual English speakers and the advanced 

Spanish learners showed stronger mismatch responses to the [ɾ]-[d] comparison than 

the native Spanish speakers. The advanced Spanish learners exhibited a stronger 

mismatch response than the native Spanish speakers on the [ð]-[d] comparison.  

 

90 

Regardless, we must confront the fact that the segments themselves pattern alike no 

matter the group. That is, the flap sound in all groups appears to elicit a higher 

amplitude mismatch response than the fricative sound. Also, peak latencies are 

consistent across language groups. The flap response peaks earlier than the fricative 

response in all cases. These facts point to the ERP waveform being a product of the 

phonetic detail of the individual segments themselves rather than phonological status 

such as allophone or phoneme. One way to test this is to calculate acoustic distances 

between the stimuli to see if mismatch response correlates with distance. To this end I 

calculated the acoustic similarity of the stimuli and compared the mean distances of the 

[ɾ] and [ð] to the standard [d]. We find that the mean acoustic distance between [ɾ] and 

[d] is larger than the mean acoustic distance between the [ð] and [d]. This difference in 

acoustic distance correlates with a larger mismatch response for the [ɾ] than the [ð]. 

Acoustic distance was calculated using Phonological CorpusTools (Hall et al., 2017; 

Mackie, et al., 2014)acoustic similarity function to compute mel-fruquency cepstrum 

coefficients (MFCC) (Delvaux & Soquet, 2007; Mielke, 2012). Distances were computed 

using dynamic time warping (Sakoe & Chiba, 1971). Figure 35 shows the mean acoustic 

distance for each pair of sounds. Essentially, these tools map the spectrum of the sounds 

onto a two-dimensional space so that a linear distance between them can be calculated. 

You can see in Figure 35 that the Euclidean distance between sounds is higher when the 

segments are different as in the case when comparing my [ɾ], [d], and [ð] stimuli. When 

 

91 

different tokens of identical segments are compared, the acoustic distance is relatively 

low. When tokens of different segments are compares, the acoustic distance is much 

higher, with the [ɾ]-[d] comparison being the most distant from one another.  

 

Figure 35: Acoustic distance calculations for each sound comparison showing 
that identical sounds have relatively low acoustic distance while  [ɾ] is further 
from [d] than [ð] is from [d]. 

 

Tying the differences to the MMN response to the phonetic categories of the 

segments themselves, rather than appealing to phonemic status seems reasonable given 

my results, but what about previous work that has provided evidence that the MMN is 

sensitive to phonemic status? It is possible that for the given sound comparisons in my 

 

92 

study subjects were simply unable to construct a standard based on phonemic 

information available in the stimuli. However, it may also be possible to reinterpret 

previous results as also showing sensitivity to phonetic, and not phonemic differences. 

The Näätänen et al., (1997) study on Finns and Estonians for example, it could be that 

the Finns simply assimilated the [õ] vowel as a poor exemplar of their existing phonetic 

category for [ö], causing the attenuated MMN to that stimuli when compared to the [ö] 

deviants. In their study on Russian and Korean stops, Kazanina et al, (2006) showed 

that in the Russian speakers, a large MMN was seen for the phonemic /t/-/d/ contrast 

using a variable standards paradigm. The lack of a similar MMN in the Korean speakers 

was claimed to be due to the voiced stop [d] being an allophone of the voiceless /t/ in 

that language. However, there is evidence for passive voicing in Korean (Jun 1996, 

Wright, 2007). If voicing is not phonological in Korean, then there would not be a clear 

phonetic category distinction between [t] and [d]. There would just be a single phonetic 

category for [t] that gets voiced only in production. Thus, if the MMN was sensitive to 

phonetic category, we would expect no MMN in the Korean speakers. In the Miglietta 

et al., (2013), where it was shown that a phonemic vowel contrast in Italian had a faster 

peak latency than an allophonic one, the authors claim that this is evidence for the 

MMN being sensitive to allophony. Because there was no cross-linguistic comparison, 

and the test was done on a single pair of contrasts, we cannot be sure that the phonetics 

 

93 

of the sounds themselves are driving these results, especially given that both contrasts 

displayed a significant MMN.  

One interesting finding is that the advanced Spanish learners have marginally 

higher peak amplitudes for both the [ɾ]-[d] (iMMN) and [ð]-[d] (MMN) comparisons 

than their native Spanish speaking counterparts (p=.06 and .07 respectively). The 

learners also have a marginally significant higher peak amplitude on the [ð]-[d] 

comparison than the monolingual English speakers in the classic MMN comparison. For 

the [ɾ]-[d] comparison, we would have expected that the advanced Spanish learners 

would have a stronger mismatch response than the monolingual English speakers if 

they were indeed creating a new category, and while the results do not reach 

significance, the trend is toward a higher peak amplitude for the learners. This 

increased amplitude could be indicative the advanced Spanish learners forming a new 

category for [ɾ]. Some caution should be taken with this interpretation due to the fact 

that the native Spanish speakers already show attenuated mismatch responses to the [ɾ]-

[d] comparison when compared to both groups of native English speakers. It could be 

that the additional ways to use [ɾ], as an allophone to both [d] and [t] in English, as well 

as a new phonemic contrast for the advanced Spanish learners, results in even greater 

perceptual contrast in these learners.  For the reasons stated above, and because the 

differences between groups are marginal if they exist at all, one should be wary of 

putting much stock in the peak amplitude results. This illustrates a more general issue 

 

94 

of researcher degrees of freedom when using ERPs in language research that I discuss 

in the section below. 

In light of the ERP results, I return to the original research questions highlighted 

in Chapter 1. Do we see evidence of L2 category formation in second language learners? 

Even though the advanced Spanish learners do show a strong mismatch response to the 

[ɾ]-[d] comparison, it would be difficult to claim that this is due to them creating a 

category for [ɾ], because the monolingual English speakers already show a similar 

mismatch response to this comparison. The learners also show a stronger mismatch 

response to the [ð]-[d] comparison than both the monolingual English speakers and the 

native Spanish speakers. So, while the advanced Spanish learners do appear to differ 

from their monolingual counterparts, they are not necessarily becoming more “Spanish-

like” in their ERP results. The advanced Spanish learners showed less discriminability 

for the [ð]-[d] contrast than the monolingual English speakers in the behavioral task, 

something that could be indicative of the Spanish spirantization rule increasing the 

amount of processing needed, they also show slightly stronger mismatch responses to 

this contrast in the ERP results. The prediction was that the Spanish learners might 

show an attenuated mismatch response to this contrast with increased L2 influence, as 

was found in Peltola et al., (2003). Instead, the increased processing demand of the 

allophonic rule could instead be having an additive effect on the mismatch response.  

 

95 

Do we see evidence that perception is modulated by level or representation? 

Using multiple real-speech tokens to build the standard is an attempt to control for 

acoustic differences between the segments. If we are able to build a standard in a 

listener’s mind from tokens that are not identical, we can be surer that the phonological 

properties of those sounds are being used to build that standard. Additionally, 

presenting both the classic MMN and iMMN results in theory would allow us to 

distinguish further mismatch responses based on acoustic differences between the 

standard and the deviant sound, and mismatches based on the phonological properties 

of those sounds. Using the variable standard paradigm coupled with measuring the 

iMMN was an attempt to get closer to the underlying level of representation. 

Ultimately, while there are some slight differences when looking at the classic MMN 

and the iMMN results, they generally pattern similarly. So, while the behavioral study 

in Chapter 3 provides evidence for this, the EEG results are less clear.  

Are we able to use the MMN to probe phonemic representations? Despite 

measuring the MMN in multiple ways and this being previously claimed in the 

literature, I find no evidence for the MMN being sensitive to phonemic category, but 

clear evidence that it is sensitive to phonetic category. 

 

96 

5.3 Conclusion 

The first human electroencephalograph was taken by Hans Berger in 1924 and 

since EEG has been utilized in a number of research and clinical settings to study the 

brain. In the last several decades, linguistics and other language sciences have taken to 

the technique. For language research, it is still early days. Various subfields have 

utilized other ERP components to test and refine their theories, such as the semantic 

N400 (Kutas & Hillyard, 1980) or syntactic P600 (Osterhout & Holcomb, 1992). The 

MMN is a relative newcomer, its use in studying phonology increasing in just the last 

twenty years. ERP research is time-consuming; initial set-up and subject preparation 

can be uncomfortable, the tasks are often monotonous, and usually recordings are quite 

long. Because of this, data is precious. From study design to analysis, researchers have 

many degrees of freedom when doing this work. One can choose from several 

measurements, electrode placements, recording settings, and a slew of post-processing 

procedures. Once the data is recorded and measured, there are additional choices to be 

made about statistical tests. It is extremely common in the literature for these choices to 

be made without mentioning the justification of one choice over another (Gelman & 

Loken, 2013; Gelman & Stern, 2006; Luck & Gaspelin, 2017; Simmons, Nelson, & 

Simonsohn, 2011). Part of the novelty of my study is simply acknowledging the myriad 

ways that researchers have analyzed ERP waveforms, specifically when utilizing the 

MMN, and attempting to be open about how that choice can affect the results in 

 

97 

sometimes dramatic ways. This is the reason I chose to look at three different ERP 

measurements; mean amplitude over an analysis window, peak amplitudes in that 

window, and peak latencies in that window. Each measurement has been used in prior 

work and they have all been said to be modulated by phonological status, or at the very 

least, strength of perceptual contrast. In addition, I chose to be explicit about how I was 

calculating the mismatch response by presenting all measurements in both the classic 

MMN (standard response – deviant response) and the iMMN (response of the target 

sound presented in a standard context – response of the target sound presented in a 

deviant context). The latter being an attempt to tap into a more ‘phonological’ response 

by controlling for any effects the standard might have as it relates to the deviant, such 

as differences in acoustics between [d] and the target sounds. This type of “ERPology” 

is a necessary step to undertake before any EEG experiment can tell us something 

meaningful about the brain.  

This dissertation contributes to the growing body of neurolinguistic work on 

second language acquisition and adds to the work on the mismatch negativity ERP 

component. It also extends previous speech perception work on second language 

learners by both replicating previous results and adding new results from 

electroencephalography that show incongruency with behavioral results. It provides 

evidence that advanced second language learners may be able to create categories that 

do not exist in the L1, and also that proficiency in an L2 may affect L1 perception. It also 

 

98 

suggests that while the Mismatch Negativity is sensitive to phonetic category, it may 

not be sensitive to phonemic category. Previous studies that have claimed as much, can 

be re-interpreted as testing phonetic category.

 

99 

 

APPENDICES

 

100 

 

APPENDIX A 

Electrode Cluster Locations 

 

101 

 

 

Figure 36: EEG Channel Locations for MMN Cluster Analysis 

 

 

 

 

102 

APPENDIX B  

Participant Survey 

 

103 

 

 

Native Spanish Speakers 

1.  What is your native language? 

 

2.  Age/Sex/Handedness? 

 

3.  In what country were you born and raised? 

 

4.  At what age was your first exposure to English? 

 

5.  At what age did you take formal English classes? 

 

6.  How long have you been in the United States? 

 

7.  What is the primary language spoken in your home? 
 
8.  Level of education? 

 

104 

 

 

Advanced Spanish Learners 

1.  What is your native language? 

 

2.  Age/Sex/Handedness? 

 

3.  When was your first exposure to Spanish? 
 
4.  How much consecutive time have you had in a majority Spanish-speaking country? 
5.  At what age did you start formal Spanish classes? 

 

6.  How many years of Spanish instruction have you had? 
 
7.  How do you plan to use Spanish in the future? 
 
8.  Level of Education? 

 

105 

 

REFERENCES

 

106 

 

REFERENCES 

Aulanko, R., Hari, R., Lounasmaa, O. V., Näätänen, R., & Sams, M. (1993). Phonetic 

invariance in the human auditory cortex. NeuroReport.  

Barrios, S., Jiang, N., & Idsardi, W. J. (2016). Similarity in L2 Phonology: Evidence from 

L1 Spanish late-learners’ perception and lexical representation of English vowel 
contrasts. Second Language Research.  

Barrios, S. L. . (2013). Similarity In L2 Phonology. Doctoral dissertation, University of 

Maryland. 

Baudena, P., Halgren, E., Heit, G., & Clarke, J. M. (1995). Intracerebral potentials to rare 

target and distractor auditory and visual stimuli. III. Frontal cortex. 
Electroencephalography and Clinical Neurophysiology.  

Beddor, P. S., & Strange, W. (2005). Cross‐language study of perception of the oral–

nasal distinction. The Journal of the Acoustical Society of America.  

Best, C. T. (1994). The emergence of native-language phonological influences in infants: 

A perceptual assimilation model. The Development of Speech Perception: The 
Transition from Speech Sounds to Spoken Words. 

Best, C. T. (1995). A direct realist view of cross-language speech perception. Speech 

Perception and Linguistic Experience. Issues in Cross-Language Research.  

Best, C. T., McRoberts, G. W., & Goodell, E. (2001). Discrimination of non-native 

consonant contrasts varying in perceptual assimilation to the listener’s native 
phonological system. The Journal of the Acoustical Society of America.  

Best, C. T., McRoberts, G. W., & Sithole, N. M. (1988). Examination of Perceptual 

Reorganization for Nonnative Speech Contrasts: Zulu Click Discrimination by 
English-Speaking Adults and Infants. Journal of Experimental Psychology: Human 
Perception and Performance. 

Best, C. T., & Tyler, M. D. (2007). Nonnative and second-language speech perception: 
Commonalities and complementarities. In Language experience in second language 
speech learning: In honor of James Emil Flege. 

Birdsong, D., & Molis, M. (2001). On the Evidence for Maturational Constraints in 

Second-Language Acquisition. Journal of Memory and Language.  

 

107 

Boersma, P., & Weenink, D. (2013). Praat: doing phonetics by computer [Computer 

program]. Glot International, Vol. 5, pp. 341–347.  

Boomershine, A., Hall, K. C., Hume, E., & Johnson, K. (2008). The impact of allophony 

versus contrast on speech perception. Contrast in Phonology - Theory, Perception, 
Acquisition, 145–172.  

Browman, C., & Goldstein, L. (1986). Towards an articulatory phonology. Phonology 

Yearbook, 3, 219–252. 

Browman, C., & Goldstein, L. (1989). Articulatory gestures as phonological units. 

Phonology, 6(2), 201–251. 

Caramazza, A., Yeni‐Komshian, G. H., Zurif, E. B., & Carbone, E. (2005). The acquisition 

of a new phonological contrast: The case of stop consonants in French‐English 
bilinguals. The Journal of the Acoustical Society of America. 

Celsis, P., Boulanouar, K., Doyon, B., Ranjeva, J. P., Berry, I., Nespoulous, J. L., & 

Chollet, F. (1999). Differential fMRI responses in the left posterior superior 
temporal gyrus and left supramarginal gyrus to habituation and change detection 
in syllables and tones. NeuroImage.  

Cheour, M., Ceponiene, R., Lehtokoski, A., Luuk, A., Allik, J., Alho, K., & Näätänen, R. 

(1998). Development of language-specific phoneme representations in the infant 
brain. Nature Neuroscience.  

Cheour, M, Korpilahti, P., Martynova, O., & Lang, A. H. (2001). Mismatch Negativity 

and Late Discriminative Negativity in Investigating Speech Perception and 
Learning in Children and Infants. Audiology and Neurotology, 6(1), 2–11.  

Cheour, Marie, Shestakova, A., Alku, P., Ceponiene, R., & Näätänen, R. (2002). 

Mismatch negativity shows that 3-6-year-old children can learn to discriminate 
non-native speech sounds within two months. Neuroscience Letters, 325(3), 187–190.  

Cook, V. J. (1992). Evidence for Multicompetence. Language Learning.  

Csépe, V., Pantev, C., Hoke, M., Hampson, S., & Ross, B. (1992). Evoked magnetic 

responses of the human auditory cortex to minor pitch changes: localization of the 
mismatch field. Electroencephalography and Clinical Neurophysiology/ Evoked Potentials.  

 

 

 

108 

Deacon, D., Gomes, H., Nousak, J. M., Ritter, W., & Javitt, D. (2000). Effect of frequency 

separation and stimulus rate on the mismatch negativity: An examination of the 
issue of refractoriness in humans. Neuroscience Letters.  

Delorme, A., & Makeig, S. (2004). EEGLAB: An open source toolbox for analysis of 

single-trial EEG dynamics including independent component analysis. Journal of 
Neuroscience Methods.  

Delvaux, V., & Soquet, A. (2007). The influence of ambient speech on adult speech 

productions through unintentional imitation. Phonetica.  

Dien, J. (1998). Issues in the application of the average reference: Review, critiques, and 

recommendations. Behavior Research Methods, Instruments, and Computers.  

Donaldson, W. (1993). Accuracy of d′ and A′ as estimates of sensitivity. Bulletin of the 

Psychonomic Society.  

Dupoux, E., Pallier, C., Sebastian, N., & Mehler, J. (1997). A destressing “deafness” in 

French? Journal of Memory and Language.  

Eulitz, C., & Lahiri, A. (2004). Neurobiological Evidence for Abstract Phonological 

Representations in the Mental Lexicon during Speech Recognition. Journal of 
Cognitive Neuroscience, 16(4), 577–583.  

Eulitz, C., & Lahiri, A. (2004b). Neurobiological evidence for abstract phonological 

representations in the mental lexicon during speech recognition. Journal of Cognitive 
Neuroscience, 16(4), 577–583.  

Face, T. L. (2006). Intervocalic rhotic pronunciation by adult learners of Spanish as a 

second language. Selected Proceedings of the 7th Conference on the Acquisition of 
Spanish and Portuguese as First and Second Languages. 

Face, T. L., & Menke, M. R. (2009). Acquisition of the Spanish voiced spirants by second 
language learners. In Selected proceedings of the 11th Hispanic Linguistics Symposium. 

Flege, J. E. (1991). Perception and production: The relevance of phonetic input to L2 

phonological learning (T. Heubner & C. W. Ferguson, Eds.). Amsterdam: John 
Benjamins. 

Flege, J. E. (1995). Second Language Speech Learning: Theory, Findings, and Problems. 

Speech Perception and Linguistic Experience: Issues in Cross-Language Research.  

 

 

 

109 

Flege, J. E. (2007). Language contact in bilingualism: Phonetic system interactions. In J. 
Cole & J. I. Hualde (Eds.), Laboratory Phonology 9 (pp. 353–381). Berlin, New York: 
Mouton de Gruyter. 

Flege, J. E. (2011). Assessing constraints on second-language segmental production and 

perception. In Phonetics and Phonology in Language Comprehension and Production: 
Differences and Similarities.  

Flege, J. E., Munro, M. J., & MacKay, I. R. A. (1995). Factors affecting strength of 

perceived foreign accent in a second language. The Journal of the Acoustical Society of 
America. 

Fowler, C. A. (1986). An event approach to the study of speech perception from a direct-

realist perspective. Journal of Phonetics, 14, 3–28. 

Fowler, C. A. (1989). Real Objects of Speech Perception: A Commentary on Diehl and 

Kluender. Ecological Psychology.  

Fowler, C. A. (2005). Auditory perception is not special: We see the world, we feel the 

world, we hear the world. The Journal of the Acoustical Society of America.  

Gelman, A., & Loken, E. (2013). The garden of forking paths: Why multiple comparisons can 
be a problem, even when there is no “fishing expedition” or “p-hacking” and the research 
hypothesis was posited ahead of time∗. 

Gelman, A., & Stern, H. (2006). The difference between “significant” and “not significant” is 

not itself statistically significant. 

Gerrits, E., & Schouten, M. E. H. (2004). Categorical perception depends on the 

discrimination task. Perception & Psychophysics, 66(3), 363–376.  

Guion, S. G. (2003). The vowel systems of Quichua-Spanish bilinguals: Age of 

acquisition effects on the mutual influence of the first and second languages. 
Phonetica. 

Halgren, E., Baudena, P., Clarke, J. M., Heit, G., Liégeois, C., Chauvel, P., & Musolino, 
A. (1995). Intracerebral potentials to rare target and distractor auditory and visual 
stimuli. I. Superior temporal plane and parietal lobe. Electroencephalography and 
Clinical Neurophysiology.  

 

 

 

110 

Halgren, E., Marinkovic, K., & Chauvel, P. (1998). Generators of the late cognitive 

potentials in auditory and visual oddball tasks. Electroencephalography and Clinical 
Neurophysiology.  

Hall, K. C., Allen, B., Fry, M., Johnson, K., Lo, R., Mackie, S., & McAuliffe, M. (2017). 

Phonological CorpusTools.  

Hari, R., Hämäläinen, M., Ilmoniemi, R., Kaukoranta, E., Reinikainen, K., Salminen, J., 

Sams, M. (1984). Responses of the primary auditory cortex to pitch changes in a 
sequence of tone pips: Neuromagnetic recordings in man. Neuroscience Letters.  

Harris, J. w. (1969). Spanish Phonolgy. Cambridge, MA: MIT Press. 

Hartshorne, J. K., Tenenbaum, J. B., & Pinker, S. (2018). A critical period for second 

language acquisition: Evidence from 2/3 million English speakers. Cognition.  

Herd, W. (2011). The preceptual and production training of /d flap r/ in L2 Spanish: 

Behavioral, psychologistics, and neurolinguistic evidence. Doctoral dissertation, 
University of Kansas. 

Herd, W., Jongman, A., & Sereno, J. (2010). An acoustic and perceptual analysis of /t/ 

and /d/ flaps in American English. Journal of Phonetics, 38(4), 504–516.  

Hestvik, A., & Durvasula, K. (2016). Neurobiological evidence for voicing 

underspecification in English. Brain and Language.  

Johnson, J. S., & Newport, E. L. (1989). Critical period effects in second language 

learning: The influence of maturational state on the acquisition of English as a 
second language. Cognitive Psychology.  

Jun Zhang, & Shane T. Mueller. (2005). A note on ROC analysis and non-parametric 

estimate of sensitivity. Psychometrika. 

Kahn, D. (1976). Syllable-based generalizations in English phonology. Doctoral dissertation, 

Massachusetts Institute of Technology. 

Kazanina, N., Phillips, C., & Idsardi, W. (2006). The influence of meaning on the 

perception of speech sounds. Proceedings of the National Academy of Sciences, 103(30), 
11381–11386.  

Kenstowicz, M. J. (1994). Phonology in generative grammar. Oxford: Blackwell. 

 

 

 

111 

Kilner, J. M., Stephan, K. E., Friston, K. J., & Garrido, M. I. (2009). The mismatch 

negativity: A review of underlying mechanisms. Clinical Neurophysiology, 120(3), 
453–463.  

Kutas, M., & Hillyard, S. A. (1980). Event-related brain potentials to semantically 
inappropriate and surprisingly large words. Biological Psychology, 11, 99–116. 

Ladefoged, P., & Johnson, K. (2011). A Course in Phonetics (6th ed). In Wadsworth.  

Lenneberg, E. H. (1967). Biological foundations of language. New York: John Wiley and 

Sons, 1967. Behavioral Science, 13(6), 493–495. 

Lopez-Calderon, J., & Luck, S. J. (2014). ERPLAB: an open-source toolbox for the 

analysis of event-related potentials. Frontiers in Human Neuroscience.  

Luck, S. J. (2005). An introduction to the event-related potential technique. Cambridge, 

Massachusetts: MIT Press. 

Luck, S. J., & Gaspelin, N. (2017). How to get statistically significant effects in any ERP 

experiment (and why you shouldn’t). Psychophysiology.  

Mackay, I. R. A., & Flege, J. E. (2004). Effects of the age of second language learning on 

the duration of first and second language sentences: The role of suppression. 
Applied Psycholinguistics.  

Mackie, S., Currie Hall, K., Allen, B., Michael, M., & Fry, M. (2014). Phonological 

CorpusTools: A free, open-source tool for phonological analysis. Presented at the 
14th Conference for Laboratory Phonology. Tokyo,Japan. 

Macmillan, N. A., & Creelman, C. D. (2004). Detection Theory: A User’s Guide. Taylor & 

Francis. 

Mäntysalo, S., & Näätänen, R. (1987). The duration of a neuronal trace of an auditory 
stimulus as indicated by event-related potentials. Biological Psychology, 24(3), 183–
195.  

Martínez-Celdrán, E., Fernández-Planas, A. M., & Carrera-Sabaté, J. (2003). Castilian 

Spanish. Journal of the International Phonetic Association.  

McCarthy, J. J. (2010). Derivations and levels of representation. In The Cambridge 

Handbook of Phonology. 

 

 

 

112 

Mielke, J. (2012). A phonetically based metric of sound similarity. Lingua.  

Miglietta, S., Grimaldi, M., & Calabrese, A. (2013). Conditioned allophony in speech 

perception: An ERP study. Brain and Language, 126(3), 285–290.  

Molholm, S., Martinez, A., Ritter, W., Javitt, D. C., & Foxe, J. J. (2005). The neural 

circuitry of pre-attentive auditory change-detection: An fMRI study of pitch and 
duration mismatch negativity generators. Cerebral Cortex.  

Näätänen, R., & Alho, K. (1995). Mismatch negativity--a unique measure of sensory 

processing in audition. Int J Neurosci, 80(1–4), 317–337.  

Näätänen, R., & Alho, K. (1997). Mismatch negativity--the measure for central sound 

representation accuracy. Audiol Neurootol, 2(5), 341–353.  

Näätänen, R., Lehtokoski, A., Lennes, M., Cheour, M., Huotilainen, M., Iivonen, A., … 

Alho, K. (1997). Language-specific phoneme representations revealed by electric 
and magnetic brain responses. Nature, 385(6615), 432–434. 

Näätänen, R., Paavilainen, P., Alho, K., Reinikainen, K., & Sams, M. (1989). Do event-
related potentials reveal the mechanism of the auditory sensory memory in the 
human brain? Neurosci Lett, 98(2), 217–221.  

Näätänen, R., Paavilainen, P., Rinne, T., & Alho, K. (2007). The mismatch negativity 

(MMN) in basic research of central auditory processing: A review. Clinical 
Neurophysiology, Vol. 118, pp. 2544–2590.  

Näätänen, R., Paavilainen, P., Titinen, H., Jiang, D., & Alho, K. (1993). Attention and 

mismatch negativity. Psychophysiology.  

Näätänen, R, Gaillard, A. W., & Mäntysalo, S. (1978). Early selective-attention effect on 

evoked potential reinterpreted. Acta Psychologica, 42, 313–329.  

Näätänen, R., & Michie, P. T. (1979). Early selective-attention effects on the evoked 

potential: A critical review and reinterpretation. Biological Psychology.  

Näätänen, R., Pakarinen, S., Rinne, T., & Takegata, R. (2004). The mismatch negativity 

({MMN}): towards the optimal paradigm. Clinical Neurophysiology, 115(1), 140–144. 

Osterhout, L., & Holcomb, P. (1992). Event-related brain potentials elicited by syntactic 

anomaly. Journal of Memory and Language, 31, 785–806. 

Paradis, M. (1985). On the representation of two languages in one brain. Language 

 

113 

Sciences. 

Patterson, D., & Connine, C. M. (2001). Variant frequency in flap production: A corpus 
analysis of variant frequency in American english flap production. Phonetica, 58(4), 
254–275.  

Peltola, M. S., Kujala, T., Tuomainen, J., Ek, M., Aaltonen, O., & Näätänen, R. (2003). 
Native and foreign vowel discrimination as indexed by the mismatch negativity 
(MMN) response. Neuroscience Letters, 352(1), 25–28.  

Peltola, M. S., Kuntola, M., Tamminen, H., Hämäläinen, H., & Aaltonen, O. (2005). Early 

exposure to non-native language alters preattentive vowel discrimination. 
Neuroscience Letters, 388(3), 121–125. 

Phillips, C., Pellathy, T., Marantz, A., Yellin, E., Wexler, K., Poeppel, D., Roberts, T. 

(2000). Auditory cortex accesses phonological categories: An MEG mismatch study. 
Journal of Cognitive Neuroscience, 12(6), 1038–1055.  

Pulvermüller, F., & Shtyrov, Y. (2006). Language outside the focus of attention: The 

mismatch negativity as a tool for studying higher cognitive processes. Progress in 
Neurobiology.  

Sabri, M., & Campbell, K. B. (2001). Effects of sequential and temporal probability of 

deviant occurrence on mismatch negativity. Cognitive Brain Research.  

Sakoe, H., & Chiba, S. (1971). A dynamic programming approach to continuous speech 

recognition. Proceedings of the Seventh International Congress on Acoustics. 

Sams, M., Paavilainen, P., Alho, K., & Näätänen, R. (1985). Auditory frequency 

discrimination and event-related potentials. Electroencephalography and Clinical 
Neurophysiology/ Evoked Potentials.  

Sancier, M. L., & Fowler, C. A. (1997). Gestural drift in a bilingual speaker of Brazilian 

Portuguese and English. Journal of Phonetics. 

Schouten, B., Gerrits, E., & van Hessen, A. (2003). The end of categorical perception as 

we know it. Speech Communication, 41(1), 71–80. 

Schrager, E. (1998). APPLICATION OF ERP MEASURES IN PSYCHOLOGICAL 

RESEARCH Measurement and interpretation of the mismatch negativity. Behavior 
Research Methods, Instruments, & Computers. 

Selinker, L. (1974). Interlanguage. International Review of Applied Linguistics in Leanguage 

 

114 

Teaching, 10(1), 209–232. 

Shestakova, A., Huotilainen, M., Ceponiene, R., & Cheour, M. (2003). Event-related 

potentials associated with second language learning in children. Clinical 
Neurophysiology, 114(8), 1507–1512.  

Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2011). False-positive psychology: 

Undisclosed flexibility in data collection and analysis allows presenting anything 
as significant. Psychological Science.  

Stanislaw, H., & Todorov, N. (1999). Calculation of signal detection theory measures. 
Behavior Research Methods, Instruments, & Computers: A Journal of the Psychonomic 
Society, Inc, 31(1), 137–149. 

Strange, W. (1995). Cross-langage studies of speech perception A historical review. 

Speech Perception and Linguistic Experience : Issues in Cross-Language Research. 

Strange, W., & Jenkins, J. J. (1978). Role of Linguistic Experience in the Perception of 

Speech. In Perception and Experience. 

Sussman, E., Winkler, I., & Wang, W. (2003). MMN and attention: Competition for 

deviance detection. Psychophysiology.  

Tavabi, K., Elling, L., Dobel, C., Pantev, C., & Zwitserlood, P. (2009). Effects of place of 

articulation changes on auditory neural activity: A magnetoencephalography 
study. PLoS ONE.  

Tervaniemi, M., Medvedev, S. V., Alho, K., Pakhomov, S. V., Roudas, M. S., Van Zuijen, 
T. L., & Näätänen, R. (2000). Lateralized automatic auditory processing of phonetic 
versus musical information: A PET study. Human Brain Mapping.  

The Mathworks Inc. (2018). MATLAB 2018a. 

Tiitinen, H., May, P., Reinikainen, K., & Näätänen, R. (1994). Attentive novelty detection 

in humans is governed by pre-attentive sensory memory. Nature.  

Waltmunson, J. C. (2005). The relative degree of difficulty of L2 Spanish / d , t /, trill , and tap 
by L1 English speakers: Auditory and acoustic methods of defining pronunciation accuracy. 
Doctoral dissertation, University of Washington. 

 

 

 

115 

Winkler, I., Kujala, T., Tiitinen, H., Sivonen, P., Alku, P., Lehtokoski, A., … Näätänen, R. 

(1999). Brain responses reveal the learning of foreign language phonemes. 
Psychophysiology, 36(5), 638–642. 

Winkler, I., Lehtokoski, A., Alku, P., Vainio, M., Czigler, I., Aaltonen, O.,  Naatanen, R. 

(1999). Pre-attentive detection of vowel contrasts utilizes both phonetic and 
auditory memory representations. Cognitive Brain Research, 7, 357–369. 

Yeni-Komshian, G. H., Flege, J. E., & Liu, S. (2000). Pronunciation proficiency in the first 

and second languages of Korean–English bilinguals. Bilingualism: Language and 
Cognition.  

Zampini, M. L. (1994). The Role of Native Language Transfer and Task Formality in the 

Acquisition of Spanish Spirantization. Hispania, 77(3), 470.  

Zanow, F., & Knösche, T. R. (2004). ASA - Advanced Source Analysis of continuous and 

event-related EEG/MEG signals. Brain Topography.  

Zue, V. W., & Laferriere, M. (2005). Acoustic study of medial /t,d/ in American English. 

The Journal of the Acoustical Society of America, 66(4), 1039–1050.  

 

 

116