Content uploaded by Jennifer Cabrelli
Author content
All content in this area was uploaded by Jennifer Cabrelli on Oct 14, 2017
Content may be subject to copyright.
The role of prosodic structure in the L2 acquisition of Spanish stop lenition (pre-proof version)
Jennifer Cabrelli Amaro, University of Illinois at Chicago
Published in Second Language Research, 33(2), 233-269.
Introduction
The study of Spanish spirantization (i.e., the weakening of an underlying stop consonant to an approximant
(/b d g/ ! [β̞ ð
̞ ɣ
̞])) is one of the most studied phenomena in L2 Spanish phonology (see e.g., Alvord and
Christensen, 2012; Face and Menke, 2009; Rogers and Alvord, 2014; Shea and Curtin, 2011; Zampini, 1997,
1998, inter alia). Spirantization has been found to be challenging for native English speakers (e.g., Díaz
Campos, 2004, 2006; Shively, 2008) since approximants are not allophones of /b d g/ in English, although it
has been shown that advanced learners produce continuants upwards of 80% of the time in postvocalic
position (Face and Menke, 2009). However, we do not fully understand how certain variables moderate the
path of the adult acquisition of spirantization, including proficiency and the operation of prosodic structure.
The purpose of this study will be to contribute to our understanding of how the interaction of these two
variables moderates acquisition.
Earlier studies of L2 spirantization examined learner productions at low levels of proficiency, and
therefore it was not possible to have a full view into the acquisition process. More recent studies have
observed cross sections of learners across proficiency levels and have provided acoustic evidence for the roles
of variables such as stress, place of articulation (POA), and word position in the acquisition process (e.g.,
Face and Menke, 2009; Rogers and Alvord, 2014; Shea and Curtin, 2011). Our primary interest in this study
is in the variable of word position. While Face and Menke (2009), Rogers and Alvord (2014), and Shea and
Curtin (2011) report differences in learner productions in word-initial versus word-medial position, little is
known about the specific role of prosodic structure in the acquisition of spirantization.
As a point of departure, we consider the analysis of impressionistic data from beginner and
intermediate L2 Spanish learners in Zampini (1997, 1998), which yielded an acquisition pattern whereby
learners acquired spirantization according to the prosodic hierarchy. The operation of phonological
hierarchies has been documented in both L1 acquisition (e.g., Branigan, 1976) and L2 acquisition (e.g.,
Cardoso, 2007). Specific to structural position, L1 phonological acquisition research indicates a relationship
between word position and consonant acquisition (e.g., Branigan, 1976; Kiparsky and Menn, 1977). While
there are no L1 studies to our knowledge that specifically examine word position and spirantization, it has
been shown, for example, that English-speaking children acquire fricatives first in word-medial position
(Edwards, 1979). As Vokic (2008) notes, structural position may also correlate with ease of acquisition in L2
learning. For example, Major (1986) indicates that perception of rhotics can vary according to structural
position, and reports that the few learners in his study that produced a native-like trill, did so only in word-
medial position. These acquisition patterns can be explained by the Subset Principle (e.g., Wexler and
Manzini, 1987), which states that acquisition occurs first at the most restrictive domain of the grammar (in
this case, word-medial position). With positive evidence, learners can then expand the domain to the superset
of the subset (e.g., Young-Scholten, 1994, Zampini, 1997, 1998). The L1 and L2 findings mentioned here,
taken together with the aforementioned studies of L2 Spanish spirantization and word position, warrant
further investigation to better understand the relationship between prosodic structure and L2 acquisition.
Herein, we provide acoustic evidence to inform the governance of prosodic structure over the path of
both categorical and gradient acquisition. A cross-sectional comparison of beginner, intermediate, and
advanced learners indicates that learners first produce underlying stops as postvocalic approximants at the
onset of the syllable (i.e., word-medial position), followed by the onset of the prosodic word (i.e., word-initial
position). Our findings only partially align with those of Zampini, whose data suggest that acquisition of
word-initial spirantization occurs first at the onset of the minimal projection of the prosodic word (i.e., across
the boundary of a lexical word and its clitic) and then at the onset of the maximal projection of the prosodic
word (i.e., across the boundary of two lexical words). As a potential explanation of the mechanisms that drive
acquisition, we model the categorical pattern found in our data within the framework of Optimality Theory
(Prince and Smolensky, 1993[2004]), following the assumption that the L2 learning task consists of the
gradual demotion of a set of prosodic positional faithfulness constraints below a markedness constraint that
prohibits voiced stops in postvocalic position.
The second contribution we aim to make regards ultimate attainment. In particular, we are interested
in a) how learners’ productions compare to those of native speakers in terms of degree of lenition and b)
residual variation.1 Our results show that advanced learners produce continuants in postvocalic position at all
of the applicable prosodic levels, which we take to indicate that this group has acquired the target constraint
ranking. However, the question remains as to whether learners’ productions, while not frequently stops, are
truly native-like. That is, do learners lenite segments to the same degree as the native speaker control group,
and does degree of lenition change across development? We will show that, for the learners that produce
continuants, there is gradual development of the acoustic target. Specifically, the difference in degree of
lenition between native speakers and learners lessens at the higher levels of the prosodic hierarchy as
acquisition progresses, and several advanced learners produce target-like segments across the three prosodic
levels. However, even these learners who consistently produce native-like continuants occasionally produce
stops. Does that mean that their constraint ranking is not native-like? Herein, we consider the possibility that
non-target-like variability and the acquisition of a native-like mental representation (i.e., constraint ranking)
1 We consider spirantization to be a type of lenition.
are not mutually exclusive. While there is general agreement in the field that virtually nothing in adult
language acquisition is categorical, the treatment of L2 ultimate attainment is often presented as a can/cannot
dichotomy: A learner either acquires a native-like L2 representation, or she does not (see e.g., Rothman, 2008,
for discussion). As a departure from this view, we demonstrate via the modeling of L2 phonological
acquisition in a stochastic (i.e., probabilistic) framework that adult L2 learners can converge on a target-like
constraint ranking and still occasionally produce outputs that are not target-like when crucial constraints
partially overlap. This study thus goes beyond the valuable first step that other studies have taken in
describing learner productions according to word position, by offering an explanation for patterns of
categorical and gradient acquisition as well as residual variation.
The remainder of this article is organized as follows: We begin with a description of postvocalic
underlying stops in Spanish followed by the constraint interaction that we adopt to explain how and in which
contexts these underlying stops are weakened. We then present an OT analysis of postvocalic stops in
English. Based on the Spanish and English analyses, we posit the L2 learning task and present our research
questions and predictions. In the sections that follow, we present the methodology for the study, followed by a
presentation of the results of our acoustic analysis. Finally, we discuss the implications of the data for
acquisition and phonological theory more generally and outline future directions for this line of inquiry.
Spanish postvocalic voiced stops
In Spanish, there is a stop/continuant alternation in which voiced stops [b d g] surface after nasal stops,
pauses, and, in the case of /d/, after /l/. On the other hand, continuants [β ð ɣ] (typically voiced approximants)
surface in other contexts. While there are a number of reports of dialectal variation informed by acoustic data
(e.g., Carrasco et al., 2012 for Costa Rican and Iberian Spanish; Colantoni and Marinescu, 2010, for
Argentine Spanish; Radu, 2014, for Colombian Spanish), continuants are thought to surface categorically in
postvocalic syllable-onset position across dialects, both within and across word boundaries. In the present
study, we focus on lenition in postvocalic syllable-onset position.
Following the hierarchy proposed by Nespor and Vogel (1986) that delimits the levels at which a
phonological phenomenon applies, Zampini (1997, 1998) determined that the prosodic domain within which
the process occurs is the intonational phrase (IP). That is, she showed that continuants surface in all instances
of post-vocalic stops within the IP and that stops surface at the left edge of the IP. As seen in examples 1a-d,
the same pattern occurs within the prosodic domains below it, including the phonological word (the terminal
element of a syntactic tree) and the clitic group (within the boundaries of a phonological word and its clitics).2
(1) Spanish stop/continuant alternation according to Nespor and Vogel’s (1986) prosodic hierarchy
a. At the left edge of the intonational phrase
[g] ~ [g]
gané la lotería [ga.ˈne la lo.t
̪e.ˈɾi.a] ‘I won the lottery’
b. Within the intonational phrase (across the boundary of lexical words)
[ɣ] ~ [g]
buena gafa [bu
̯e.na 'ɣ
̞a.fa] ‘good hook’
buen gato ['bu
̯eŋ ˈga.t
̪o] ‘good cat’
c. Clitic group (across the boundary of a functional and lexical word)
[ɣ] ~ [g]
la gafa [la 'ɣ
̞a.fa] ‘the hook’
un gato [uŋ 'ga.t
̪o] ‘a cat’
2 For support of the inclusion of a clitic group level in the prosodic hierarchy, see Hayes (1989) and Nespor
and Vogel (1986).
d. Phonological word (word-medial)
[ɣ] ~ [g]
agua ['a.ɣ
̞u
̯a] ‘water’
manga ['maŋ.ga] ‘sleeve’
Traditional generative analyses treat the stop/continuant alternation as categorical (see ‘Theoretical analysis’
for an OT account of this alternation). While context-dependent degree of lenition should not affect the
phonological representation as long as the segment is not produced with complete occlusion, there is ample
evidence that the degree of lenition in postvocalic position is variable. Acoustic studies (e.g., Carrasco et al.,
2012; Cole et al., 1999) have measured the intensity difference (in dB) of CV sequences, whereby a larger
intensity difference indicates a greater degree of occlusion of the consonant and a smaller intensity difference
indicates a greater degree of openness (i.e., lenition). The gradience of lenition is thought to be conditioned by
several factors, including POA of the segment, flanking vowels, lexical stress, and position within the word or
prosodic phrase (see e.g., Simonet et al., 2012, for a review). The last variable is of primary interest to the
current study. It has been found that segments are weaker in word-medial positions than in word-onset
position (Carrasco et al., 2012, for Costa Rican Spanish; Eddington, 2011 for /b/ and /d/), although Simonet et
al. (2012) did not find a difference in word position for Madrid Spanish and Rogers and Alvord (2014) did not
find a difference among speakers from three varieties of Latin American Spanish. To our knowledge, there are
no studies that address the role of segment position beyond a word-medial versus word-initial distinction.
Specifically, in word-initial position, we do not have acoustic evidence of whether the prosodic domain that
governs the word in question (i.e., the prosodic word or intonational phrase) conditions the degree of lenition.
This question is relevant to the present study since one of the central questions is whether learners acquire
spirantization according to the prosodic hierarchy. If learners’ degree of lenition correlates with the prosodic
hierarchy, it is possible that there are prosodic constraints at play that are part of the grammar. We now turn to
the presentation of the Optimality Theoretic analyses we use to make predictions for L2 development.
Theoretical analysis
As seen in the examples in (1), voiced continuants surface in postvocalic position in Spanish within the
intonational phrase. While several generative accounts of spirantization are based on assimilation (e.g.,
Goldsmith, 1981; Harris, 1969, 1984; Holt, 2002; Mascaró, 1984, 1991) or effort reduction (e.g., Kaplan,
2010; Kirchner, 1998, 2004; Piñeros, 2002; Recasens, 2002), we assume that the basis for stop lenition is
perceptual enhancement via sonority promotion (Kingston, 2007). Kingston puts forth the notion that lenition
of voiced stops, while not due to effort reduction, might be a result of undershoot of an articulatory target.
That is, a speaker shortens the closure duration to sustain vibration of the vocal folds. Maintenance of voicing
comes at the expense of closure of the oral cavity, resulting in an approximant. Kingston posits that whether a
stop lenites depends on prosodic structure, whereby lenition occurs more frequently inside a prosodic
constituent than at its edge. The weakening of a stop (which is less sonorous than an approximant and thus is
thought to interrupt the acoustic signal more) within the word or at the beginning of a word communicates to
the hearer that the word is within a prosodic constituent (p. 17).3 Since the domain of the stop/continuant
alternation in Spanish is the intonational phrase (IP; Zampini, 1997, 1998), we follow here that voiced stops
will surface only at the left edge of an IP.
Herein, we adopt a modified version of Kingston’s (2007) constraint-based analysis.4 This analysis
consists of a (prosodic) positional faithfulness constraint (Beckman, 1997) that outranks a more general
3 See Harris (2003) for a similar analysis.
4 The modification comes in the form of the markedness constraints. In his analysis, Kingston addresses an
instance of lenition in which the input is [-cont] and the output is [+cont]. While this is adequate for Spanish
spirantization, since we are working with L1 English/L2 Spanish speakers, it is ideal to adopt an analysis that
also accounts for English lenition of voiced stops (i.e., flapping of /d/), a case in which both the input and
output are [-cont].
faithfulness constraint and contextual markedness constraints. The general faithfulness constraint is IDENT-IO
(CONT) and the positional faithfulness constraint IDENT-IO (CONT) ONSET IP dictates preservation of identity
at the left edge of an IP.
IDENT-IO (CONT)
The specification for the feature [continuant] of an input segment must be preserved in its output
correspondent.
IDENT-IO (CONT) ONSET IP
The specification for the feature [continuant] of an input segment at the onset of an intonational
phrase must be preserved in its output correspondent.
The markedness constraints prohibit a group of segments from surfacing in a particular context. This set of
constraints forms a hierarchy which aligns with the sonority scale such that the highest-ranked constraints
refer to groups of segments that are the lowest in sonority (i.e., voiceless stop >> … glide). The two
markedness constraints that play an important role in our analysis of Spanish spirantization refer to a) voiced
stops and b) approximants, with the former outranking the latter.
*VOICED STOP/V_5
There are no postvocalic voiced stops in the output.
*APPROXIMANT/V_
There are no postvocalic approximants in the output.
5 These markedness constraints typically indicate the context as X_Y, with X and Y specifying minimal
sonority of the flanking segments. Since the relevant context for spirantization is postvocalic, we use the
shorthand notation ‘V_’ throughout.
In Spanish, the high ranking of *VOICED STOP/V_ in relation to the IDENT-IO (CONTINUANT) forces
continuants to surface in postvocalic position. If the positional faithfulness constraint IDENT-IO
(CONTINUANT)IP dominates the markedness constraint *VOICED STOP/V_ (2), postvocalic approximants will
surface within the IP (Table 1) and stops will surface at the onset of the IP (Table 2).
(2) Spanish ranking:
IDENT-IO (CONT) ONSET IP >> *VOICED STOP/V_ >> IDENT-IO (CONT) >> *APPROXIMANT/V_
Table 1. Postvocalic continuant output in Spanish within the IP.
/agua/6
IDENT-IO (CONT)
ONSET IP
*VOICED STOP/V_
IDENT-IO (CONT)
*APPROXIMANT/V_
! ['a.ɣ
̞u
̯a]
*
*
['a.gu
̯a]
*!
Table 2. Postvocalic continuant output in Spanish at the left edge of the IP.
/buena/IP
IDENT-IO (CONT)
ONSET IP
*VOICED STOP/V_
IDENT-IO (CONT)
*APPROXIMANT/V_
[ˈβ̞u
̯e.na]
*!
*
*
! [ˈbu
̯e.na]
6 While several constraint accounts propose that the input is a [+continuant] segment (e.g., Barlow, 2003; de
Lacy, 2001; Maddieson, 2011; Smith, 2005), we assume a [-continuant] input based on Piñeros (2003). His
analysis is based on the assumption that IDENT (FEATURE) (input and output correspondents must agree in
their specifications for all features) constraints outrank the constraint *VOICED STOP (voiced obstruent stops
are prohibited). This assumption has recent empirical support in the form of developmental data (Fabiano-
Smith et al., 2015), for which children show higher accuracy on voiced stops than continuants even at age 8;2.
L2 acquisition of Spanish spirantization
Having established an L2 target, we can proceed with a discussion of the L2 learning task and predictions for
the current study. We start with an analysis of English postvocalic stops to establish the constraint reranking
task, followed by a description of an algorithm used to model phonological acquisition. We then conclude
with a brief review of the L2 literature that informs our study.
English postvocalic stops
The phonemic inventory of American English contains voiced stops /b d g/ in simple and complex onset and
coda position, as well as intervocalically (Celce-Murcia et al., 1996). In intervocalic position, the output of /b/
and /g/ is faithful to the input. However, /d/ surfaces as an alveolar flap ([ɾ]), a voiced consonant without a
release burst, violating IDENT-IO (FLAP)7 (e.g., Piñeros, 2003).
IDENT-IO (FLAP)
The specification for the feature [flap] of an input segment must be preserved in its output
correspondent.
(3) American English postvocalic voiced stops
[b] ‘baby’ [ˈbeɪ.bɨ]
[d] ~ [ɾ] ‘daddy’ ['dӕ.ɾɨ]
[g] ‘muggy’ ['mʌ.gɨ]
7 Bouavichith and Davidson (2013), Warner and Tucker (2011), inter alia, show that English speakers lenite
stops in fast speech, particularly in unstressed syllables (Bouavichith & Davidson, 2013). However,
Bouavichith and Davidson conclude that this phonetic weakening is not phonological in nature, as evidenced
by the prioritization of maintenance of acoustic cues to stress by English speakers. Even in the environment
most susceptible to weakening (unstressed position), 49% of underlying intervocalic voiced stops still surface
as stops. Moreover, the data that point to weakening come from tasks that are less formal than the recitation
task that we use in the present study. As a result, we assume that intervocalic /b g/ lenition is not a
phonological process in English and therefore do not presume that lenition in Spanish production is
transferred from English.
Flapping is a type of stop lenition in which the period of closure is shortened (e.g., Herd, Jongman, and
Sereno, 2010). Flaps are higher in sonority than their stop counterpart, and have been shown to have higher
root mean square (RMS) amplitude than stops (Herd et al., 2010; Lavoie, 2000). Bonet and Mascaró (1997)
propose that a flap has sonority akin to that of glides. Therefore, it does not violate *VOICED STOP/V since the
markedness constraints employed in this analysis are sonority-based. Rather than conflating a flap and a glide
(which differ in terms of [continuant]) and assuming that a flap violates *GLIDE/V_, we propose that a flap in
the output violates *FLAP/V_ (C. Wiltshire, personal communication, November 19, 2015).
*GLIDE/V_
There are no postvocalic glides in the output.
*FLAP/V_
There are no postvocalic flaps in the output.
Just as the feature constraint IDENT-IO (CONT) dominated by *VOICED STOP/V_ yields a continuant output, the
domination of IDENT-IO (FLAP) by *VOICED STOP/V_ yields a postvocalic flap. The feature [continuant] is not
relevant here, since both stops and flaps are [-continuant]. Because there are no stop/continuant input/output
relationships in English, IDENT-IO (CONT) is undominated (Table 3). Note that IDENT-IO (CONTINUANT)
ONSET IP is not part of the English ranking since there is no evidence that it is undominated in English.
(4) English ranking:
IDENT-IO (CONT) >> VOICED STOP/V_ >> IDENT (FLAP), *FLAP/V_8
8 We assume that IDENT(FLAP) and *FLAP/V_ do not have a strict ranking relationship since the same
candidate wins regardless of the constraints’ relative order.
Table 3. English postvocalic /d/ in the word ‘daddy’.
/dӕdi/
IDENT (CONT)
*VOICED STOP/V_
IDENT (FLAP)
*FLAP/V_
[ˈdӕ.dɨ]
*!
! [ˈdӕ.ɾɨ]
*
*
[ˈdӕ.ð
̞ɨ]
*!
Now that we have established the Spanish and English constraint interactions that yield the surface outputs
attested in each language, we can define the L2 Spanish initial state and outline the L2 learning task.
Gradual acquisition and the L2 learning task
Due to the lack of a stop-continuant alternation in English, learners first have to acquire the Spanish surface
distribution of stops and continuants. While the degree of lenition will vary across dialects and speech styles,
learners will minimally have to determine that underlying stops are produced as continuants in postvocalic
position. In the following sections, we describe the mechanism that we assume to underlie phonological
acquisition, followed by the learning task and the potential role of the prosodic hierarchy in the learning task.
The Gradual Learning Algorithm. In our analysis, we assume that there is a learning mechanism that drives
allophonic acquisition. We follow Boersma (1997) and Boersma and Hayes (2001), who posit that acquisition
of the distribution occurs via an error-driven learnability algorithm called the Gradual Learning Algorithm
(GLA). The GLA is based on stochastic (i.e., probabilistic) OT and simulates both L1 and L2 phonological
acquisition. Boersma and Hayes claim that each constraint has an absolute ranking value and occupies a range
on a continuous scale, which they interpret as a probability distribution. The amount of space between
constraints indicates the strictness of the relative ranking. During production, selection points (i.e., actual)
ranking values from within the range that a constraint occupies) are picked and the constraints are ranked
according to these points. If this constraint ranking is a mismatch with that of the surface form, the algorithm
goes to work to alter the constraint ranking values so that the grammar will be more likely to generate the
optimal candidate the next time. Since constraints’ probability distributions can continuously be adjusted, the
GLA predicts L2 convergence upon a native-like ranking, although it is of course possible that variable
outputs can persist in the case of two constraints with ranking values that are close together (just as they do, to
a lesser extent, in native grammars).9 While the assignment of actual ranking values to the constraints in our
analysis is outside the scope of this paper, we will demonstrate in the results and discussion that the tenets of
the GLA presented here are attestable.
In the only study that we know of that has formally modeled L2 acquisition of Spanish postvocalic
continuants based on acoustic data, Shea and Curtin (2011) set out to identify acquisition as categorical or
gradient. In their investigation of low intermediate and high intermediate learners, they hypothesized that
uniform acquisition across places of articulation (POA) would support a categorical generative account (either
rule-based or constraint-based). On the other hand, if learners were to acquire the alternation at different rates
according to POA, the authors proposed that such gradient acquisition would favor a gradual input-based
system. A difference in high intermediate productions according to POA led Shea and Curtin to claim support
for an exemplar-based model in which frequency of input drives acquisition under the assumption that
gradient acquisition and a generative approach are incompatible. However, frequency is a key component in
the GLA and a mismatch between aural input and a current constraint ranking drives acquisition. Gradient
9 An anonymous reviewer brings up the question of how the GLA accounts for fossilization (i.e., lack of
ultimate convergence on the L2 target). Although Boersma and Hayes do not explicitly address the question,
we assume that fossilization occurs as a result of a lack of parsing failures. That is, learners continue to use L1
parsing routines and “map interpretations onto the stimuli and ignore unanalyzable parts of speech” (Carroll,
2001, p. 305).
acquisition is not only compatible with OT, but is in fact predicted under the GLA. Depending on the
constraints at play, rate and path of acquisition could indeed vary according to the frequency according to
POA in the input. We therefore posit that the GLA can also explain Shea and Curtin’s (2011) data.
The constraint reranking task. Assuming that learners’ L2 initial state is a copy of the L1 phonological
grammar, L1 English/L2 Spanish learners come to the task with a ranking in which the faithfulness constraint
IDENT-IO (CONT) is undominated. Thus, to produce Spanish postvocalic underlying stops as continuants, the
faithfulness constraint IDENT-IO (CONT) will have to be demoted below *VOICED STOP/V_ so that the
markedness constraint can block a postvocalic voiced stop from surfacing. Initially, learners will evaluate
Spanish surface candidates against their L1 rankings. It is predicted that since the L2 input will violate
undominated IDENT-IO (CONT), EVAL (evaluator, the mechanism that determines the relationship between
the candidates and the universal constraint set) will initially rule out the Spanish candidates with fatal
violations. As input increases, however, the GLA will successively alter the probability distributions until the
target ranking of *VOICED STOP/V_>> IDENT-IO (CONT) is reached.
The role of the prosodic hierarchy in the L2 learning task. As we have noted, recent L2 research has
considered effects of position on the production of postvocalic underlying stops, finding segments to be more
approximant-like in word-medial versus word-initial context even at lower levels of proficiency (Rogers and
Alvord, 2014; Shea and Curtin, 2011). These authors show that although position distinctions are maintained
across proficiency levels, there is an overall reduction in intensity difference, which indicates progression
towards convergence on the native-like target. Zampini (1997, 1998) also observed a pattern with regard to
the position of the segment, finding that production by beginner and intermediate learners differed according
to position in the prosodic hierarchy as a function of proficiency. In light of Zampini’s impressionistic
evidence that learners acquire postvocalic spirantization according to different domains of the prosodic
hierarchy, we hypothesize that there could be additional constraint interactions that go beyond those already
described. Herein, we present these constraints, working from Itô and Mester’s (2007) more recent
instantiation of the prosodic hierarchy for reasons of economy, and an ONSET-driven analysis that is
compatible with the positional faithfulness approach to spirantization.
Itô and Mester (2007) address several proposals of the levels to be included in the prosodic
hierarchy, including that of Nespor and Vogel (1986). They posit that a more minimal framework with fewer
categories is necessary, in which prosodic adjunction is central in the parsing of phonological strings (Figure
1).
Figure 1. Prosodic hierarchy (Selkirk, 1986).
In the spirit of syntactic constituent structure, each prosodic category has a maximal and minimal projection
and has the option of recursivity. To illustrate, Figure 2 demonstrates that for a prosodic category (in this case,
that of the prosodic word), the larger structure is the maximal projection. The structure consists of the
innermost subconstituent (the lexical word) and the functional words that adjoin to it, and the innermost
subconstituent is the minimal projection.
Figure 2. Minimal and maximal projections of the prosodic word (Itô and Mester, 2007).
In this analysis, function words occupy positions within the maximal prosodic words. Itô and Mester (2007)
provide support for this analysis via a discussion of onset requirements for the levels of the prosodic hierarchy
and their faithfulness properties, and argue that this economical approach yields flexibility for whatever
distinctions are necessary without the noise of additional categories such as the clitic group (Nespor and
Vogel, 1986) and intermediate and accentual phrases (Pierrehumbert and Beckman, 1988). Taking this idea of
onset markedness and faithfulness at the different prosodic levels as a point of departure, we can illustrate
Spanish spirantization across prosodic levels in terms of the location of the onset that surfaces unfaithfully in
postvocalic onset position. Word-medially, a lenited postvocalic onset is dominated by the syllable,
demonstrated in Figure 3 with the word agua ‘water’.
Figure 3. Prosodic level of the syllable; dominates word-medial syllable onset.
Across the boundary of a lexical word and the function word(s) adjoined to it, the onset of the minimal
projection of a prosodic word is lenited. This is seen in Figure 4 with la gafa ‘the hook’.
Figure 4. Minimal projection of the prosodic word; dominates word-initial onset across the boundary of a
functional and lexical word.
Finally, across the boundary of two lexical words, the onset of the maximal projection of a prosodic word is
lenited, as in Figure 5 with buena gafa ‘good hook’.
Figure 5. Maximal projection of the prosodic word; dominates word-initial onset across the boundary of two
lexical words.
Based on Zampini’s (1997, 1998) findings, it is predicted that the learners will start with the most restrictive
setting for the alternation on the prosodic hierarchy (in this case, syllable onset). As previously mentioned in
‘Introduction’, this prediction follows from the Subset Principle (Dell, 1981; Wexler and Manzini, 1987),
which assumes that the learning mechanism maps input to the most restrictive (subset) grammar. The Subset
Principle was developed to account for L1 acquisition patterns and has been found to be operative in L2
acquisition as well, particularly when the L2 learning task consists of acquiring a superset grammar and not a
subset grammar (see Kilpatrick, 2009, for discussion). In the case of L2 Spanish spirantization, the most
restrictive grammar is the lowest level of the prosodic hierarchy at which spirantization occurs, which is the
onset of the syllable. With positive evidence, learners will incrementally extend the domain of spirantization
to less restrictive (superset) grammars, first in word-initial position across the boundary of a word and its
clitic and then across the boundary of two lexical words. If we find that L2 Spanish learners acquire
postvocalic spirantization according to prosodic level, we can posit that there are separate stricture
faithfulness constraints for the onset of each prosodic level and that learners gradually demote onset stricture
faithfulness constraints. With evidence, they will extend the domain of spirantization until they reach a native-
like ranking in which *VOICED STOP/V_ outranks all onset stricture faithfulness constraints. If our data
support Zampini’s findings, the following onset stricture faithfulness constraints are thus predicted to be part
of the ranking:
IDENT-IO (CONT) ONSET SYLL
The specification for the feature [continuant] of an input segment at the level of the syllable must be
preserved in its output correspondent.
IDENT-IO (CONT) ONSET PW-MIN
The specification for the feature [continuant] of an input segment at the level of the minimal
projection of the prosodic word must be preserved in its output correspondent.
IDENT-IO (CONT) ONSET PW-MAX
The specification for the feature [continuant] of an input segment at the level of the maximal
projection of the prosodic word must be preserved in its output correspondent.
Under this analysis, IDENT-IO (CONT) will be the first constraint to be demoted because it is violated every
time a continuant appears in the evidence, independent of the phonological domain. Given that each level of
the prosodic hierarchy subsumes the levels below, a violation at a higher level will mean a violation at each of
the levels below. As such, the constraints will be demoted at different rates according to the number of
violations incurred and are demoted in order from the lowest level to the highest level of the hierarchy at
which continuants are evidenced. As the learner is exposed to surface forms that the learner’s current ranking
would not yield, IDENT-IO (CONT) ONSET SYLL will demote, followed by IDENT-IO (CONT) ONSET PW-MIN,
and finally by IDENT-IO (CONT) ONSET PW-MAX. By demoting the stricture faithfulness constraints below
*VOICED STOP/V_, learners are predicted to converge on the target grammar. Assuming the GLA, reranking is
gradual and constraints are expected to overlap throughout the acquisition process even in the grammars of
near-native speakers that have acquired the target ranking. As discussed, overlap of crucially ranked
constraints can lead to variable outputs, and multiple constraints can overlap one another. As a result, upon
convergence it is highly probable that an underlying postvocalic stop surface as a continuant regardless of the
prosodic level in which the stop occurs.
Convergence on the L2 acoustic target. While we posit that constraint reranking is a critical component of L2
acquisition of Spanish spirantization, there is another component involved. As mentioned, if a learner’s
constraint ranking reflects the target Spanish ranking, it can be expected that a learner will reliably produce
postvocalic continuants. However, that does not mean that a learner’s productions will reflect native-like
degree of lenition. Therefore, the other aspect of the learning task is to develop the acoustic target such that
occlusion is reduced to the same extent as a native speaker’s productions.
Research questions. Recall from the introduction that the goals of this study are twofold. First, we aim to
account for the acoustic patterns observed at different levels of proficiency across domains of the prosodic
hierarchy. Our second goal is to determine whether late L2 learners can revise an L1 ranking in their L2, and
to illustrate that a native-like ranking/acoustic target and non-target-like productions are not mutually
exclusive. To meet these objectives, we put forth the following questions:
1a. Can learners acquire spirantization in L2 Spanish across all of the relevant levels of the prosodic
hierarchy?
1b. Does the acquisition of spirantization in L2 Spanish occur in stages according to the prosodic hierarchy?
That is, do learners first produce postvocalic underlying stops as continuants at the onset of the syllable,
followed by the onset of the minimal and maximal projections of the prosodic word, respectively?
Face and Menke (2009) present categorical data from 13 PhD students that show development
towards a native-like representation of postvocalic underlying stops in word-medial (85%) and word-initial
(73%) positions, although learners still produced significantly more stops in word-initial position than in
word-medial position. Comparing the PhD students to fourth semester Spanish undergraduate students and
Spanish majors, there was an overall reduction of stops produced across proficiency levels while all groups
maintained a significant distinction between word positions. Therefore, additional data are warranted to
determine whether and how the prosodic hierarchy moderates the learners’ acquisition path, and whether
learners can reliably produce continuants even at the onset of the PW-max.
2a. For the learner groups that produce approximants at each prosodic level, is the degree of lenition similar to
that of the native controls' approximants?
2b. Is there evidence of an increase in degree of lenition as proficiency increases?
To answer this question, an analysis of intensity ratio or difference is necessary. Two L2 acquisition
studies have used this method: Shea and Curtin (2011) and Rogers and Alvord (2014). Shea and Curtin report
a significant difference in lenition in word-medial position when comparing low intermediate learners, high
intermediate learners, and native speakers. This finding indicates an increase in degree of lenition as
proficiency increases, at least in word-medial position, although degree of lenition is not native-like. The
authors do not report between-group pairwise comparisons for word-initial position; however, descriptively
we see an increase in degree of lenition in word-medial and word-initial position when comparing the two
groups. Based on their word-medial pairwise comparison, we assume word-initial lenition is not native-like.
Rogers and Alvord (2014) found that a group of advanced learners’ (n=4) that had spent two years abroad
produced postvocalic segments with a degree of lenition that was descriptively similar to that of three native
speakers in word-medial and word-initial positions. They also compared the degree of lenition between word
positions for the advanced learners and learners that had completed two years of university Spanish (n=4), and
indicate an overall reduction of occlusion in both positions. We anticipate that the examination of a larger
sample size and the use of inferential statistics will further inform our understanding of the development of
learners’ acoustic targets.
Methodology
As stated in the introduction, spirantization is one of the most studied phenomena in L1 English/L2 Spanish
phonology research, although a review of the literature shows a need for further empirical inquiry to better
understand acquisition from the initial state to L2 target convergence. Specifically, there is a lack of 1) data
from advanced/near-native learners and control participants, 2) proficiency assessment, and 3) acoustic
analysis from both gradient and binary accounts of stricture. We address these issues throughout the section.
Participants
The cross-sectional study consisted of three learner groups with a mean age of acquisition of Spanish of 13
years (SD = 1.98) at the beginner (n = 11), intermediate (n = 13), and advanced (n = 20) proficiencies, and a
control group of native speakers (n = 5).
Table 4. Participant age, gender, and proficiency scores.
Group
age (M (SD))
Gender
M (SD)
proficiency
score (out of
50)
M
F
Beginner
21 (.69)
0
11
25.09 (2.87)
Intermediate
21 (.74)
4
9
33.15 (2.51)
Advanced
23 (3.99)
6
14
44.2 (3.03)
Control
21 (1.78)
2
3
N/A
The inclusion of advanced learners and a control group addresses a gap in the literature. While a considerable
body of research reports persistent difficulty in target-like productions (e.g., Díaz-Campos, 2004, 2006; Elliot,
1997; Shively, 2008) with only González Bueno (1995) and Lord (2005) evidencing target production by
intermediate learners surpassing the 50% mark, few studies (Alvord and Christensen, 2012; Face and Menke,
2009; Rogers and Alvord, 2014) test advanced speakers and ultimate attainment to determine if and when
postvocalic underlying stops surface as continuants. However, in the Face and Menke (2009) study, the most
advanced group of participants was categorized based solely on their standing as students in a Spanish PhD
program, and the authors recognize overlap in the learner profiles from the three groups tested (fourth
semester students, graduating Spanish majors, and PhD students). While the PhD students produced
approximants 82% of the time, significantly more than the other groups, there is no control data with which to
make a statistical comparison. Similarly, Alvord and Christensen (2012) reported an 81% accuracy rate in
intervocalic approximant production by returning missionaries that had spent two years abroad, but there was
no control group as a comparison (although the authors claim that spirantization is categorical, p. 22). To our
knowledge, only Rogers and Alvord (2014) have tested near-native speakers (n = 4) and compared them with
native controls (n = 3) using a continuous measurement. However, the authors were unable to statistically
compare the learners and controls due to small sample sizes.
The control group in the present study (n = 5) was composed of native speakers of Cuban Spanish
that had moved to the US and acquired English after late childhood, rather than monolingual Spanish
speakers. The speakers had lived in the US for an average of 5.3 years (SD = 2.74), and had demonstrated
sufficient English proficiency to study in American universities at the undergraduate level. The use of a
mirror image of the experimental group follows Hopp and Schmid (2013), who posit that a monolingual
native norm is not an appropriate measurement of L2 ultimate attainment. It has been established that a
bilingual speaker is not equivalent to two monolinguals, even in the case of ‘balanced’ bilinguals (see e.g.,
Grosjean, 1998). Therefore, “the choice [to use monolingual productions]…serves to move the yardstick of
nativelikeness to a point which may, by definition, be out of reach for most bilinguals” (Hopp and Schmid, p.
354). To this end, Hopp and Schmid suggest that a more appropriate control group is a mirror image of the
experimental group. That is, if the population under investigation is composed of L1 English/L2 Spanish
speakers, the control group should consist of L1 Spanish speakers that have acquired L2 English after a
certain age and are living in an immersion context. We acknowledge, however, that there is a disparity in
usage between the control and learner groups given that the learners are not in an L2 immersion context.
Proficiency measurement
All participants completed a background questionnaire to assess age of acquisition, length of residence in
Spanish immersion environments, formal education in Spanish, and self-reported use (context/frequency) of
Spanish. Participants did not report fluency in any languages other than English and Spanish. Most studies do
not apply an objective proficiency assessment, instead classifying learners according to course or program
enrollment (e.g., Face and Menke, 2009; González Bueno, 1995; Rogers and Alvord, 2014; Shea and Curtin,
2011; Zampini, 1997, 1998). This practice can be problematic considering the variation that can be found
even when proficiency is carefully controlled for via independent assessment. To address this issue,
proficiency was assessed via a 50-item test composed of portions of the Diploma of Spanish as a Foreign
Language (DELE) and Modern Language Association (MLA) Spanish proficiency exams. This test has been
used extensively in generative L2 Spanish research (see, e.g., Montrul, 2008), and consists of a passage-based
cloze section (30 items) and sentence-based cloze section (20 items). As in other studies (e.g. Giancaspro,
2013, Slabakova and Montrul, 2003, inter alia), learners with a score of 40-50 were classified as advanced,
learners with a score of 30-39 were classified as intermediate, and learners with a score below 30 were
classified as beginner.
Task and stimuli
Participants completed a sentence recitation task which consisted of 54 tokens in the carrier phrase Diga
______, por favor ‘Say ______, please’ (Appendix 1). The task was selected to control for task effects (see
e.g., Radu, 2014; Zampini, 1994) and speech rate. Following Shea and Curtin (2011), items differed in terms
of POA of the underlying stop (/b d g/), following vowel (/a i u/), and stress (stressed/unstressed). While
Shea and Curtin also investigated word position (word-initial/word-medial), our items were designed to be
equally distributed across the three onset positions introduced in ‘The role of the prosodic hierarchy in the L2
learning task’ (syllable (word medial), minimal prosodic word (across the boundary of a word and its clitic),
maximal prosodic word (across lexical word boundaries), see (5) for example stimuli).10,11 All tokens from the
10 In our discussion of the acquisition of spirantization in this paper, we focus on postvocalic productions and
do not report on data from contexts that favor stop productions. However, data from a study in preparation by
Authors show that advanced learners that pattern with native controls in postvocalic position, also pattern
with controls in that they produce stops in post-pausal and post-nasal positions.
syllable onset condition (word-medial position) were trisyllabic or quadrisyllabic, and the critical segment
was never in the final syllable (Shea and Curtin, 2011). To summarize, the stimuli comprised a 3 (POA) ×
vowel (3) × stress (2) × prosodic level (3) design. However, due to researcher error, three conditions at the
onset of the syllable (stressed /ba/, stressed /bi/, stressed /da/), and three at the onset of the maximal prosodic
word (unstressed /gu/, stressed /du/, stressed /gi/) were excluded from analysis. Since contextual vowels were
not considered a factor in our analysis, our error does not necessarily impact the results negatively.
(5) Example stimuli by prosodic level (critical segment in bold)
a. Onset of the syllable (word medial)
duda /duda/ ‘doubt’
b. Onset of the minimal projection of the prosodic word (across the boundary of two lexical words)
la duda /la duda/ ‘the doubt’
c. Onset of the maximal projection of the prosodic word (across the boundary of two lexical words)
es una vida digna /es una bida digna/ ‘(it) is a dignified life’
Frequency was controlled for, using only stimuli from the 5000 most frequent Spanish words (taken
from Davies, 2006). Past participle forms were excluded, as they have been found to be more lenited due to
frequency effects (e.g., Díaz-Campos and Gradoville, 2011; Eddington, 2011). We also excluded items with
orthographic ‘v’, as its inclusion has been found to yield fricative production in L2 studies (Face and Menke,
2009; Zampini, 1994) and studies of heritage speaker Spanish (Fabiano Smith et al., 2014; Ronquest, 2014).
11 Although our primary interest is in how the position of an underlying stop conditions lenition across L2
development, we incorporated POA, following vowel, and stress into our design to add meaningful variability
to the dataset in terms of contextual factors found to influence native speech. Given the focus of the current
study, however, we limit our discussion to the variable of prosodic level.
Procedure
All communication with the researchers was conducted in Spanish to avoid language mode effects.
Participants were asked to read the list of sentences at a normal rate of speech and to pause for two seconds
between each sentence. They completed the list of words twice to control for a novelty effect, and the second
reading was used for analysis. All recordings were made in a quiet room using a Marantz PMD660 recorder
and a head-mounted Shure SM10A microphone at a sampling rate of 44.1 kHz.
Analysis
Most existing L2 work has been based on an impressionistic binary analysis of stop versus approximant
production, with few exceptions. While Face and Menke (2009) and Alvord and Christensen (2012) analyzed
acoustic data, they coded their data only for manner of articulation. Our analysis methodology aligns with that
of Shea and Curtin (2011) and Rogers and Alvord (2014), who are the only L2 researchers to investigate the
phenomenon via acoustic analysis of the intensity curve, in line with evidence from monolingual data as
described earlier.
Measurements of intensity (an acoustic index of lenition, Parker, 2002, 2008) were used to determine
the degree of constriction of each segment. Specifically, the minimum intensity (measured in decibels, dB) of
the consonant in a CV syllable was subtracted from the maximum intensity of the following vowel (e.g.,
Hualde et al., 2011; Soler and Romero, 1999). Intensity difference was measured to control for variation
across speakers and recordings. The lower the intensity difference, the more open the constriction (i.e., the
higher the degree of lenition).
To measure intensity difference, each underlying stop and following vowel was segmented in Praat
(Boersma and Weenink, 2014). We followed Lavoie (2001) and Shea and Curtin (2011), marking the onset of
the consonant as the offset of the previous vowel’s second formant (F2) and the offset as the onset of the
following vowel’s F2. F2 onset and offset was used to segment the following vowel segment. In the case that
F2 was not a clear enough indicator, we followed Hualde et al. (2011), marking the onset of the underlying
stop at the location of a dip in Praat’s intensity curve. The offset of the consonant was then marked at the
location of an increase in the waveform’s amplitude. Once all segments were marked, we ran a script in Praat
that calculated the difference (in dB) between the maximum intensity of the vowel and the minimum intensity
of the consonant. A total of 2307 items were analyzed. Figure 6 shows a continuant production from a control,
and Figure 7 shows a stop production from a beginner.
Figure 6. Continuant production of /b/ in hábito ‘habit’ by a native control participant (intensity difference =
2.85 dB). Note: Transcription is phonemic.
Figure 7. Stop production of /b/ in hábito ‘habit’ by a beginner participant (intensity difference = 21.16 dB).
Results
To examine the relationship between proficiency and prosodic level that conditions the production of
postvocalic underlying stops in L2 Spanish, we report descriptive and inferential statistics. Our descriptive
statistics include means, 95% confidence intervals, and effect sizes (Hedges’ g). We also computed a Linear
Mixed Model in SPSS using the MIXED procedure. The dependent variable was Intensity Difference,
measured in decibels (dB). The between-subjects effect was Group (control, beginner, intermediate,
advanced). Because existing experimental data shows that native Spanish speakers categorically produce
continuant in postvocalic position (see section on Spanish postvocalic voiced stops), the native speaker group
data provide an appropriate baseline for comparison with the learner data for determining whether learner
productions are lenited to the same degree as the native control. The within-subject fixed effect was Prosodic
Level (onset of syllable, onset of PW-min, onset of PW-max). We included the Prosodic Level×Group
interaction in the model and planned pairwise comparisons for the interaction (Bonferroni adjustments to
observe differences in means and TOSTs to observe similarity in means).12 A random intercept and a random
slope for prosodic level were included at the participant level. We also included a random intercept at the item
level in order to capture variance in the dependent variable across items. Alpha was set at p =.05. Significant
main effects were found for Group (F (3, 45.05) = 10.67; p =.000) and Prosodic Level (F (2, 91.07) = 17.59;
p =.000). Considering the questions that drive the study, the significant interaction of Prosodic Level×Group
(F (6, 87.363) = 3.348; p =.005) is of primary interest (see Figure 8 for a visual presentation of the
interaction). Looking at the figure, there are two observable patterns. First, the intensity difference increases
across prosodic levels for all groups except for the advanced group. Second, the intensity difference of the
spirants for each group varied as a function of proficiency group; the control group has the lowest intensity
difference values and the beginners have the highest ones.
We will explore this interaction of proficiency and prosodic level descriptively and inferentially throughout
the rest of the section as we address the relevant results for each research question.
12 To our knowledge, equivalence tests as a method of similarity testing have not been used in SLA research
before (see Juzek and Kizach, submitted), and therefore a brief explanation is warranted. The equivalence test
we have implemented consists of two one-sided t-tests (TOST) and is attributed to Westlake (1976) and
Schuirmann (1981). The null hypothesis assumes a difference in means and a rejection of the null hypothesis
allows a conclusion that the means are equivalent. The first test evaluates the difference in means plus delta
(absolute difference) and the second test evaluates the difference in means minus delta. If both t-tests yield
positive results, the means are concluded to be similar. The reason TOSTs have not been used in linguistic
research before now is the difficulty in setting δ, which is the critical parameter that controls the TOST. Juzek
and Kizach (submitted) outline how to objectively set δ “to achieve the narrowest similarity range that meets
common error rates of 1 − β = 80% and 1 − α = 95% (where β is the rate of false negatives and α the rate of
false positives)” (p. 2). We have used Juzek’s (2016) web-based TOST calculator, which sets δ and returns a
result of positive (p < .05) or negative (p > .05).
Figure 8. Intensity difference (in dB) according to group and prosodic level. Error bars represent standard
error.
Research question 1
To determine a) whether learners acquire spirantization in L2 Spanish across all of the relevant levels of the
prosodic hierarchy and b) if the hierarchy moderates the path of acquisition, it was first necessary to see
whether the learner groups produced stops or continuants segments at each level. While our dependent
variable (intensity difference) is continuous, we categorized each mean as a stop or continuant using one-
sided 97.5% confidence intervals.13 A plausible lower limit of intensity difference for a stop was calculated by
measuring intensity difference in a pseudorandom sample distributed among the variables crossed in the
stimuli set of 150 stops from the learner data. Tokens came from 38 learners, although there was naturally a
higher ratio of beginner and intermediate tokens since they produced a larger number of stops.14 A segment
was classified as a stop based on the presence of a release burst in the spectrogram. Upon noting the
differences in mean intensity difference across prosodic levels, we treated the data points at each prosodic
level separately (n = 50 each). Items with an intensity difference value larger than 2 SDs above the mean for
the relevant prosodic level were removed from the data set, leaving 47 items at the syllable and PW-min
levels and 48 at the PW-max (n = 142). T-tests comparing the stop data at each prosodic level revealed a
significant difference between the syllable and PW-min (p =.003) and PW-max (p =.000), and between the
PW-min and PW-max (p =.001). In light of these results, we constructed separate CIs for the syllable (M =
13.48, SD = 3.94), PW-min (M = 16.07, SD = 4.42), and PW-max (M = 20.08, SD = 7.28). Table 5 details a)
the CI of the segments with release bursts for each prosodic level and b) the mean intensity difference and CI
for each learner group and the control group. Cells with group means that fall above the one-sided CI are
shaded.
13 This method was implemented subsequent to our initial gradient analysis to address the question of
categorical stop/continuant production. Future analysis will entail categorization of the full data set as stops
or continuants according to presence/absence of burst separate analyses of tokens with and without bursts.
14 We assume independence of the 142 responses, as a check of the interclass correlation indicated a
correlation of .104 among responses from the same individual.
Table 5. Intensity difference in dB (estimated marginal means and 95% two-sided CIs) according to group
and prosodic level. The 97.5% one-sided CI for the lower limit of the sample of segments with a release burst
is included as a reference.
CI
Beginner
Intermediate
Advanced
Control
Syllable
[12.35]
12.26
[10.04-15.16]
11.74
[9.35-14.11]
10.35
[8.37-12.34]
5.80
[2.16-9.44]
PW-min
[14.81]
16.33
[13.82-18.83]
14.45
[12.09-16.76]
11.83
[9.88-13.78]
6.43
[2.84-10.02]
PW-max
[18.02]
18.86
[16.32-21.41]
16.22
[13.85-18.58]
11.38
[9.40-13.36]
7.36
[3.71-11.00]
Based on the categorization of means as stops or continuants, the beginner group produces
continuants at the level of the syllable and stops at the PW-min and PW-max. Within-group pairwise
comparisons (Table 6) indicate that the degree of occlusion increases significantly with each prosodic level,
although effect sizes are very small.15 The intermediate group produces continuants at all three levels.
Segments are significantly less occluded at the level of the syllable than at the level of the PW-min and PW-
max, again with very small effect sizes. There is no significant difference between the PW-min and PW-max,
although a TOST did not yield a positive (similar) result. The advanced group also produces continuants at all
15 When evaluating effect sizes, we follow Plonsky and Oswald (2014), who propose based on meta-analysis
that small, medium, and large effects in second language acquisition research correspond respectively to
absolute values of .40, .70, and 1.00 for between-groups analyses and .60, 1.00, and 1.40 for within-groups
analyses.
three prosodic levels. Like the control group, none of the within-group pairwise comparisons across prosodic
levels are statistically significant. While only the PW-min/PW-max comparison returns a positive TOST for
the advanced group and control group, all within-group effect sizes for the two groups fall well below .70 and
the CIs overlap substantially in each case. That said, it remains to be seen whether the advanced group’s
productions are statistically different from those of the control group and how the other groups compare with
the control group as well.
Table 6. Within-groups pairwise comparison p values from Bonferroni post hoc tests and effect sizes
(Hedges’ g, in parentheses) according to prosodic level. TOST results are included when Bonferroni test
yields p > .05.
Beginner
Intermediate
Advanced
Control
TOST
TOST
TOST
Syllable
PW-min
.001* (.49)
.017* (.43)
.166 (.28)
p > .05
1.000 (.22)
p > .05
Syllable
PW-max
.045* (.73)
<.001* (.63)
.573 (.22)
p > .05
.944 (.30)
p > .05
PW-min
PW-max
<.001* (.25)
.179 (.22)
p > .05
1.000 (.04)
p < .05
1.000 (.16)
p < .05
*p < .05
Research question 2
To answer our question of whether learners’ degree of lenition is similar to that of the native controls'
approximants, we evaluate between-groups pairwise comparisons and effect sizes at each prosodic level.
Specifically, we are interested in the comparisons between the control group and the learner groups (Table 7).
Table 7. Between-groups pairwise comparison p values and effect size values according to prosodic level.
TOST results are included when Bonferroni results are p > .05.
Beginner
Intermediate
Advanced
p
g
p
g
p
g
TOST
Control
Syllable
.013*
1.05
.035*
.95
.146
.75
< .05*
PW-min
.000*
1.12
.001*
1.10
.044*
.77
PW-max
.000*
1.05
.000*
.93
.278
.46
> .05
*p < .05
Recall that the beginners produce continuants only at the onset of the syllable. Their mean intensity difference
is significantly larger than that of the control group at all levels, and the effect sizes are all around 1,
indicative of a large effect size according to Plonsky and Oswald (2014). The outcome of the comparison
between the control group and intermediate group is similar; again, there are significant differences between
groups at each of the three levels with large effect sizes. Therefore, although the intermediate group produces
continuants at the three levels, the degree of lenition is significantly less than in the control group’s
productions. The advanced learners do not differ significantly from the control group at the syllable onset, and
a positive TOST result indicates that the degree of lenition at this prosodic level is similar. The Bonferroni
tests used to compare the advanced and control groups at the PW-min and PW-max yield a null result.
Although the TOSTs are negative and therefore indicate an inconclusive outcome, the effect size values
suggest very low practical significance in both cases. Although the beginner and intermediate learners’
intensity difference values are not native-like at the prosodic levels in which they produce continuants (i.e.,
the syllable onset for both beginners and intermediates; the PW-min and PW-max onset for intermediates), we
were interested in whether learners’ intensity difference values decrease as proficiency increases. At the onset
of the syllable, a pairwise comparison of the beginner and intermediate learners yields a non-significant result
(p =1.000) and a positive TOST (p < .05) with an effect size of .13. Comparing the intermediate and advanced
learners at the onset of the PW-min, there was not a significant difference (p =.396) and the effect size was
.35. However, the TOST was negative (p > .05) and the intermediate mean (14.45 dB) falls above the upper
limit of the advanced group’s CI ([9.88-13.78]). At the onset of the PW-max, there was a significant
difference (p =.006) between the intermediate and advanced means.
Discussion
In this section, we begin with a summary of the results, and then address these results in light of our research
questions, followed by a consideration of future research.
Summary
Our analysis of acoustic data suggests that all of the groups produce continuants at the prosodic level of the
syllable onset (word-medial position). The beginner and intermediate groups’ intensity difference values are
similar to each other (TOST p < .05) and lower than that of the control group, while the advanced group and
control group produce continuants with similar intensity difference values (TOST p < .05). At the onset of
PW-min (at the boundary of a lexical word and its clitic), the acoustic data suggest that beginners produce
stops while the other learner groups and controls produce continuants. The intermediate and advanced groups
produce continuant segments with a lower degree of lenition than the control group, and there is an increase in
lenition between intermediate and advanced proficiencies. At the onset of PW-max (at the boundary of two
lexical words), the intermediate, advanced, and control groups produce continuants, and the advanced and
control groups produce segments with a higher degree of lenition than the intermediate group. While the
results of the statistical comparison between the advanced and control groups were inconclusive at this
prosodic level (i.e., neither different nor similar), practical significance was very low.
Discussion of research question 1
We used cross-sectional data to determine a) whether late learners of L2 Spanish can acquire spirantization at
all relevant levels of the hierarchy and b) whether acquisition occurs in stages according to the prosodic
hierarchy. Recall that the intermediate and advanced learners’ intensity difference values at the prosodic level
of the PW-max (and the levels below it) fall below the one-sided confidence interval established for a stop,
which leads us to posit that these learners have acquired spirantization at the highest relevant level of the
prosodic hierarchy. This finding patterns in part with what is reported in Face and Menke (2009) since the
learners produce approximants in word-initial position. However, unlike in Face and Menke (2009), the
advanced learners do not appear to produce more continuants in word-medial position than across word
boundaries and therefore pattern with the control group. Since there is no evidence that neither the advanced
nor the control group differentiate among prosodic levels, data from these two groups do not warrant further
consideration of the acquisition of postvocalic spirantization as a developmental process that spans multiple
stages.
The beginner and intermediate data contain evidence suggestive of gradual categorical acquisition
moderated by prosodic structure. Our results partially support Zampini’s (1997, 1998) original hypothesis
that postvocalic stop lenition is acquired according to the prosodic hierarchy; spirantization is acquired first at
the most restrictive domain (i.e., the onset of the syllable), as predicted by the Subset Principle. However,
there is no evidence that continuants are produced first at the onset of the PW-min and subsequently at the
onset of the PW-max. Instead, the data from the intermediate and advanced learners only allow us to posit that
the prosodic levels that moderate the acquisition of spirantization include a) the syllable and b) the prosodic
word (without the distinction of minimal and maximal projections). While our data are limited to evidence of
two domains that are operative in this process of acquisition, these data add to the evidence for the operation
of phonological hierarchies in L2 acquisition that we reference in ‘Introduction’. We now propose a revised
constraint ranking that captures the stages of acquisition of this phenomenon and accounts for residual
variation.
Assuming that learners start with an L1 English ranking in which the stricture faithfulness constraint
IDENT-IO (CONT) outranks *VD STOP/V_, evidence from the beginner and intermediate learners indicates that
L2 Spanish learners gradually acquire spirantization according to prosodic level. The acoustic data are in line
with our hypothesis that there are separate stricture faithfulness constraints, specifically for the onset of the
syllable and prosodic word. While it is possible that there are separate PW-min and PW-max constraints that
are simultaneously demoted, there is no evidence for this and we therefore assume here that there is a single
PW constraint. Therefore, the two stricture faithfulness constraints in our analysis are IDENT-IO ONSET
(CONT) SYLL and IDENT-IO ONSET (CONT) PW. As learners extend the domain of spirantization, they gradually
demote IDENT-IO (CONT) ONSET SYLL, followed by IDENT-IO (CONT) ONSET PW. Once they demote IDENT-
IO (CONT) ONSET PW, *VOICED STOP/V_ dominates all of the identity constraints and the target ranking is
acquired:
(4) *VOICED STOP /V_>> IDENT-IO (CONT) ONSET PW >> IDENT-IO (CONT) ONSET SYLL >> IDENT-IO (CONT)
>> *APPROXIMANT /V_
The tableaux in Table 7 illustrate the beginners’ ranking. Recall that IDENT-IO (CONT) is the first constraint to
be demoted because it is violated every time a continuant appears in the evidence, independent of prosodic
domain. Since the beginners reliably produce continuants in word-medial position, we propose that IDENT-IO
(CONT) ONSET SYLL has been demoted below *VOICED STOP /V_ (Table 7). However, since IDENT-IO (CONT)
ONSET PW still dominates *VOICED STOP /V_, a stop is predicted to surface at the onset of the PW whether it
shares a boundary with a function word or lexical word.
Table 7. Spanish postvocalic underlying stops – Beginners.
prosodic level
and input
output candidates
IDENT-IO
(CONT)
ONSET
PW
*VOICED
STOP/ V_
IDENT-IO
(CONT)
ONSET
SYLL
IDENT
(CONT)
*APPROXI
MANT/V_
a. syllable
/agua/
! [ˈa.ɣu
̯a]
*
*
*
[ˈa.gu
̯a]
*!
b. PW
/buena gafa/
[ˈbu
̯e.na.ˈɣa.fa]
*!
*
*
*
! [ˈbu
̯e.na.ˈga.fa]
*
As an anonymous reviewer points out, the existence of approximants in English in word-medial position (see
footnote 7) may facilitate the task of the beginners, i.e, their continuant production could be a result of
transfer and not acquisition. This would align with findings from studies such as Major’s (1986) investigation
of L2 Spanish rhotics. He found that learners produced Spanish [ɾ] in word-medial position with a higher rate
of accuracy than in word-final position, and posited that this difference was due to the presence of [ɾ] in word-
medial position in English. This is an empirical question which could be answered by comparing these
beginners’ productions with those of learners at the very initial stages of L2 Spanish acquisition.
By examining the intermediate and advanced learners’ productions, we can address the other part of
this first research question, which is whether learners can acquire a native-like categorical representation.
Since the intermediate and advanced groups produce continuants across the prosodic levels we tested, we
posit that a native-like representation has been acquired in which *VOICED STOP /V_ dominates the
prosodically conditioned identity constraints. This means that a continuant will likely surface in postvocalic
position at the onset of the syllable and PW (Table 8).
Table 8. Spanish postvocalic underlying stops – Intermediate/Advanced/Control.
prosodic level
and input
output candidates
*VOICED
STOP/ V_
IDENT-IO
ONSET
PW
IDENT-IO
ONSET
SYLL
IDENT
(CONT)
*APPROXI
MANT /V_
a. syllable
/agua/
! [ˈa.ɣu
̯a]
*
*
*
[ˈa.gu
̯a]
*!
b. PW
/buena gafa/
! [ˈbu
̯e.na.ˈɣa.fa]
*
*
*
*
[ˈbu
̯e.na.ˈga.fa]
*!
Anonymous reviewers bring up the question of whether acquisition is mediated by prosodic structure
or by the frequency with which /b d g/ appear in a leniting context as a consequence of prosodic structure.
They note that a postvocalic underlying stop in word-medial position will always surface as a continuant, and
therefore continuants will be most frequent in the input in this context. At the PW-min, /b d g/ is likely to
frequently be in a lenition context since a word will be preceded by a clitic, whereby there will be more
variability in terms of the preceding segment at the PW-max. If segments at the onset of the PW-min appear
more frequently in a leniting context than those at the onset of the PW-max (and less frequently than those at
the onset of the syllable), then we would expect to find a difference in the learner data when comparing the
two higher prosodic levels, which we did not. However, we agree that if the difference in frequency of
segments in a leniting context is that much greater in word-medial position than in word-initial position, it is
possible that the observed reranking occurs independently of the inherent ranking of the positional identity
constraints. With that said, we cannot ignore the independent evidence of L1 and L2 acquisition according to
prosodic structure that has been reported for other types of segments that may not have a similar confound of
frequency and structure (see ‘Introduction’). This is an empirical question that could be pursued going
forward by determining the frequency of leniting contexts in each position and running a simulation of the
GLA with this frequency information.
Discussion of research question 2
We have presented evidence that the L1 English/L2 Spanish learners in the present study acquire postvocalic
spirantization in stages according to prosodic structure, and that a native-like constraint ranking can be
acquired. That said, a target-like constraint ranking does not guarantee that the continuants that learners
produce will be native-like in their degree of lenition. Within this study, we consider two pieces of the
acquisition puzzle, the first being the acquisition of categorical representation and the second being the
development of the acoustic representation of the target approximant category.
To address native-like acoustic target development, we compared the learners with the control group.
While the beginners produce continuants at the onset of the syllable, the control group’s segments are
significantly more lenited than the beginners’, which indicates that the group has not fully developed the
acoustic representation of the target. Comparing the control group with the intermediate learners, who
produce continuants at the three levels16 tested, there are significant differences at all three levels. Thus, we
can conclude that these learners’ continuant productions are not target-like. While the advanced and control
groups produce continuants with a similar degree of lenition at the onset of the syllable, their comparison at
16 While the categorical data do not show evidence of separate stages of acquisition at the PW-min and PW-
max, we did not collapse the PW-min and PW-max data in order to determine whether there were gradient
differences at these two prosodic levels.
the two higher levels is not as clear, although individual data inform the question of whether at least a subset
of the learners lenite postvocalic stops to the same degree as the native control group. At the PW-min, there is
a significant difference in the degree of lenition with a medium effect size. However, seven out of 20
advanced learners fall below the upper limit of the control group’s 95% CI (10.02 dB), that is, they produce
segments that are lenited to the same degree as the controls. At the PW-max, 10 learners fall below the upper
limit of the control group CI (11.00 dB) and 10 fall above. A total of six of the learners produce target-like
segments across the three levels. Thus, it would appear that several individuals have converged on the
acoustic target as well as the target constraint ranking across the prosodic levels.
The variability in native-like attainment among the advanced learners leads us to a consideration of
the source of difficulty in producing a native-like continuant. One possibility is that some have not been
accurately categorized as advanced learners. We recognize that the proficiency measurement used to
categorize the participants might not adequately reflect proficiency as it relates to phonological production,
and that the oral proficiency of some of the learners might be more advanced than the others. Since carrying
out this project, we have begun implementing foreign accent ratings as a global oral proficiency measurement
(e.g., Authors, 2016), the use of which could result in placement of some of these advanced participants in a
lower proficiency group. Another possibility is that acquisition has fossilized as a result of one or more
factors that have been reported to impede native-like L2 production (see e.g., Broselow and Kang, 2013 for a
review of these factors). While additional factors may be at play, we posit that speech rate might be a locus of
the difference in degree of lenition between these advanced learners and controls. Degree of lenition increases
as a function of speech rate; given that L2 speech is typically slower than native speech (e.g., Derwing and
Munro, 1997), it is possible that higher rates of intensity difference could be due to slower speech rate.
Learner speech rate might also be reduced in this case because of the task type; spontaneous speech samples
could potentially yield lower intensity difference.
In addition to ultimate attainment, we were interested in whether there was evidence of target
development according to word position. There are several indications of this type of development: First, the
beginner and intermediate learners produce segments that are less lenited than the controls’ segments while
the advanced group produces segments that are equally lenited, suggesting development at the most restrictive
setting for postvocalic stop lenition. Second, the intermediate learners produce more lenited segments at the
level of the syllable than at the PW-min and PW-max, which again points to differential development of the
target at the lowest prosodic level. Third, the intermediate learners produce less lenited segments than the
advanced group at the PW-min and PW-max. This last finding highlights the benefit of approaching the
acquisition of L2 spirantization from both categorical and gradient perspectives. If we were to limit our
investigation to categorical acquisition, we might conclude that the data indicate that intermediate learners
had fully acquired L2 spirantization. However, the difference in degree of lenition between the intermediate
and advanced/control groups suggest that there are two parts to this acquisition story.
Comparing these results with those of the two previous studies that have examined degree of lenition
in L2 Spanish spirantization, there are commonalities in terms of both acquisition and ultimate attainment.
Data from Shea and Curtin (2011) indicate that the high intermediate learners they tested produced segments
that were (descriptively) more lenited than the segments that low intermediate learners produced in both
word-medial and word-initial positions, although not native-like. Our data show that there is further reduction
in intensity difference between intermediate and advanced proficiencies, and that it is possible for learners to
eventually develop a native-like acoustic target. These findings mirror the smaller-scale findings in Rogers
and Alvord (2014) in terms of a) development of the acoustic target across proficiency levels and b) ultimate
attainment, and do so with a larger sample size (n=20) and with inferential statistical support.
Residual variation
A frequent question that comes up in the study of second language acquisition is why an adult L2 learner can
so often be identified as a non-native speaker even at a very advanced level of proficiency, and we believe it
is possible to use L2 spirantization as a case study to consider this question. As reported, our data indicate
that there are several advanced learners that have acquired the categorical representation but not developed
the relevant acoustic target. However, what about the advanced learners that have successfully developed the
target? Take advanced participant 1003, for example, whose lenition is native-like across the three prosodic
levels. In Authors (2016), a 15-second speech sample from this participant was submitted to a foreign accent
rating whereby native Spanish speakers assigned a score between 1 (very strong foreign accent) and 7
(without a doubt is a native speaker). The score given to the participant was 5.50 out of 7, which indicates that
his global oral proficiency falls short of that of a native speaker. There are undoubtedly numerous variables
that raters attend to when making their judgments (e.g., intonational patterns, other segmental phenomena,
verbal fluency, etc.), but we would like to propose that one of the variables that might be responsible for a
non-native accent is residual optionality, which can be explained by overlapping constraints. If we examine
the individual data points of the learners who appear to have converged on the target ranking as well as the
acoustic target, five of the learners produce at least one stop at each prosodic level. Even participant 1053,
who exclusively produces continuants at the two lower levels, produces two stops at the PW-max. These
findings align with those of Face and Menke (2009), who found that the PhD students they tested continued to
produce stops upwards of 20% of the time. Recall that within a stochastic model of OT, variable outputs can
persist when two constraints’ ranking values are close to one another. It is therefore possible that although the
probability distributions have been adjusted such that these learners have acquired the native-like ranking
shown in in Table 9, the ranges that *VOICED STOP /V_ and the relevant IDENT constraint(s) occupy are
overlapping. As long as the selection points do not occur where there is overlap (Figure 9), a continuant will
be produced. However, if the selection points for the relevant constraints occur in the space in which the
constraints overlap (Figure 10), a stop will be produced.
Figure 9. Selection points for two constraints that are made outside of the range in which there is overlap.
Figure 10. Selection points for two constraints that are made in the range in which there is overlap.
Considering that there are 18 tokens for each prosodic level, if a learner produces one stop, we can assume
that the constraint overlap is minimal since the probability that a continuant will surface is very high. On the
other hand, consider advanced participant 1014, who produces 10/18 stops across the boundary of two lexical
words; the constraint overlap is such that there is a higher probability that the selection points will result in an
IDENT-IO (CONT) ONSET PW >> *VOICED STOP /V_ ranking than a *VOICED STOP /V_ >> IDENT-IO (CONT)
ONSET PW ranking.
Future directions
The findings in this study lead us the considerations for future investigations related to L2 acquisition and L1
acquisition, respectively.
Although our data do not support Zampini’s (1997, 1998) finding of a separate stage of acquisition at
the onset of the PW-min, it is possible that our cross-sectional investigation does not capture the full path of
acquisition. After all, this is the only study to use acoustic data to address segment position beyond a word-
medial/word-initial distinction and thus additional research would be welcome. In particular, a longitudinal
examination of L2 spirantization would afford observation of the acquisition process across the same
individuals without the obstacle of classifying learners according to proficiency. Such a study would allow us
to a) determine at the initial stages whether learners transfer spirantization in word-medial position from
English and b) identify any intermediate stages of acquisition.
Our second consideration is that of L1 acquisition of postvocalic stop lenition. Adult L2 data can
inform phonological theory in a way in which adult monolingual data cannot, because we are able to observe
development, including the effects of the prosodically conditioned constraints that we have proposed here.
Indeed, we provide further experimental evidence in support of the Subset Principle (e.g., Wexler and
Manzini, 1987), suggesting that adult L2 learners start with the most restrictive grammar and expand their
grammar according to positive evidence in the input. That is, postvocalic lenition is first evident in word-
medial position (at the onset of the syllable), and as proficiency increases, learners lenite across word
boundaries at the onset of the prosodic word. However, we can only make claims for L2 learners, as we
cannot be sure that the L2 developmental path is the same as the L1 developmental path. As noted in
Schuttenhelm (2013), although the acquisition process has led to a native-like target (at least for some), the
underlying representation (in this case, the constraint ranking) for an L2 learner might be different from that
of a native speaker of Spanish. The only way to verify this is of course via investigation of children’s
productions. While several studies have investigated children’s acquisition of spirantization (e.g., Barlow,
2003; Fabiano-Smith et al., 2014; Murillo, 1978), they have not explicitly addressed position of the segment.
Conclusion
The primary aim of this study has been to test the hypothesis that adult L2 acquisition of postvocalic
continuants is moderated by the prosodic hierarchy. Acoustic analysis of cross-sectional data indicates that
prosodic structure indeed governs the path of both categorical and gradient acquisition, with learners first
producing postvocalic underlying stops as approximants in word-medial position and then in word-initial
position. Within a stochastic Optimality Theoretic framework, we proposed an analysis of gradual frequency-
driven acquisition according to prosodic structure and provide acoustic data that supports that L2 Spanish
learners acquire postvocalic spirantization according to prosodic structure. We also determined that while it
appears that learners can acquire a native-like representation (i.e., constraint ranking), the development of
native-like acoustic representations is harder to come by. Importantly, the analysis accounts for data which
support advanced learners’ attainment of a native-like target, while also explaining why we observe non-
native outputs even at near-native proficiency.
Appendix 1. Recitation task items according to prosodic level
Syllable
arriba
hábito
ubica
dibujo
abusar
duda
rodilla
adivinar
maduro
madurez
regar
hígado
seguir
águila
aguja
agujero
PW-min
mi barrio
la ballena
la biblia
mi bigote
lo busqué
su burro
su dado
lo dañaba
la dicha
su dinero
la duda
de dureza
mi gato
su ganancia
una guía
la guitarra
su gusto
tu gusano
PW-max
es un curso básico
me quiero bajar
es algo bíblico
es mi primera bicicleta
es un estilo buque
quiere buscar a ella
necesito darme cuenta
no puedo dañarme
es una vida digna
ella quiere disfrazarse
es una chica dudosa
no tengo ganas
le puede garantizar
odio guiar a ellos
tiene gusto de fútbol
References
Alvord SM and Christiansen DE (2012) Factors influencing the acquisition of Spanish voiced stop spirantization
during an extended stay abroad. Studies in Hispanic and Lusophone Linguistics 5: 239-276.
Archibald J (2009) Phonological feature re-assembly and the importance of phonetic cues. Second Language Research
25: 231-233.
Barlow J (2003) The stop-spirant alternation in Spanish: Converging evidence for a fortition acount. Southwest Journal
of Linguistics 22: 51-86.
Best C (1995) A direct realist view of cross-language speech perception. In: Strange W (ed) Speech perception and
linguistic experience: issues in cross-language research. Baltimore: York Press, 171-204.
Best C and Tyler M (2007) Nonnative and second- language speech perception: Commonalities and
complementarities. In: MJ Munro and Bohn OS (eds) Second Language Speech Learning: The Role of
Language Experience in Speech Perception and Production. Amsterdam: John Benjamins, 13-34.
Birdsong D (2003) Authenticité de la pronciation en français L2 chez de apprenants tardifs anglophones: analyses
segmentales et globales. Acquisition et Instruction en Language Étrangère 18: 17-36.
Birdsong D and Molis M (2001) On the evidence for maturational constraints in second language acquisition. Journal
of Memory and Language 44: 235-249.
Boersma P (1997) How we learn variation, optionality, and probability. Proceedings of the Institute of Phonetic
Sciences of the University of Amsterdam 21: 43-58.
Boersma P and Hayes B (2001) Empirical tests of the Gradual Learning Algorithm. Linguistic Inquiry 32: 45-86.
Boersma P and Weenink D (2014) Praat: Doing Phonetics by Computer. Available at:
http:www.fon.hum.uva.nl/praat/.
Bonet E and Mascaró J 1997 On the representation of contrasting rhotics. In: Martínez-Gil and Morales-Front
(eds) Issues in the Phonology and Morphology of the Major Iberian Languages. Washington, D.C.:
Georgetown University Press, 103-126.
Bongaerts T (1999) Ultimate attainment in L2 pronunciation: The case of very advanced late learners. In: Birdsong D
(ed) Second Language Acquisition and the Critical Period Hypothesis. Mahwah, NJ: Lawrence Erlbaum
Associates, 133-160.
Bouavichith D and Davidson L (2013) Segmental and prosodic effects on intervocalic voiced stop reduction
in connected speech. Phonetica 70: 182-206.
Branigan G (1976) Syllabic structure and the acquisition of consonants: The great conspiracy in word
formation. Journal of Psycholinguistic Research 5: 117-133.
Broselow E, Chen SI and Wang C (1998) The emergence of the unmarked in second language phonology. Studies in
Second Language Acquisition 20: 261-280.
Broselow E and Kang Y (2013) Second language phonology and speech. In: Herschensohn J and Young-Scholten M
(eds) The Cambridge Handbook of Second Language Acquisition. Cambridge: Cambridge University Press,
529-554.
Brown G (1990) Listening to spoken English, New York: Longman.
Brownman CP and Goldstein L (1992) Articulatory phonology: An overview. Phonetica 49: 155-180.
Cardoso W (2007) The variable development of English word-final stops by Brazilian Portuguese speakers: A
stochastic optimality theoretic account. Language Variation and Change 19: 219-248.
Carrasco P, Hualde JI and Simonet M (2012) Dialectal differences in Spanish voiced obstruent allophony: Costa Rican
versus Iberian Spanish. Phonetica 69: 149-179.
Carroll S (2001). Input and evidence: The raw material of second language acquisition. Amsterdam: John Benjamins.
Celce-Murcia M, Brinton DM and Goodwin JM (1996) Teaching pronunciation: a reference for teachers of English as
a second language. Cambridge: Cambridge University Press.
Colantoni L and Marinescu I (2010) The scope of stop weakening in Argentine Spanish. In: Ortega Llebaria (ed)
Selected Proceedings of the 4th Conference on Laboratory Approaches to Spanish Phonology. Somerville,
MA: Cascadilla Proceedings Project, 100-114.
Cole J, Hualde JI and Iskarous K (1999) Effects of prosodic and segmental context on /g/-lenition in Spanish. In:
Fujimora O, Joseph BD and Palek B (eds) Language Processing. Prague: The Karolinum Press, 575-589.
Colina S (2013) On onset clusters in Spanish: Voiced obstruent underspecification and /f/. Paper presented at the 17th
Hispanic Linguistics Symposium (Ottawa, Canada).
Davidson L (1997) An Optimality Theoretic approach to second language acquisition. PhD Thesis, Brown University,
USA.
Davies M (2006) A Frequency Dictionary of Spanish: Core Vocabulary for Learners. New York: Routledge.
De Lacy P (2001) Markedness in prominent positions. In: Matushansky O (ed) Proceedings of the 1st HUMIT Student
Conference in Language Research (HUMIT 2000). Cambridge, MA: MIT Press, 53-66.
Dell FC (1981) On the learnability of optional phonological rules. Linguistic Inquiry 12: 21-37.
Derwing TM and Munro MJ (1997) Accent, intelligibility, and comprehensibility: Evidence from four L1s. Studies in
Second Language Acquisition 19: 1-16.
Diaz-Campos M (2004) Context of learning in the acquisition of Spanish second language phonology. Studies in
Second Language Acquisition 26: 249-273.
Diaz-Campos M (2006) The effect of style in second language phonology: An analysis of segmental acquisition in
study abroad and regular classroom students. In: Klee C and Face T (eds) Selected Proceedings of the 7th
Conference on the Acquisition of Spanish and Acquisition of Spanish and Portuguese as First and Second
languages. Somerville, MA: Cascadilla Proceedings Project, 26-39.
Díaz-Campos M and Gradoville M (2011) An analysis of frequency as a factor contributing to the diffusion of variable
phenomena. Evidence from Spanish data. In Ortiz López L (ed) Selected Proceedings of the 13th Hispanic
Linguistics Symposium. Somerville, MA: Cascadilla Proceedings Project, 224-238.
Eddington D (2011) What are the contextual phonetic variants of in colloquial Spanish? Probus 23: 1-19.
Elliot AR (1997) On the teaching and acquisition of pronunciation within a communicative approach. Hispania 80: 95-
108.
Fabiano-Smith L, Oglivie T, Maiefski O and Schertz J (2015) Acquisition of the stop-spirant alternation in bilingual
Mexican Spanish-English speaking children: Theoretical and clinical implications. Clinical Linguistics and
Phonetics 29: 1-26.
Face T and Menke MR (2009) Acquisition of the Spanish voiced spirants by second language learners. In: Collentine
J, García M, Lafford B, and Marín FM (eds) Selected Proceedings of the 11th Hispanic Linguistics
Symposium. Somerville, MA: Cascadilla Proceedings Project, 39-52.
Flege J (1995) Second language speech learning: Theory findings and problems. In: Strange W (ed) Speech Perception
and Linguistic Experience: Issues in Cross-Language Research. Baltimore: York Press, 233-277.
Flege JE (1991) Perception and production : The relevance of phonetic input to L2 phonological learning.
Crosscurrents in second language and acquisition and lingusitic theories. 2: 249-289.
Giancaspro D (2013) L2 learners' and heritage speakers' judgments of code switching at the auxiliary-VP boundary. In
Cabrelli Amaro J, Lord G, de Prada Pérez A and Aaron JA (eds) Selected Proceedings of the 14th Hispanic
Linguistics Symposium. Somerville, MA: Cascadilla Proceedings Project, 56-69.
Gimson AC (1989) An Introduction to the Pronunciation of English. London: Edward Arnold.
Goldsmith J (1981) Phonology as an intelligent system. In Napoli DJ and Kegl JA (eds) Bridges between Psychology
and Linguistics: A Swarthmore Festschrift for Lila Gleitman. Hillsdale, NJ: Lawrence Erlbaum, 247-267.
Gong J (2013) Computational modelling of sound perception in a second language. PhD Thesis, University of the
Basque Country, Spain.
Gonzalez-Bueno M (1995) Adquisición de los alófonos fricativos de la oclusivas sonoras españolas por aprendices de
español como segunda lengua. Estudios de lingustica aplicada 13: 64-79.
Granena G and Long M (2013) Age of onset, length of residence language aptitude, and ultimate L2 attainment in
three linguistic domains. Second Language Research 29: 311-343.
Grosjean F (1998) Studying bilinguals: Methodological and conceptual issues. Bilingualism: Language and Cognition
1: 131-149.
Hall TA (1997) The Phonology of Coronals. Amsterdam: John Benjamins.
Hancin-Bhatt B (2000) Optimality in second language phonology: Codas in Thai ESL. Second Language Research 16:
201-232.
Hancin-Bhatt B (2008) Second language phonology in optimality theory. In Hansen Edwards J and Zampini M (eds)
Phonology and Second Language Acquisition. Amsterdam: John Benjamins, 117-146.
Hancin-Bhatt B and Bhatt RM (1997) Optimal L2 syllables. Studies in Second Language Acquisition 19: 331-378.
Harris J (1993) Integrity of Prosodic Constituents and the Domain of Syllabification Rules in Spanish and Catalan.
Cambridge, MA: MIT Press.
Harris JW (1969) Sound change in Spanish and the theory of markedness. Language 45: 538-552.
Harris JW (1984) La espirantización en castellano y la representación fonológica autosegmental. In: Estudis
gramaticals 1. Barcelona: UAB Press, 149-167.
Harris JW and Lindsey G (1995) The elements of phonological representation. In: Durand J and Katamba F (eds)
Frontiers of Phonology: Atoms, Structures, Derivations. Harlow, Essex: Longman, 34-79.
Hayes BP (1989) The prosodic hierarchy in meter. In: Kiparsky P and Youmans G (eds) Phonetics and
Phonology, Volume 1: Rhythm and Meter. San Diego: Academic Press, Inc., 201-260.
Hayes BP (1999) Phonetically driven phonology. In: Darnell M, Moravcsik EA, Newmeyer FJ and Wheatley K (eds)
Functionalism and Formalism in Linguistics. Amsterdam: John Benjamins, 243-285.
Hayes-Harb R (2005) Optimal L2 speech perception: Native speakers of English and Japanese consonant length
contrasts. Journal of Language and Linguistics 4: 1-29.
Holt E (2002) The articulator group and liquid geometry: Implications for Spanish phonology present and past. In:
Wiltshire C and Camps J (eds) Romance Phonology and Variation. Amsterdam: Benjamins, 85-99.
Hopp H and Schmid MS (2013) Perceived foreign accent in first language attrition and second language acquisition:
The impact of age acquisiton and bilingualism. Applied Psycholinguistics 34: 361-394.
Hualde JI (1988) A lexical phonology of Basque. PhD Thesis, University of Southern California, USA.
Hualde JI, Shosted R and Scarpace D (2011) Acoustics and articulation of Spanish /d/ spirantization. In: Lee W-S and
Zee E (eds) Proceedings of the International Congress of Phonetic Sciences XVII. Hong Kong: City
University of Hong Kong, 906-909.
Hualde JI, Simonet M and Nadeu M (2011) Consonant lenition and phonological recategorization. Laboratory
Phonology 2: 301-329.
Hyltenstam K and Abrahamsson N (2000) Who can become native-like in a second language? All, some or none? On
the maturational constraints controversy in second language acquisition. Studia Linguistica 54: 150-166.
Hyltenstam K and Abrahamsson N (2003) Maturational constraints on SLA. In: Doughty CJ and Long MH (eds) The
handbook of second language acquisition. Oxford: Blackwell, 539-588.
Hyltenstam N and Abrahamsson K (2009) Age of onset and nativelikeness in a second language: Listener perception
versus linguistic scrutiny. Language Learning 59: 249-306.
Ito J and Mester A (2007) Prosodic adjunction in Japenese compounds. In: Miyamoto Y and Ochi M (eds) Formal
Approaches to Japanese Linguistics 4. Cambridge, MA: MIT Press, 97-111.
Ito J, Mester A and Padgett J (1995) Licensing and underspecification in Optimality Theory. Linguistic Inquiry 26:
571-613.
Jakobson R (1941) Kindersprache, Aphasie und allgemeine Lautgesetza, Uppsala: Almqvist & Wiksell.
Juzek T and Kizach J (2016). The TOST as a method of similarity testing in linguistics. Unpublished manuscript.
Kaplan A (2010) Phonology shaped by phonetics: The case of intervocalic lenition. PhD Thesis, University of
California-Santa Cruz, USA.
Kaplan A (2010) Phonology Shaped by Phonetics: The Case of Intervocalic Lenition Ann Arbor, MI.
Kaplan A (2010) Phonology shaped by phonetics: The case of intervocalic lenition.: University of California-Santa
Cruz.
Kenstowicz MJ (1994) Phonology in Generative Grammar. Cambridge, MA: Blackwell.
Kilpatrick C (2009) The acquisition of ungrammaticality: Learning a subset in L2 phonotactics. PhD Thesis,
University of California – San Diego, USA.
Kingston J (2008) Lenition. In Colantoni L and Steele J (eds) Selected proceedings of the 3rd conference on laboratory
approaches to Spanish phonology. Somerville, MA: Cascadilla Proceedings Project, 1-31.
Kiparsky P and Menn L (1977) On the acquisition of phonology. In MacNamara J (ed) Language and thought. New
York: Academic Press, 47-48.
Kirchner R (1998) An effort-based approach to consonant lenition. PhD Thesis, University of California Los Angeles,
USA.
Kirchner R (2004) Consonant lenition. In: Hayes B and Steriade D (eds) Phonetically Based Phonology. Cambridge:
Cambridge University Press, 313-345.
Krause A (2014) What to do about nonsense words?: The case of Spanish dipthongs. Paper presented at Current
Approaches to Spanish and Portuguese Second Language Phonology (CASPSLaP). Georgetown University.
Lavoie L (1996) Consonant strength: Results of a data base development project. Working Papers of the Cornell
Phonetics Laboratory 11: 269-316.
Lavoie LM (2001) Consonant Strength: Phonological Patterns and Phonetic Manifestations. New York: Routledge.
Lombardi L (2003) Markedness and the typology of epenthetic vowels. Unpublished manuscript, University of
Maryland.
Lombardi L (2003) Second language data and constraints on Manner: Explaining substitutions for the English
interdentals. Second Language Research 19: 225-250.
Long M (1990) Maturational constraints on language development. Studies in Second Language Acquisition 12: 251-
285.
Long M (2005) Problems with supposed counter-evidence to the critical period hypothesis. International Review of
Applied Linguistics in Language Teaching 43: 287-317.
Lord G (2005) How can we teach foreign language pronunciation? On the effects of a Spanish phonetics course.
Hispania 88: 557-567.
Lozano MD (1979) Stop and spirant alternations: Fortition and spirantization processes in Spanish phonology. PhD
Thesis, Indiana University, USA.
Maddieson I (1984) Patterns of Sounds. Cambridge: Cambridge University Press.
Maddieson I (2011) Phonological complexity in linguistic patterning. In: Lee W-S and Zee E (eds) Proceedings of the
17th International Congress of Phonetic Sciences. Hong Kong: City University of Hong Kong, 28-34
Major RC (1986) The ontogeny model: Evidence from L2 acquisition of Spanish r. Language Learning 36: 453-504.
Major RC (2001) Foreign Accent: The Ontogeny and Phylogeny of Second Language Phonology. New York:
Routledge.
Major RC (2004) Gender and stylistic variation in second language phonology. Language Variation and Change 16:
169-188.
Manzini MR and Wexler K (1987) Parameters, binding theory, and learnability. Linguistic Inquiry 18: 413-444.
Marinova-Todd SH (2003) Know your grammar: What the knowledge of syntax and morphology in an L2 reveals
about the critical period for second/foreign language acquisition. In: Garcia-Mayo MP and Garcia-Lecumberri
ML (eds) Age and the Acquisition of English as a Foreign Language: Theoretical Issues and Field Work.
Clevedon: Multilingual Matters, 59-76.
Martohardjono G, Valian V and Klein E (2012) The tense puzzle in second language acquisition: What part
representation? What part performance? Unpublished manuscript, The Graduate Center at CUNY.
Mascaró J (1984) Continuant spreading in Basque, Catalan, and Spanish. In: Aronoff M and Oehrle R (eds) Catalan
working papers in linguistics 1991. Cambridge, MA: MIT Press, 287-298.
Mascaró J (1991) Iberian spirantization and continuant spreading. Catalan Working Papers in Linguistics 1, 167-179.
Montrul S (2008) Incomplete Acquisition in Bilingualism:Re-examining the Age Factor. Amsterdam: John Benjamins.
Murillo C (1978) The integration of linguistic constraints on spirantization in Argentinian Spanish. Master’s
Thesis, San Diego State University, USA.
Nespor M and Vogel I (1986) Prosodic Phonology, Dordrecht: Foris.
Ortega-Llebaria M (2004) Interplay between phonetic and inventory constraints in the degree of spirantization of
voiced stops: Comparing intervocalic /b/ and intervocalic /g. Spanish and English. In: T Face (ed) Laboratory
Approaches to Spanish Phonology. Berlin: Mouton de Gruyter, 237-253.
Ortega-Llebaría M and Prieto P (2011) Acoustic correlates of stress in Central Catalan and Castilian Spanish.
Language and Speech 54: 73-97.
Parker S (2002) Quantifying the sonority hierarchy. PhD Thesis, University of Massachusetts Amherst, USA.
Parker S (2008) Sound level of protusions as physical correlates of sonority. Journal of Phonetics 36: 55-90.
Pierrehumbert J and Beckman M (1988) Japanese Tone Structure. Cambridge, MA: MIT Press.
Piñeros CE (2002) Markedness and laziness in Spanish obstruents. Lingua 112: 379-413.
Piñeros CE (2003) Accounting for the instability of Palenquero voiced stops. Lingua 113: 1185-1222.
Prince A and Smolensky P (1993/2004) Optimality Theory: Constraint Interaction in Generative Grammar. New
Jersey: Wiley-Blackwell.
Radu M (2014) Intervocalic stop lenition in Columbian Spanish. Paper presented at the 18th Hispanic Linguistics
Symposium. West Lafayette, Indiana.
Recasens D (2002) Weakening and strengthening in Romance revisited. Italian Journal of Linguistics 14: 327-374.
Rogers B and Alvord SM (2014) The gradience of spirantization: Factors affecting L2 production of intervocalic
Spanish. Spanish in Context 11: 402-424.
Ronquest R (2014) Lenition of intervocalic /bdg/ in hertiage Spanish: An acoustic analysis of an emerging bilingual
variety in the Southeastern United States. Paper presented at Current Approaches to Spanish & Portuguese
Second Language Phonology. Washington, D.C.
Rothman J (2008) Why all counter-evidence to the Criticial Period Hypothesis in second language acquisition is not
equal or problematic. Language and Linguistics Compass 2: 1063-1088.
Schmidt L (2014) Contextual variation in L2 Spanish: Voicing assimilation in advanced learner speech. Studies in
Hispanic and Lusophone Linguistics 7: 1-35.
Schuirmann D. J. (1981) On hypothesis testing to determine if the mean of a normal distribution is contained in a
known interval. Biometrics 32: 617.
Schuttenhelm L (2013) Perception of the non-native phone [g] in Dutch: where lies the VOT boundary? MA Thesis,
University of Amsterdam, the Netherlands.
Selkirk E (1986) On derived domains in sentence phonology. Phonology 3: 371-405.
Shea CE and Curtin S (2011) Discovering the relationship between context and allophones in a second language.
Studies in Second Language Acquisition 32: 581-606.
Shea CE and Curtin S (2011) Experience, representations and the production of second language allophones. Second
Language Research 27: 229-250.
Shively RL (2008) L2 Acquisition of [β], [ð], and [ɣ] in Spanish: Impact of experience, linguistic environment and
learner variables. Southwest Journal of Linguistics 27: 79-114.
Simonet M, Hualde JI and Nadeu M (2012) Lenition of /d/ in spontaneous Spanish and Catalan. In: Proceedings of the
13th Annual Conference of the International Speech Communication Association (InterSpeech). Red Hook,
NY: Curran Associates, 1414-1417.
Slabakova R and Montrul S (2003) Competence similarities between native and near-native speakers: An investigation
of the preterite/imperfect contrast in Spanish. Studies in Second Language Acquisition 25: 351-398.
Solé MJ (2012) Natural and unnatural patterns of sound change? In: Solé MJ and Recasens D (eds) The Initiation of
Sound Change: Perception, Production, and Social Factors. Amsterdam: John Benjamins, 123-146.
Soler A and Romero J (1999) The role of duration in stop lenition in Spanish. In: Ohala JJ, Hasegawa Y, Ohala M,
Granville D and Bailey AC (eds) Proceedings of the 14th International Congress of Phonetic Sciences.
Berkeley, CA: University of California Press, 483-486.
Tesar B and Smolensky P (1998) Learnability in Optimality-Theoretic grammars. Linguistic Inquiry: 29, 229-268.
Warner N and Tucker BV (2011) Phonetic variability of stops and flaps in spontaneous and careful speech. The
Journal of the Acoustical Society of America 27: 1606-1617.
Westlake, W. J. (1976) Symmetric confidence intervals for bioequivalence trials. Biometrics 32: 741-744.
Wexler K and Manzini MR (1987) Parameters and learnability in binding theory. In: Roeper T and Williams E (eds)
Parameter Setting. Dordrecht: Reidel, 41-76.
White L (2000) Second language acquisition: From initial to final state. In: Archibald J (ed) Second Language
Acquisition and Linguistic Theory. New York: Blackwell, 130-155.
Widdison KA (1997) Physical parameters behind the stop-spirant alternation in Spanish. Southwest Journal of
Linguistics 16: 73-84.
Wode H (1995) Speech perception, language acquisition and linguistics: Some mutual implications. In: Strange W (ed)
Speech Perception and Linguistic Experience: Issues in Cross-Language Research. Timonium, MD: York
Press, 321-347.
Wode H (1997) Perception and production in learning to talk. In: Hannahs SJ and Young-Scholten M (eds) Focus on
Phonological Acquisition. Amsterdam: John Benjamins, 17-46.
Young-Scholten M (1994) On positive evidence and ultimate attainment in L2 phonology. Second Language Research
10: 193-214.
Young-Scholten M and Archibald J (2000) Second language syllable structure. In: Archibald. J (ed) Second Language
Acquisition and Linguistic Theory. Oxford: Wiley-Blackwell, 64-97.
Zampini ML (1994) The role of native language transfer and task formality in the acquisition of Spanish spirantization.
Hispania: 470-481.
Zampini ML (1997) L2 Spanish spirantization, prosodic domains and interlanguage rules. In: Hannahs SJ and Young-
Scholten M (eds) Focus on Phonological Acquisition. Amsterdam: John Benjamins, 209-234.
Zampini ML (1998) L2 Spanish spirantization: A prosodic analysis and pedagogical applications. Hispanic Linguistics
10: 154-188.
