Investigating the relation between L2 pauses,syntactic complexity,and pause location: Longitudinal data from L2-Spanish study-abroad learners

Abstract

Inverse relations, or ‘trade-off effects’, are a common outcome of interlanguage development: a learner may increase performance in one linguistic domain while simultaneously decreasing performance in another. In this study, we investigate the relationships between one aspect of fluency (pause usage) and two aspects of syntactic complexity (utterance length and subordination) in relation to the location of pauses (between-clause or within-clause) in second-language (L2) oral narratives. The longitudinal analysis is based on monologic data produced by 16 English-speaking L2 learners of Spanish who participated in a seven-week study-abroad program in Spain. Overall, the learners decreased their silent-pause rate over the course of the program while concurrently increasing their number of syntactically complex clauses. Notably, the data suggest a systematic trade-off between pausing and complexity: the learners consistently produced more pauses (i.e. decreased fluency performance) during the elocution of the most complex clauses involving clausal subordination (i.e. increased complexity performance) in comparison to utterances lacking such subordination. We contextualize the findings within models of oral production and discuss how this research generates new insight into the processing factors that modulate pause usage in L2 speech.

Keywords

fluency pauses Spanish study-abroad subordination syntactic complexity trade-offs

I Introduction

Disfluent speech is a well-known trait of spoken human language (Bortfeld et al., 2001; Tree, 1995). Although the type and frequency of disfluencies (e.g. pauses, repairs, reformulations) vary across languages (Campione and Véronis, 2002), these speech phenomena cannot be systematically avoided (Goldman-Eisler, 1961, 1968). Pioneering studies on first-language (L1) disfluencies linked their production to psycholinguistic factors such as cognitive load, working memory, and speech planning (Barr, 2001; Beattie, 1979). Regarding second-language (L2) speech, research commonly shows that L2 users produce filled pauses (FPs) and silent pauses (SPs) at greater rates and with longer durations than comparable L1 speakers of a target language (De Jong, 2016; Kahng, 2014). While there may be similarities between L1 and L2 pause production (De Jong et al., 2015), the use of pauses in the L2 is also inextricably linked to cognitive and processing factors such as a learner’s monitoring and formulation of phonological, lexical, and morphosyntactic constructions (Kormos, 2006; Skehan et al., 2016; see Swerts, 1998).

Of interest to the present study, research shows that the syntactic complexity of an utterance can modulate L2 pause usage (Mirdamadi and De Jong, 2015; Towell et al., 1996), with the understanding that more complex syntax involves longer utterances and increased subordination (Norris and Ortega, 2009; Pallotti, 2015). Within the complexity–accuracy–lexis–fluency framework (CALF; see Michel, 2017), studies consider the possibility of trade-off effects between complexity and fluency wherein increased complexity performance brings about diminished fluency performance (Skehan and Foster, 2012; Towell, 2012; Yuan and Ellis, 2003). However, most studies in this area implement cross-sectional designs, leaving open the question of how trade-offs emerge after increased exposure and interaction in the target language (see Ferrari, 2012; Vercellotti, 2017). Although they are undoubtedly more challenging to conduct, longitudinal studies are highly desirable in L2 research, particularly as they offer a more complete view of interlanguage development (Han and Tarone, 2014; Selinker, 1972). The present study confronts this gap of knowledge by implementing a longitudinal design that considers whether trade-off effects emerge between syntactic complexity and pause usage in L2 speech. We examined oral retells from English-speaking L2 learners of Spanish who participated in a seven-week study-abroad program.

II Background

1 Models of speech production

‘Fluent’ speech is generally understood as smooth and effortless message communication that lacks disfluencies such as pauses, repairs, repetitions, lengthenings, and reformulations (Goldman-Eisler, 1961, 1968; Schmidt, 1992). Models of speech production establish a cognitive link between the use of disfluencies and interruptions in the speech-production process. Levelt’s (1983, 1989) ‘blueprint’ for the act of speaking in the monolingual mind, for example, begins at the site of conceptualization, or preverbal message generation (the Conceptualizer), followed by message formulation through grammatical and phonological encoding (the Formulator), and ends at articulation, where the phonetic plan is realized orally (the Articulator). In subsequent publications, De Bot (1992, 2003), Kormos (2006), and Segalowitz (2010, 2016) have adapted Levelt’s model to explain disfluency production in L2 speech. Empirically, cognitive difficulties can be assessed through the measurement of a learner’s utterance-fluency patterns, such as their speed fluency (e.g. rate of speech), breakdown fluency (e.g. pausing rate), or repair fluency (e.g. repetitions, self-corrections, and reformulations; see Skehan, 2003).

2 Pause production in L2 speech

This study presents data on two speech units of breakdown fluency, namely FPs (e.g. uh, um, eh, em) and SPs (periods of silence typically lasting over 250 ms in duration). As synthesized in de Leeuw (2007), researchers generally take one of two approaches when studying pause production: the ‘signal-hypothesis’ approach or the ‘symptom-hypothesis’ approach. Under the signal hypothesis, pauses serve a pragmatic or social function based on the discursive needs of a speaking context, such as floor-holding or turn-taking (Swerts, 1998). Clark and Tree (2002), for instance, explain that speakers may use pauses when they forget a word or name; pauses in such situations are a signal that the interlocutor may complete the utterance by providing the missing information. Alternatively, under the symptom hypothesis, pauses reflect cognitive difficulties experienced by the speaker through speech planning (Barr, 2001; Beattie, 1979) or lexical access and retrieval (Maclay and Osgood, 1959). Pauses in such instances afford speakers time to self-monitor grammar and pronunciation without bringing their speech delivery to a halt (Goldman-Eisler, 1968: 26).

Supporting the symptomatic view of pause production, studies of L2 speech show that learners produce fewer and shorter pauses as they gain proficiency in the L2 (e.g. De Jong, 2016; Mora and Valls-Ferrer, 2012). Within the study-abroad (SA) context, Huensch and Tracy-Ventura (2017a), for instance, showed that L2-English participants significantly decreased their pause durations in addition to their overall pausing rates over the course of a 12-month sojourn abroad. García-Amaya (2015) similarly showed that L2-Spanish learners decreased their production of long SPs, between 2,000 and 3,000 ms in length, upon the completion of a short-term SA program.

Pausing research emphasizes the need to consider the location of pauses in discourse; a common distinction is made with respect to between-clause pauses (i.e. pauses at clause boundaries) and within-clause pauses (i.e. pauses produced after the onset of a clause). L1 research shows that between-clause pauses aid in long-distance syntactic and message planning (Kircher et al., 2004; Levelt, 1983) whereas within-clause pauses aid in lexical retrieval (Maclay and Osgood, 1959). Regarding L2 research, learners are shown to produce within-clause pauses at higher rates than L1 speakers of the target language (Kahng, 2014; Riazantseva, 2001; Tavakoli, 2011). Skehan et al. (2016) argue that between-clause pauses are specifically linked to difficulties in L2 ‘macro-planning’, or message formulation, while within-clause pauses arise from difficulties in ‘micro-planning’ processes such as grammatical encoding and morphosyntactic computation.

Motivated by the premise that the location of L2 pauses derives from fundamentally different levels of cognitive processing, the present study will explore whether the relation between pausing and complexity undergoes fluctuations in the between-clause and within-clause positions over the course of a seven-week SA program. While listener-oriented frameworks demonstrate the communicative value that pauses hold in interactive settings (see Pickering and Garrod, 2004), the primary aim of the present analysis is to disentangle the roles played by syntactic complexity and clause location during L2 production and thus control for the discursive role held by pauses in interaction (de Leeuw, 2007; Swerts, 1998). The present research design was therefore limited to monologic oral narratives.

Pause production can vary from language to language, both in terms of pausing rate and pause duration (García-Amaya and Lang, 2020). For example, French speakers typically use fewer pauses than English speakers, yet with overall longer durations (Grosjean and Deschamps, 1975). Though English and Spanish, which are the respective L1 and L2 of this study’s participants, have been sparsely studied from a comparative standpoint, the existing data suggest that FPs are considerably more common among L1-English speakers than L1-Spanish speakers, while the SP rates of the two languages are more analogous (de Johnson et al., 1979; García-Amaya et al., 2019).

3 L2 syntactic complexity

L2 research generally associates complex syntax with an increased range and sophistication of forms, such as increased overall length and greater degree of subordination (see Ortega, 2003). Barring a single definition, Norris and Ortega (2009) adopt a multifaceted perspective of syntactic complexity which acknowledges the need to account for multiple sub-constructs of complexity, including length, subordination, coordination, and phrasal complexity. Expanding on this approach, Pallotti (2015) proposes ‘structural complexity’ as encompassing the number of linguistic elements present in an L2 utterance in addition to their relationships, which is composed of two categories: stylistic syntactic complexity, which reflects the number of interconnected constituents in an utterance or phrase; and grammatical syntactic complexity, which involves defining syntactic combinations and ranking their complexity.

Given this range of approaches, Norris and Ortega (2009) make two specific recommendations for researchers inquiring into L2 syntactic complexity. First, Norris and Ortega argue that a single measure is insufficient to capture the myriad dimensions of L2 syntactic complexity; most researchers thus characterize more complex utterances as longer utterances as well as those involving subordination (see Larsson and Kaatari, 2020; Pallotti, 2015). Second, Norris and Ortega recommend that researchers should tailor complexity measures to the nature of the analysed data. To this point, oral data is generally analysed through analysis of speech units (ASUs), defined as ‘a single speaker’s utterance consisting of an independent clause, or sub-clausal unit, together with any subordinate clause(s) associated with either’ (Foster et al., 2000: 365).

The present study explores the relationship between syntactic complexity and pause production in L2 oral narratives. Although previous studies provide qualitative evidence suggesting that increased syntactic complexity results in increased hesitations in the L2 (Kahng, 2014; Towell et al., 1996), the relation between these two areas of linguistic performance is rarely studied in a systematic fashion. To our knowledge, Mirdamadi and De Jong (2015) is the single study to evaluate whether utterances with varying degrees of syntactic complexity mediate learners’ pause usage. Using a controlled sentence-construction experiment, Mirdamadi and De Jong (2015) tested the effect of syntactic complexity on pausing patterns in which Dutch-speaking L2 learners of English produced active and passive sentences based on pre-established computer prompts. In showing that the learners produced more hesitations in passive sentences, the authors argued that increased hesitation rates originate from the complex syntactic operations required by the less automatized passive structure. The present research builds upon Mirdamadi and De Jong’s study by implementing a longitudinal design using oral narratives. In addition, this study includes a more fine-grained account of pause usage by coding for FPs separately from SPs, and by taking into account whether pauses produced in syntactically complex utterances are more common in the between-clause or within-clause position.

With respect to SA research, previous studies do not analyse the possible trade-offs between syntactic complexity and pause production. However, Long and Solon (2021) investigated the impact of a short-term SA program on syntactic complexity alone in spontaneous speech. The authors measured complexity via three dimensions – ASU length, number of clauses per ASU, and length of clause – and found no significant gains in the learners’ speech after four weeks studying L2 Spanish in comparison to at-home learners. Although Long and Solon conclude that the four-week period is not long enough to bring about sufficient gains in syntactic complexity, other SA programs with durations longer than three months appear to yield more robust developmental gains (see Jensen and Howard, 2014; Lennon, 1990).

4 CALF research and ‘trade-off’ effects in L2 oral performance

CALF, or the complexity–accuracy–lexis–fluency framework, is an analytic approach in second language acquisition (SLA) from which researchers study L2 learners’ oral performance under different cognitive, planning, and task-based demands (Housen et al., 2012; Michel, 2017). Towell (2012) summarizes the importance of CALF studies for understanding interlanguage development in the following way: ‘For second language acquisition to succeed and for learners to be able to use complex language accurately and fluently, it is essential for all three dimensions to be successful and to be integrated with each other’ (p. 47).¹ A key tenet driving CALF research is that learners’ performance under different variations of a production task is regulated by the same cognitive mechanisms that promote successful interlanguage development, such as noticing, attention and working memory (Michel et al., 2007; Robinson, 2011). Crucially, since learners do not possess native-like competence in their L2, they develop strategies for focusing on one domain of L2 knowledge while temporarily bypassing or ‘trading-off’ others (De Jong et al., 2015; Housen and Kuiken, 2009).

Two frameworks make predictions related to trade-off effects in L2 performance vis-à-vis cognitive capacities. First, Skehan’s (1998) Limited Attention Capacity Hypothesis rests on the notion that knowledge is not automatized in the L2; learners’ limited attentional resources and working memory inhibit their ability to simultaneously attend to all linguistic components. Skehan’s framework presumes a close relation between L2 disfluencies and Levelt’s (1989) three stages of oral production. Second, Robinson’s (2003, 2011) Cognition Hypothesis is grounded in the idea that learners have more constant attentional capacities during speech production and can therefore draw from many pools of cognitive resources. Although Robinson argues that ‘trade-offs’ do not exist under a strict reading of the term, Robinson hypothesizes that learners are more likely to redirect their attentional resources toward the production of more complex and accurate speech while engaging in complex tasks. The present study draws upon areas where Skehan and Robinson converge: the notion that increased cognitive processing, especially in terms of morphosyntactic computation, results in decreased fluency performance (Skehan, 2018: 56–67, 80). The next section lays out the research questions and hypotheses that motivated the present study.

III Research questions

The previous sections synthesized the need for systematic analyses that probe how syntactic complexity is intertwined with pause usage in the L2. The present study was guided by four research questions, and each research question was further grounded in a specific hypothesis (hypotheses 1 to 4).

Research question 1: How do pausing phenomena develop in a seven-week SA experience?

Per hypothesis 1, we hypothesize that learners’ pause patterns, operationalized here through SP and FP rates, will decrease over the course of their stay abroad. This hypothesis is supported by the findings of previous research showing that learners’ overall pause usage diminishes as they gain experience in the target language (García-Amaya, 2015; Huensch and Tracy-Ventura, 2017a; Mora and Valls-Ferrer, 2012).

Research question 2: How does syntactic complexity develop in a seven-week SA experience?

Per hypothesis 2, we hypothesize that learners’ capacities to produce complex syntax, measured through length and subordination, will remain stable over the course of their stay abroad. We base this hypothesis on a near-consistent finding in the SA research demonstrating that short-term SA programs are the least likely to facilitate gains in syntactic complexity (Long and Solon, 2021).

Research question 3: What is the effect of syntactic complexity on L2 pause production?

Per hypothesis 3, we hypothesize that a trade-off should emerge between learners’ pause usage and the complexity of the surrounding utterance. Specifically, we expect that learners will show greater pause usage in more complex utterances (i.e. longer utterances and utterances involving clausal subordination) than in less complex utterances (Mirdamadi and De Jong, 2015; Robinson, 2011; Skehan, 2018).

Research question 4: What is the combined effect of syntactic complexity and pause type on the location of L2 pauses?

Per hypothesis 4, we hypothesize that learners will produce more within-clause pauses, compared to between-clause pauses, during the delivery of the most complex utterances. Hypothesis 4 rests on the premise that learners’ use of within-clause pauses originates from difficulties in morphosyntactic computation and syntactic processing (Kahng, 2014; Tavakoli, 2011) due to difficulties experienced in online Formulator work (Levelt, 1989). Per the findings of Kahng (2014, 2020), we additionally predict that learners will rely on within-clause SPs to a greater extent than on within-clause FPs while uttering the most complex clauses.

IV Materials and methods

1 Participants

The design of this study included 16 native speakers of English who were L2 learners of Spanish. All learners participated in a seven-week SA summer program in León, Spain and had just completed their junior year of high school in the United States. Before arriving, they were expected to have reached a Spanish proficiency-level equivalent to the A2 or B1 level of the Common European Framework of Reference for Languages. Table 1 provides demographic information about the learner group gathered via the Language Contact Profile (Freed et al., 2004). While abroad all learners lived with host families and enrolled in five courses lasting six hours per day, which amounted to 144 hours of formal instruction (in the seventh week, no official classes were held). The learners additionally signed a language pledge, which stipulated that they would speak in the target language at all times during their stay abroad except in cases of emergency. The participants and their parents agreed to this language pledge as part of the application process.

Table 1.

Demographic and language-background information for the 16 study-abroad (SA) participants (12 female; 4 male).

	M	SD
• Age (years)	17.04	0.45
• Years of previous Spanish instruction	4.41	2.04
• Years of previous English instruction	11.48	0.78
• Number of Spanish courses taken prior to onset of the study	5	0

Notes. The age range for the participants was 16–18 years. Since most participants were under 18 years, both they and their parents were required to provide consent to participate in this study.

2 Data collection

The data collection took place at three times during the SA experience: the first day of the first week of classes (Time 1, or T1), the last day of the third week of classes (Time 2, or T2), and the last day of the sixth week of classes (Time 3, or T3). At each Time, the learners saw a Microsoft PowerPoint presentation which guided them through the monologic speaking task. The PowerPoint included instructions for the learners to watch a video, embedded in the succeeding slide, and specified that they should watch the video in its entirety prior to retelling the plot that they had just watched. The instructions clearly stated that the learners should not re-access the video during their retells. At each Time, the learners watched two unique videos of Simon Tofield’s ‘Simon’s Cat’, for a total of six retells per learner during the study. The ‘Simon’s Cat’ videos are 1–2 minutes long and feature a mischievous cat with its owner Simon. The videos are fully animated and include sound effects, although there is no dialogue between the characters. We did not collect data in the learners’ L1 English since they had signed a language pledge prior to sojourning abroad.

The speech data were recorded using a PMD620 Marantz digital recorder and a Shure head-mounted microphone. The learners sat alone in a quiet university classroom while completing the task. In case they had questions about the on-screen instructions, they knew to call the attention of the researcher, who sat outside of the classroom while the recordings took place. For the 16 learners, we amassed a corpus of 12,742 words, 1,504 ASUs, 5,292 SPs and 1,836 FPs (for per-learner sums, see Appendix 1, Table A1). The total recording time across all learners and Times was 96 minutes.

3 Data analysis

a ASU transcriptions and pause labeling

Each video retell was transcribed in CHAT format using the CLAN software (MacWhinney, 2000). The transcriptions included all speech in addition to codes for FPs and SPs. Within each CHAT transcription, a single line constituted a distinct analysis of speech unit (ASU), following the guidelines provided in Foster et al. (2000), as in the example shown in example (1). It is important to acknowledge that SPs can be further classified into ‘micro pauses’ (or articulation-motivated SPs) or hesitation-motivated SPs (see De Jong and Bosker, 2013). Some common cut-off points for differentiating between these two types of SPs include 100 ms (Riazantseva, 2001), 250 ms (Kahng, 2014), and 400 ms (Tavakoli, 2011). Per De Jong and Bosker (2013), the threshold of 250 ms demonstrates the strongest correlations with levels of L2 proficiency; we thus implemented 250 ms as the threshold in our analysis. Each SP’s duration was confirmed to be above this threshold through manual analysis in Praat (Boersma and Weenink, 2021) and was subsequently marked in the CLAN transcripts. FPs were labeled auditorily and followed vowel-only (uh), nasal-only (mm), or combination vowel-and-nasal (um) vocalization patterns. FPs are not typically cataloged based on overall duration values, thus all FPs were included in the analysis. To our knowledge, there is no previous research suggesting that different lengths of FP production are linked to separate cognitive processes.

(1) | FP al principio FP SP el gato SP FP SP FP salió FP la casa SP por una puerta muy pequeña SP FP para SP FP SP animales domésticos |

| SP y FP hay mucha nieva SP FP en la césped |

| SP el gato SP estaba muy confundido :: FP y quiere que Simón SP le FP SP darle comida |

Translation:

‘| FP At the start FP SP the cat SP FP SP FP left FP the house SP through a very small door SP FP for SP FP SP domestic animals |

‘| SP and FP there is a lot of snow SP FP on the lawn |

‘| SP the cat SP was very confused :: FP and wanted Simon SP FP SP to feed him |’

b Clause labeling and clause position

Following the guidelines established in Norris and Ortega (2009), we chose our methods of analysis for the syntactic coding based on the specificity of our data. Given the oral-narrative aspect of our methodology, we measured complexity via length through mean ASU length. We evaluated complexity via subordination by coding the clauses within each ASU based on their degree of subordination (Bulté and Housen, 2012: 37; Pallotti, 2015). Each clause was therefore tagged for its upper-level clause (ULC), for which there were three mutually exclusive possibilities: independent clauses (ICs); matrix clauses (MCs); and complex clauses (XCs).² This original classification differentiates between verbal and clausal complexity such that ICs contain a single verbal element lacking any type of subordination, MCs contain an auxiliary or modal verb in addition to a non-finite verb form (i.e. verbal subordination is present), and XCs contain at least two finite verbal elements (i.e. clausal subordination is present) and are the most complex of the three ULC types. Comprehensive definitions are given below:

IC: An independent clause (IC) contains a subject and at least one conjugated verb without any verbal or clausal subordination (e.g. [Simón entra IC] ‘Simon enters’).

MC: A matrix clause (MC) contains a subject and at least one conjugated verb in the way that an IC does, but is more complex in that it also uses a non-finite form, such as an infinitive, gerund, or past participle that results in verbal subordination (e.g. [Simón está [dibujando] en su habitación MC] ‘Simon is drawing in his room’).

XC: A complex clause (XC) contains a subject and at least one conjugated verb in the way that an IC or MC does, but also contains at least one lower-level clause subordinate to it, such as an adverbial clause, noun clause, or relative clause. An example of an XC is provided here, in which the lower-level clause is a relative clause: [hay un pájaro [que tiene el gato en sus manos] XC] ‘there is a bird that the cat has in its hands’.

Our inclusion of MCs as an element of syntactic complexity distinct from ICs and XCs is motivated by Norris and Ortega (2009) who note that tallies of verb classes (e.g. auxiliaries, conditionals, or modals) are a common proxy for measures of syntactic sophistication. Bulté and Housen (2012: 37) further argue that fine-grained measures of syntactic subordination and sophistication are necessary to achieve an informed understanding of the individual constructs driving global complexity outcomes such as subordination ratios.³

The three ULC codes were transcribed in tandem with the labeling of each pause type and its syntactic location. To code pause location, between-clause pauses, tagged as ‘bw’, were defined as any pause produced prior to the onset of vocalization. In contrast, within-clause pauses, tagged as ‘wi’, were defined as any pause embedded within the units of speech of the respective clause. Two CHAT transcriptions are included in examples (2) and (3).

(2) [&sp_ic_bw y &fp_ic_wi hay mucha nieva &sp_ic_wi &fp_ic_wi en la césped IC]

‘and there is a lot of snow on the grass’

(3) [&fp_xc_bw y quiere que Simón &sp_xc_wi le &fp_xc_wi &sp_xc_wi darle comida XC]

‘and (the cat) wants Simon to give him food’

Although our three-way classification of ULCs is intended to differentiate clause production based on the degree of subordination, we recognize that increased length is a common by-product of clausal subordination (Bulté and Housen, 2012: 22; Mirdamadi and De Jong, 2015: 113). To determine the potential overlap between degree of subordination and clause length, we calculated the mean pruned word count for each clause type. To calculate the pruned word count, we first summed the total number of words produced for each retell (this was the unpruned count) and then eliminated all repetitions, false starts, and reformulations (Foster et al., 2000; García-Amaya, 2018).⁴ This calculation verified that XCs (12.7 pruned words on average) were indeed longer than MCs (8.1 pruned words on average), which were in turn longer than ICs (6.8 pruned words on average). These differences proved significant through a mixed-effect model and pairwise tests performed with ULC TYPE as the predictor variable (p < .001).

c Dependent variables

There were seven dependent variables in this study, which were calculated via Microsoft Excel formulas. First, to operationalize the learners’ pausing patterns (per research question 1), we calculated the SP RATE and FP RATE for each video retell. These rates were computed by dividing the total number of pauses by each retell’s pruned word count. The resulting figures (see Section V.1) will display the rates per 100 words (rather than per 1 word) for ease of interpretation, although the rates themselves were not normalized as pauses per 100 words.

Regarding syntactic complexity (per research question 2), we followed the guidelines proposed by Norris and Ortega (2009), who suggest that two core dimensions of syntactic complexity include overall length and subordination. Our measure of overall length was MEAN LENGTH OF ASU, calculated by dividing the pruned word count of each retell by the total number of ASUs. We additionally included two subordination measures. The first subordination measure was CLAUSES/ASU, calculated by dividing the total number of clauses by the total number of ASUs produced in each retell. For the second subordination measure, we averaged the ULC COUNT for each ULC TYPE (i.e. ICs, MCs, and XCs) per each Time and each participant.

To evaluate the relation between pause and clause production (per research question 3), we developed a PAUSE-TO-CLAUSE RATIO, which divides each learner’s pause counts by their clause counts. To evaluate the relative use of within- and between-clause pauses (per research question 4), we created a PAUSE–LOCATION RATIO. This PAUSE–LOCATION RATIO first averages over both videos the number of within-clause pauses, number of between-clause pauses, and pruned word count for each ULC TYPE, PAUSE TYPE, TIME, and speaker; afterwards the averaged numbers of within-clause pauses are divided by the averaged numbers of between-clause pauses and the averaged pruned word counts (again for each speaker, TIME, PAUSE TYPE, and each ULC TYPE).

4 Statistical analysis

For research question 1, the two dependent variables were FP RATE and SP RATE. We fitted generalized linear mixed-effects models (LMEMs) to the two dependent variables using the lme4 package (Bates et al., 2015) within the statistical software package R, version 4.0.5 (R Core Team, 2021). To ensure that the assumption of normality of the residual terms was met, both dependent variables were log-transformed. As fixed factors, we included the effects of TIME (the predictor variable of interest) and VIDEO ORDER (a control variable). VIDEO ORDER was included to account for differences in fluency behavior over the course of each two-video recording session (for a summary of the effects of this variable, see Table 2). We further included random-intercept effects for SPEAKER to account for within-speaker dependencies.

Table 2.

Structures of the final regression models used in this article.

Research question	Outcome	3-way	2-way	2-way	2-way	Main	Main	Main	Control
1	FP RATE						TIME		VIDEO ORDER
1	SP RATE						TIME		–
2	MEAN LENGTH OF ASU						TIME		–
2	CLAUSES/ASU						TIME		–
2	ULC COUNT				ULC TYPE * TIME	ULC TYPE	TIME
3	PAUSE-TO-CLAUSE RATIO	–	TIME* PAUSE TYPE	–	–	ULC TYPE	TIME	PAUSE TYPE
4	PAUSE–LOCATION RATIO	–	TIME* PAUSE TYPE	–	TIME* ULC TYPE	ULC TYPE	TIME	PAUSE TYPE

Notes. A blank cell indicates that an interaction or main effect was not tested; a dash (‘–’) indicates that the effect was removed due to lack of significance. For the last three dependent variables, VIDEO ORDER was not included in the models because the outcome measures were averaged across both videos. For the first four models, TIME was included as the lone variable of interest, while for the last three models, the top-down approach was implemented. For FP RATE, the effect of VIDEO ORDER was that learners had a lower FP RATE in Video 2 than in Video 1. ASU = analysis of speech unit. FP = filled pause. SP = silent pause. ULC = upper-level clause.

For research question 2, we followed the same procedures described in the previous paragraph for the dependent variables MEAN LENGTH OF ASU and CLAUSES/ASU. Regarding the subordination index (i.e. IC, MC, XC), we established one dependent variable: ULC COUNT. For ULC COUNT, we fitted an LMEM using the main effects of ULC TYPE and TIME, their interaction, and the main effect of VIDEO ORDER (as a control variable) as fixed factors, and a random-intercept effect for SPEAKER.

For research question 3, we developed the PAUSE-TO-CLAUSE RATIO, computed for each PAUSE TYPE, per ULC TYPE, per SPEAKER, per TIME. For research question 4, we developed a PAUSE–LOCATION RATIO (again, per PAUSE TYPE, per ULC TYPE, per SPEAKER, per TIME). The latter ratio’s distribution was skewed and included a high number of zero values; we thus fitted a beta-regression model with a logit link using the R package glmmTMB (Brooks et al., 2017) for this outcome variable. This model included TIME, PAUSE TYPE, ULC TYPE, and their interactions as fixed effects (including the three-way interaction), as well as VIDEO ORDER as a control variable. The random-intercept effect was SPEAKER.

For the dependent variables ULC COUNT, PAUSE-TO-CLAUSE RATIO, and PAUSE–LOCATION RATIO, we followed the top-down strategy for fixed-effects selection (West et al., 2014: 39). Table 2 shows the structures of the final regression models used in this article.

We computed type III F-statistics for the predictor variables of interest and their interactions. As a measure of effect size, we computed marginal R² (i.e. the percentage of variance explained by the reported effect adjusted for the other predictors in the model) using the R package r2glmm (Jaeger, 2017; Jaeger et al., 2017) for the LMEMs. In Section V, we report marginal R² values for the test results discussed in the text only. This data reporting may lead to difficulties in comparing the effect-size values of the main effects with those of the interactions, since the interaction values are reported without considering their main-effect values. To facilitate holistic interpretations, we include the complete list of marginal R² values in Tables A2–A4 in Appendix 1. As an estimate of effect size for the beta-regression modeling, we use the differences of whole-model marginal R² computed by the package performance (Lüdecke et al., 2021) from the models including and excluding the effect of interest. Throughout this article, we apply a significance level of α = 0.05 for model selection; for pairwise comparisons we adjust the significance level using the numbers of comparisons performed (see Section V.2).

Regarding data visualization, we show descriptive plots for the results from research questions 1 and 2. For the results from research questions 3 and 4, we computed marginal predicted-means plots with 95%-confidence intervals based on the fitted models using the R package ggeffects (Lüdecke, 2018). This visualization facilitates model-based inference regarding the interaction effects.

V Results

1 Pause rates

Figure 1a plots the FP RATE values aggregated for all learners at each Time (for the means and standard deviations, see Table 3). For this outcome, there were minimal differences across the three Times, with a mean increase of 0.1 FPs per 100 words between T1 and T2, and a mean decrease of 0.9 FPs per 100 words between T2 and T3. The LMEM did not return a significant effect of TIME on FP RATE (F(2,77) = 0.012, p = .988, marginal R² = 0.000).

Figure 1.

(a) FP RATE (FP count/pruned word count) per TIME. (b) SP RATE (SP count/pruned word count) per TIME.

Table 3.

Mean and SD values for FP RATE and SP RATE by TIME (values are per 100 words).

TIME	FP RATE		SP RATE
TIME	Mean	SD	Mean	SD
T1	15.0	11.1	56.5	17.1
T2	15.1	9.9	44.4	13.6
T3	14.2	8.4	37.1	14.2

Notes. FP = filled pause. SP = silent pause.

Figure 1b plots the learners’ SP RATE values (see also Table 3). The learners’ mean SP RATE decreased over time; the decrease was of 12.1 SPs per 100 words between T1 and T2, and of an additional 7.3 SPs per 100 words between T2 and T3. The LMEM returned a significant effect of TIME on SP RATE (F(2, 78) = 31.613, p < .001, marginal R² = 0.219).

2 Complexity

We calculated one measure of complexity via length, MEAN LENGTH OF ASU, which is plotted in Figure 2a (see also Table 4). For this measure the learners demonstrated only small differences between each data-collection session, and the LMEM did not return a significant effect of TIME (F(2, 78) = 1.720, p = .186, marginal R² = 0.020).

Figure 2.

(a) MEAN LENGTH OF ASU (pruned word count/ASU count) per TIME. (b) CLAUSES/ASU per TIME.

Table 4.

Mean and SD values for MEAN LENGTH OF ASU and CLAUSES/ASU by TIME.

TIME	LENGTH OF ASU		CLAUSES/ASU
TIME	Mean	SD	Mean	SD
T1	8.17	1.78	1.30	0.26
T2	8.41	1.27	1.14	0.14
T3	8.56	1.19	1.24	0.15

Next, we measured complexity via subordination. The first subordination measure was CLAUSES/ASU, plotted in Figure 2b (see also Table 4). The learners’ mean number of clauses per ASU decreased from T1 to T2 (decrease of 0.16 clauses/ASU) and then increased from T2 to T3 (increase of 0.10 clauses/ASU). The LMEM returned a significant effect of TIME on CLAUSES/ASU (F(2, 78) = 5.975, p = .004). Of importance, the model returned a marginal R² value of 0.136 for TIME, which indicates that only 13.6% of the total variance is explained by this predictor.

We also coded each clause per the subordination index described in Section IV.3.b. Figure 3 plots the values for ULC COUNT at each time point (for aggregated counts for all learners, see Appendix 1, Table A5; for mean ULC COUNTS by TIME, see Table A6). Figure 3 shows that the ICs were consistently more frequent than both MCs and XCs. The MCs were generally more frequent than the XCs, with the dispersion between them being the greatest at Time 3. The LMEM returned a significant result for the interaction of TIME * ULC TYPE (F(4, 120) = 2.195, p = .047, marginal R² = 0.380).

Figure 3.

Model-predicted marginal means and 95% confidence intervals for ULC COUNT per TIME and ULC TYPE.

We additionally performed stepwise comparisons for ULC COUNT (i.e. T1 versus T2; T2 versus T3), shown in Table 5. These comparisons revealed significant differences between T1 and T2 for both ICs and MCs, and between T2 and T3 for XCs.

Table 5.

Pairwise, stepwise comparisons for ULC TYPE.

ULC TYPE	Comparison	Test result
IC	T1 vs. T2	DF = 120, t-ratio = −3.302, p-value = 0.001
IC	T2 vs. T3	DF = 120, t-ratio = −0.088, p-value = 0.930
MC	T1 vs. T2	DF = 120, t-ratio = −3.385, p-value = 0.001
MC	T2 vs. T3	DF = 120, t-ratio = −2.435, p-value = 0.016
XC	T1 vs. T2	DF = 120, t-ratio = −0.625, p-value = 0.533
XC	T2 vs. T3	DF = 120, t-ratio = −2.851, p-value = 0.005

Notes. Bold p-values indicate test results that are below the adjusted alpha level of 0.008 (calculated as .05 divided by 6, since six comparisons were tested). ULC = upper-level clause.

To summarize Sections V.1 and V.2, the learners demonstrated a significant decrease in SP RATE over time, while their values for FP RATE, MEAN LENGTH OF ASU, and CLAUSES/ASU remained stable between T1 and T3 (although, for CLAUSES/ASU, the T2 values were lower than T1 and T3). Regarding the subordination index, the learners experienced the greatest gains in the usage of the least complex ICs and MCs during the first half of the SA, and of the more complex XCs during the second half of the SA. Section V.3 next considers whether trade-offs may have occurred between pausing and complexity in the learners’ speech.

3 Pause-to-clause and pause–location ratios

To explore the relation between pausing and syntactic complexity, we computed the PAUSE-TO-CLAUSE RATIO, which divides learners’ pause counts by the respective clause counts, plotted on the y-axis of Figure 4 (see also Appendix 1, Table A7). For this outcome, values above 1 indicate that learners produced more than one pause per clause type on average, while values below 1 indicate that learners produced less than one pause per clause type. The x-axis separates the data based on PAUSE TYPE, the vertical panels separate the data according to TIME, and the colors signify the three categories of ULC TYPE. For the outcome PAUSE-TO-CLAUSE RATIO, the model returned a significant main effect of ULC TYPE (F(2, 261.09) = 77.412, p < 0.001, marginal R² = 0.239), as well as of the two-way interaction of TIME * PAUSE TYPE (F(2, 261.02) = 8.108, p < 0.001, marginal R² = 0.032). Regarding the main effect of ULC TYPE, the ratios were consistently higher for the most complex clauses (XCs) than for the lesser complex MCs and ICs (by a predicted margin of 1.4–2.4 for SPs, and of 0.7–1.0 for FPs). Regarding the interaction effect of PAUSE TYPE * TIME, Figure 4 further shows that the learners’ SP rates decreased over time congruently across the three ULCs, whereas their FP rates remained stable.

Figure 4.

Model-predicted marginal means for PAUSE-TO-CLAUSE RATIO (calculated as number of pauses/number of clauses), per PAUSE TYPE, per TIME, and per ULC TYPE.

The second ratio was the PAUSE–LOCATION RATIO, which divides the number of within-clause pauses by the number of between-clause pauses by the ULC TYPE-specific pruned word count (see Appendix 1, Table A8). In the marginal-effects plots in Figures 5 and 6, higher values indicate that learners produced more within-clause pauses than between-clause pauses. The LMEM uncovered a significant effect of the two-way interaction of ULC TYPE * TIME (Chi²(4) = 11.4246, p = .022, marginal R² = 0.039). Figure 5 shows that the learners’ ratios generally remained stable for ICs and MCs during the SA. For XCs, however, a notable data pattern is present between T2 and T3, with the learners demonstrating the sharpest decrease in their ratios during the second half of the program.

Figure 5.

Model-predicted marginal means for PAUSE–LOCATION RATIO, calculated by dividing the averaged numbers of within-clause pauses by the averaged numbers of between-clause pauses and the averaged pruned word counts (for each speaker, TIME point, PAUSE TYPE, and each ULC TYPE), shown per the significant interaction of TIME * ULC TYPE.

Figure 6.

The LMEM additionally uncovered a significant two-way interaction of PAUSE TYPE * TIME (Chi²(2) = 7.222, p = .027, marginal R² = 0.022). Per Figure 6, the learners’ ratios decreased over time for SPs, while in general they remained stable for FPs. These patterns mirror the general findings for SP RATE and FP RATE reported in Section V.1.

VI Discussion

1 Responding to the research questions and hypotheses

Research question 1: Pause rates

The first research question asked how FP and SP rates develop during a seven-week SA experience. The findings were divergent for the two dependent variables, with the model for SP RATE showing a significant effect of TIME, but the model for FP RATE showing no effect.

Research question 2: Syntactic complexity

The second research question considered the learners’ syntactic-complexity patterns over time. We analysed the data through three dependent variables: MEAN LENGTH OF ASU (complexity via length); CLAUSES/ASU (ratio of complexity via subordination); and ULC COUNT (complexity via degree of subordination). We found TIME to be a significant predictor for the two subordination measures, but not for the measure of length. The data for CLAUSES/ASU showed lower T2 values compared to T1 and T3; however, the effect size for TIME was small, and there was high variation at Time 1 as interpreted through the SD values (Table 4). Regarding ULC COUNT, the ICs were the most common clause type at each Time, followed by the more complex MCs and XCs. The pairwise comparisons showed that the learners increased their use of ICs and MCs the most between T1 and T2, but increased their use of XCs the most between T2 and T3. Taken together, these findings offer partial confirmation for hypothesis 2 that the learners’ syntactic complexity patterns should remain stable over time. As support for this hypothesis, the measure for overall length remained stable. However, contrary to hypothesis 2, the learners showed significant development in their subordination capacities at all recording sessions after Time 1.

Research question 3: The relation between pauses and complexity

The third research question considered the relation between pause usage and syntactic complexity through examining the learners’ pause-to-clause ratios. The data showed that: (1) there were higher ratios in the production of XCs (the most complex clauses) compared to MCs and ICs; (2) the ratios for SPs were higher than for FPs; and (3) the SP ratios decreased over time across the three clause types, whereas the FP ratios remained stable. These findings offer partial confirmation for hypothesis 3: the learners showed more frequent pause production during the elocution of XCs in comparison to MCs and ICs; however, they did not show appreciable differences between MCs and ICs.

Research question 4: Complexity and pause location

The results for the pause–location ratio were as follows: (1) the IC and MC ratios remained stable over time; (2) the XC ratios were higher at Time 2 than at Time 3, indicating that learners were the most prone to use such pauses as a within-clause hesitation strategy at mid-program; and (3) the ratios for SPs decreased over time while they remained stable for FPs. Taken together, the findings offer partial support for hypothesis 4: (1) when we found evidence that the learners favored within-clause pauses over between-clause pauses for a specific ULC TYPE, it was for the most complex clauses (XCs at Time 2); (2) however, we did not find evidence to support the expectation that the learners would use within-clause SPs to a greater extent than within-clause FPs in such situations.

2 Trade-offs between complexity and fluency

This study uncovered a trade-off between complexity and fluency that was consistent across the three recording sessions. Specifically, the learners’ production of ULCs involving clausal subordination (i.e. increased performance on complexity) coincided with increased pause usage (i.e. diminished performance on fluency) in comparison to ULCs lacking clausal subordination (Figure 4). Mirdamadi and De Jong (2015) report an analogous outcome in their analysis of active and passive sentences produced by Dutch-speaking L2 learners of English. The authors proposed that the learners’ increased hesitation rates in passives resulted from the syntactic operations required by this more complex structure. The present results build on the findings of Mirdamadi and De Jong by demonstrating that increased pause usage coincides with the elaboration of another syntactically complex element, namely clausal subordination. Whereas the study by Mirdamadi and De Jong grouped FPs and SPs together in the labeling of hesitation occurrences, we classified the production of the two pause types separately. In doing so, we showed that the learners did not differentially favor either of the two pause types when producing more complex syntax.

Our results are additionally consistent with the proposal from Robinson’s (2011) Cognition Hypothesis that monologic tasks should promote fluency performance (operationalized here through pausing rates) over complexity performance. When comparing the learners’ pausing and complexity abilities over the course of the SA, we saw that decreases in SP rates (i.e. increased performance on fluency between T1 and T2) were a precursor to increases in the usage of clausal subordination (i.e. increased performance on complexity between T2 and T3). Put differently, the findings suggest that fluency gains precede complexity gains in a monologic task. Had we collected the data in a dialogic context, we might predict the inverse outcome such that the learners would experience complexity gains prior to fluency gains. In this respect, Tavakoli (2018) examined data produced by L2-English speakers over a four-week interval, incorporating both monologic and dialogic tasks in the research design. The learners from that study experienced significant gains in syntactic complexity in a dialogic task requiring discussion and persuasion, but not in a monologic task involving narrative retells. With respect to the learners’ fluency performance, however, this dimension remained stable over the course of the study.

3 Implications for models of speech production

The findings of this study generate implications for models of speech production. The three core components of Levelt’s (1989) model of speech production include preverbal message planning (the Conceptualizer), grammatical and phonological encoding of lemmas (the Formulator), and the translation of the phonetic plan to speech output (the Articulator). The learners of the present study decreased their use of within-XC pauses (relative to between-XC pauses) between Times 2 and 3. Tavakoli (2011: 77) suggests that L2 users’ production of mid-clause pauses is highly reflective of the greater information-processing load associated with complex grammatical structures in the L2. Building on this premise, the fact that the learners showed higher rates of XC-internal pause usage at Time 2 suggests that they were experiencing difficulties at the level of morphosyntactic encoding (the Formulator) required by clausal subordination. Importantly, when the learners demonstrated a marked increase of XC usage between Times 2 and 3 (Figure 3), this was coupled with a marked decrease in the XC’s PAUSE–LOCATION RATIO values (Figure 5). The learners’ increased XC usage between Times 2 and 3 likely made the elocution of these complex clauses a more automatized process by the end of the SA, which is reflected in their diminished usage of XC-internal pauses at Time 3.

Our finding that the ICs and MCs showed generally similar values for each of the pause-to-clause and pause–location ratios suggests that the syntactic sophistication associated with auxiliary or modal structures, such as MCs, did not create extra processing difficulties for the learners in comparison to clauses lacking any type of subordination. Future studies might further scrutinize the effect of syntactic variety on L2 hesitations (e.g. Ellis and Yuan, 2005). While in the present study we grouped auxiliaries and modals conjointly under the MC label, previous studies suggest that these structures may differentially influence L2 oral performance (see Bulté and Housen, 2012).

4 Implications for conceptualizing FP and SP usage in L2 speech

This study generates new findings regarding the cognitive functions reflected by SPs and FPs in L2 speech. Although Skehan (2003) did not originally hypothesize whether FPs should comport differently from SPs in L2 speech, Skehan (2018: 240–242) more recently suggests that the differential use of FPs and SPs is more likely to be enhanced in dialogic rather than in monologic tasks. Primarily⁵ synthesizing data from L2-English research, Skehan (2018) suggests that FPs produced in dialogic settings reflect a greater communicative engagement on the part of the speaker. For example, L2 learners often employ FPs as a floor-holding strategy when completing online Formulator work, whereas they resort to SPs to overcome more serious processing challenges that require engagement at the level of the Conceptualizer. In monologic settings, Skehan explains that learners are likely to favor mid-clause pauses to overcome difficulties originating from morphosyntactic computation or lemma retrieval. Of importance, Skehan draws no distinction between types of pauses and their utterance location, offering no commentary as to whether learners might differ in their usage of within-clause FPs or SPs. Through the present methodology, we opened space to pursue this topic by specifically testing whether a significant interaction was present between ULC TYPE, PAUSE TYPE, and TIME for the learner’s pause–location ratios. The interaction was not significant, which suggests that the learners indiscriminately used the two pause types in the clause-internal position. While this result adds valuable nuance to Skehan’s (2018) most recent proposal, it contrasts with Kahng (2020), who showed that L2-English learners’ rates of within-clause SP usage, but not of within-clause FP usage, correlate with their L2-specific speed of syntactic encoding. Of note, there are important differences between the present study and Kahng (2020). In addition to the different language combinations, the learners from Kahng’s study had spent more time abroad prior to study participation.

Regarding measures of FP usage, studies show that such metrics strongly correlate with the equivalent metric from a learner’s L1 (De Jong et al., 2015; Kahng, 2020). Although we cannot speak to the L1 pausing patterns for the learners from the present study, our results highlight the importance of implementing a granular approach to encoding the syntactic environment that modulates FP usage. By examining FP usage in tandem with syntactic complexity, we uncovered a subtle, but marked, relation between FP usage and clausal subordination: the learners used approximately two FPs on average when producing XCs, but approximately one FP on average when producing MCs and ICs (see Figure 4). Consistent with our finding, other studies report learners’ preference for FPs to be present before longer and more complex clauses (Watanabe, 2003).

To conclude this section, the present findings emphasize the need for disaggregated, multifaceted approaches to studying L2 pauses in combination with a design that tracks learner performance at multiple stages of development (Jensen and Howard, 2014). Future research should examine learners’ pause distributions in tandem with their pause durations (e.g. De Jong et al., 2015; Kahng, 2014) and further investigate the differential usage of pauses through a more controlled experimental design (e.g. Mirdamadi and De Jong, 2015). Additionally, it will be important to gather new data on pausing patterns from L1 speakers of Spanish and English to evaluate the extent to which learners approximate target-language norms (see Pellegrino et al., 2011).

5 Implications for study-abroad research

The previous literature presents a range of findings with regard to SA learners’ syntactic-complexity performance. With respect to long-term SA programs lasting more than three months, some studies show longitudinal gains in learners’ complexity measures (Jensen and Howard, 2014; Lennon, 1990), while others show more stable patterns (Mora and Valls-Ferrer, 2012; Serrano et al., 2011). Regarding short-term SA programs lasting less than three months, such experiences do not typically promote gains in learners’ syntactic-complexity abilities (see Long and Solon, 2021). Our results are not consistent with the latter findings, as in the present study the learners significantly increased the production of XCs during the second half of the SA. The increase in XC usage was coupled with the finding that the learners’ values for CLAUSES/ASU increased from T2 to T3 (see Figure 2b).

One possible explanation driving the difference between our study and previous studies on short-term SA programs might be the increased target-language engagement experienced by our participants. As part of their SA program, the learners resided with host families, signed a language pledge, and attended Spanish-language courses for approximately six hours daily. Previous research on participants from the same program (García-Amaya, 2021) suggests that the language pledge encouraged these learners to minimize L1 interactions, which likely contributed to their extensive L2 use (see also García-Amaya, 2017; Paige and Vande Berg, 2012). Under the premise that L2 linguistic gains are best promoted when in-person L2 interaction is maximized (Long, 1996), it is possible these learners’ augmented L2 use helps to explain the complexity gains reported here.⁶

6 Limitations

This study has several limitations, including: (1) the absence of data from the learners’ L1; (2) the need to disentangle the constructs of syntactic length and syntactic complexity; and (3) the absence of a dialogic task.

Regarding the first limitation, research shows that measures capturing L1 fluency behavior can aid in predicting a learner’s L2 fluency patterns (e.g. De Jong et al., 2015; Huensch and Tracy-Ventura, 2017b). Further exploration into individual learners’ L1 performance, as well as the incorporation of metrics of speed fluency (Skehan, 2003; Tavakoli and Skehan, 2005), will help to distinguish the influence of the L1 from the influence exerted by the processing and monitoring of complex syntax in the L2 (Kormos, 2006; Pang and Skehan, 2021; Tavakoli, 2018).

A second limitation concerns the need to discern whether overall length or clausal subordination ultimately drives learners’ increased pause frequency when elaborating complex syntax. In our study, XCs were differentiated from MCs and ICs in that the former included clausal subordination and were also longer in terms of words per clause (see Section IV.3.b). Although the distinction between number of units or hierarchy of relationships poses challenges for the operationalization of syntactic complexity (Bulté and Housen, 2012: 22), future studies should consider novel methods of disentangling these two constructs.

A third limitation is that we did not account for the communicative value held by pauses in interactive settings (Pickering and Garrod, 2004) along the lines of de Leeuw’s (2007) ‘signal-hypothesis’ approach. One notable result of our study was that the learners’ diminished use of within-XC pauses was coupled with a significant increase in the overall production of XCs themselves. As a caveat to this finding, the monologic task may have redirected learners’ attentional resources toward increased self-monitoring during syntactic processing. In dialogic contexts, self-monitoring is intertwined with the need to pay attention to the content of an interlocutor’s message (Segalowitz, 2010: 175) while also accessing lexical items pertinent to the topic of conversation (Kircher et al., 2004; Tavakoli, 2018).

VII Conclusions

The primary aim of this article was to address a lacuna in L2 research by examining whether trade-offs are present between complexity (syntactic length and subordination) and fluency (pause usage) in L2 Spanish. While previous findings support the idea that syntactic structure modulates pause usage in the L2 (Mirdamadi and De Jong, 2015; Towell et al., 1996), a systematic comparison between complexity and pausing in L2 narratives had yet to be performed using a longitudinal design. As a further innovation, we proposed a three-way index for coding complexity based on subordination that facilitates a more granular assessment of learners’ grammatical development. By studying learners’ pausing rates conjointly with linguistic factors, this study has established more direct connections between learners’ hesitation patterns and their assumed processing difficulties.

The first major finding was that the learners decreased their silent-pause rate over the course of the SA while at the same time increasing the number of complex structures. Second, at each data-collection session, a trade-off occurred such that the learners produced more pauses during the elaboration of utterances involving clausal subordination in comparison to utterances lacking such subordination. We thus inferred that clausal subordination, in combination with the construction of longer clauses, continued to be a relatively controlled process for the learners throughout the duration of their stay abroad. Third, we showed that the learners’ usage of within-clause pauses decreased in tandem with an increased complex-clause usage between Times 2 and 3. This combination of findings likely indicates that there is less cognitive burden as the production of complex syntax becomes more frequent and automatic, precipitating a reduction of clause-internal pauses. Moving forward, future work would do well to examine factors such as L1–L2 pairing, task type, and proficiency levels as additional predictors of L2 hesitations (e.g. De Jong, 2016).

Footnotes

Appendix 1

Table A8.

PAUSE-LOCATION RATIO per ULC.

TIME	ULC	FP	FP 95% CI	SP	SP 95% CI
T1	IC	0.039	[0.019, 0.058]	0.121	[0.047, 0.196]
	MC	0.054	[0.014, 0.093]	0.094	[0.054, 0.134]
	XC	0.073	[–0.003, 0.148]	0.153	[0.047, 0.258]
T2	IC	0.040	[0.022, 0.059]	0.058	[0.027, 0.089]
	MC	0.042	[0.019, 0.065]	0.060	[0.034, 0.087]
	XC	0.090	[0.045, 0.136]	0.121	[0.083, 0.159]
T3	IC	0.060	[0.023, 0.098]	0.062	[0.041, 0.083]
	MC	0.044	[0.014, 0.074]	0.060	[0.034, 0.087]
	XC	0.056	[0.020, 0.091]	0.066	[0.033, 0.099]

Notes. FP = filled pause. SP = silent pause. IC = independent clause. MC = matrix clause. XC = complex clause. ULC = upper-level clause.

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Lorenzo García-Amaya

Notes

References

Barr

(2001) A computer model for the Pauline epistles. Literary and Linguistic Computing 16: 233–50.

Bates

Mächler

Bolker

Walker

(2015) Fitting linear mixed-effects models using lme4. Journal of Statistical Software 67: 1–48.

Beattie

(1979) Contextual constraints on the floor-apportionment function of speaker-gaze in dyadic conversations. British Journal of Social and Clinical Psychology 18: 391–92.

Boersma

Weenink

(2021) Praat: Doing phonetics by computer: Version 6.1.42 [computer program]. Available at: http://www.praat.org (accessed January 2023).

Bortfeld

Leon

Bloom

Schober

Brennan

(2001) Disfluency rates in conversation: Effects of age, relationship, topic, role, and gender. Language and Speech 44: 123–47.

Brooks

Kristensen

van Benthem

, et al. (2017) glmmTMB balances speed and flexibility among packages for zero-inflated generalized linear mixed modeling. The R Journal 9: 378–400.

Bulté

Housen

(2012) Defining and operationalising L2 complexity. In: Kuiken

Vedder

Housen

(eds) Dimensions of L2 performance and proficiency: Complexity, accuracy and fluency in SLA. Philadelphia, PA: John Benjamins, pp. 21–46.

Campione

Véronis

(2002) A large-scale multilingual study of silent pause duration. In: Bel

Marlien (eds) Proceedings of Speech prosody 2002, Aix-en-Provence, France, 11–13 April.

Clark

Tree

JEF

(2002) Using uh and um in spontaneous speaking. Cognition 84: 73–111.

10.

De Bot

(1992) A bilingual production model: Levelt’s ‘speaking’ model adapted. Applied Linguistics 13: 1–24.

11.

De Bot

(2003) Bilingual speech: From concepts to articulation: Mündliche Produktion in der Fremdsprache. Fremdsprachen Lehren und Lernen 32: 92–103.

12.

de Johnson

O’Connell

Sabin

(1979) Temporal analysis of English and Spanish narratives. Bulletin of the Psychonomic Society 13: 347–50.

13.

De Jong

(2016) Predicting pauses in L1 and L2 speech: The effects of utterance boundaries and word frequency. International Review of Applied Linguistics in Language Teaching 54: 113–32.

14.

De Jong

Bosker

(2013) Choosing a threshold for silent pauses to measure second language fluency. In: Eklund

(ed.) Proceedings of the 6th workshop on disfluency in spontaneous speech. Stockholm: KATH Royal Institute of Technology, pp. 17–20.

15.

De Jong

Groenhout

Schoonen

Hulstijn

(2015) Second language fluency: Speaking style or proficiency? Correcting measures of second language fluency for first language behavior. Applied Psycholinguistics 36: 223–43.

16.

de Leeuw

(2007) Hesitation markers in English, German and Dutch. Journal of Germanic Linguistics 1: 85–114.

17.

Ellis

Yuan

(2005) The effects of careful within-task planning on oral and written task performance. In: Ellis

(ed.) Planning and task performance in a second language. Amsterdam: John Benjamins, pp. 167–92.

18.

Ferrari

(2012) A longitudinal study of complexity, accuracy and fluency variation in second language development. In: Housen

Kuiken

Vedder

(eds) Dimensions of L2 performance and proficiency: Complexity, accuracy and fluency in SLA. Amsterdam / Philadelphia, PA: John Benjamins, pp. 277–97.

19.

Foster

Tonkyn

Wigglesworth

(2000) Measuring spoken language: A unit for all reasons. Applied Linguistics 21: 354–75.

20.

Freed

Segalowitz

Dewey

(2004) Context of learning and second language fluency in French: Comparing regular classroom, study abroad, and intensive domestic immersion programs. Studies in Second Language Acquisition 26: 275–301.

21.

García-Amaya

(2015) A longitudinal study of filled pauses and silent pauses in second language speech. In: Lickley

Eklund

(eds) Proceedings of disfluency in spontaneous speech. Edinburgh, pp. 23–28.

22.

García-Amaya

(2017) Detailing L1 and L2 use in study-abroad research: Data from the daily linguistic questionnaire. System 71: 60–72.

23.

García-Amaya

(2018) Utterance fluency in the study abroad context: An overview of research methodologies. In: Sanz

Morales-Front

(eds) The Routledge handbook of study abroad research and practice. New York: Routledge, pp. 181–92.

24.

García-Amaya

(2021) A change of setting: Measuring language use in an overseas immersion context. Study Abroad Research in Second Language Acquisition and International Education 6: 32–58.

25.

García-Amaya

Lang

(2020) Filled pauses are susceptible to cross-language phonetic influence: Evidence from Afrikaans–Spanish bilinguals. Studies in Second Language Acquisition 42: 1077–15.

26.

García-Amaya

Boland

Coetzee

(2019) Syntactic complexity and hesitation phenomena in L1 and L2 speech: An experimental approach. Unpublished paper presented at the International Symposium of Bilingualism, University of Alberta, Alberta, Canada.

27.

Grosjean

Deschamps

(1975) Analyse contrastive des variables temporelles de l’anglais et du français: Vitesse de parole et variables composantes, phénomènes d’hésitation [A contrastive analysis of English and French temporal variables: Speech rate and component variables, hesitation phenomena]. Phonetica 31: 144–84.

28.

Goldman-Eisler

(1961) The distribution of pause durations in speech. Language and Speech 4: 232–37.

29.

Goldman-Eisler

(1968) Psycholinguistics: Experiments in spontaneous speech. London: Academic Press.

30.

Han

Z-H

Tarone

(eds) (2014) Interlanguage: Forty years later. Amsterdam: John Benjamins.

31.

Housen

Kuiken

(2009) Complexity, accuracy, and fluency in second language acquisition. Applied Linguistics 30: 461–73.

32.

Housen

Kuiken

Vedder

(eds) (2012) Dimensions of L2 performance and proficiency: Complexity, accuracy and fluency in SLA. Amsterdam: John Benjamins.

33.

Huensch

Tracy-Ventura

(2017a) L2 utterance fluency development before, during, and after residence abroad: A multidimensional investigation. The Modern Language Journal 101: 275–93.

34.

Huensch

Tracy-Ventura

(2017b) Understanding second language fluency behavior: The effects of individual differences in first language fluency, cross-linguistic differences, and proficiency over time. Applied Psycholinguistics 38: 755–85.

35.

Jaeger

(2017) r2glmm: Computes R squared for mixed (multilevel) models: R package version 0.1.2. Available at: https://cran.r-project.org/web/packages/r2glmm/ (accessed January 2023).

36.

Jaeger

Edwards

Das

Sen

(2017) An R² statistic for fixed effects in the generalized linear mixed model. Journal of Applied Statistics 44: 1086–1105.

37.

Jensen

Howard

(2014) The effects of time in the development of complexity and accuracy during study abroad: A study of French and Chinese learners of English. EUROSLA Yearbook 14: 31–64.

38.

Kahng

(2014) Exploring utterance and cognitive fluency of L1 and L2 English speakers: Temporal measures and stimulated recall. Language Learning 64: 809–54.

39.

Kahng

(2020) Explaining second language utterance fluency: Contribution of cognitive fluency and first language utterance fluency. Applied Psycholinguistics 41: 457–80.

40.

Kircher

TTJ

Brammer

Levelt

Bartels

McGuire

(2004) Pausing for thought: Engagement of left temporal cortex during pauses in speech. NeuroImage 21: 84–90.

41.

Kormos

(2006) Speech Production and Second Language Acquisition. New York: Lawrence Erlbaum.

42.

Larsson

Kaatari

(2020) Syntactic complexity across registers: Investigating (in)formality in second-language writing. Journal of English for Academic Purposes 45: 100850.

43.

Lennon

(1990) Investigating fluency in EFL: A quantitative approach. Language Learning 40: 387–417.

44.

Levelt

WJM

(1983) Monitoring and self-repair in speech. Cognition 14: 41–104.

45.

Levelt

WJM

(1989) Speaking: From intention to articulation. Cambridge: MIT Press.

46.

Long

Solon

(2021) The impact of a short-term stay abroad on L2 Spanish syntactic complexity development in narratives. Study Abroad Research in Second Language Acquisition and International Education 6: 163–88.

47.

Long

(1996) The role of the linguistic environment in second language acquisition. In: Ritchie

Bhatia

(eds) Handbook of second language acquisition. New York: Academic Press, pp. 413–68.

48.

Lüdecke

(2018) ggeffects: Tidy data frames of marginal effects from regression models. Journal of Open Source Software 3: 772.

49.

Lüdecke

Ben-Shachar

Patil

Waggoner

Makowski

(2021) performance: An R package for assessment, comparison and testing of statistical models. Journal of Open Source Software 6: 3139.

50.

Maclay

Osgood

(1959) Hesitation phenomena in spontaneous English speech. WORD 15: 19–44.

51.

MacWhinney

(2000) The CHILDES Project: Tools for analyzing talk. 3rd edition. Mahwah, NJ: Lawrence Erlbaum.

52.

Michel

(2017) Complexity, accuracy and fluency (CAF). In: Loewen

Sato

(eds) The Routledge handbook of instructed second language acquisition. New York: Routledge, pp. 50–68.

53.

Michel

Fuiken

Vedder

(2007) The influence of complexity in monologic versus dialogic tasks in Dutch L2. International Review of Applied Linguistics 45: 241–59.

54.

Mirdamadi

De Jong

(2015) The effect of syntactic complexity on fluency: Comparing actives and passives in L1 and L2 speech. Second Language Research 31: 105–16.

55.

Mora

Valls-Ferrer

(2012) Oral fluency, accuracy, and complexity in formal instruction and study abroad learning contexts. TESOL Quarterly 46: 610–41.

56.

Norris

Ortega

(2009) Towards an organic approach to investigating CAF in instructed SLA: The case of complexity. Applied Linguistics 30: 555–78.

57.

Ortega

(1995) The effect of planning in L2 Spanish narratives. Unpublished research Note 15, University of Hawai’i Second Language Teaching and Curriculum Center, Honolulu, HI, USA.

58.

Ortega

(1999) Planning and focus on form in L2 oral performance. Studies in Second Language Acquisition 21: 109–48.

59.

Ortega

(2003) Syntactic complexity measures and their relationship to L2 proficiency: A research synthesis of college-level L2 writing. Applied Linguistics 24: 492–518.

60.

Ortega

(2005) What do learners plan? Learner-driven attention to form during pre-task planning. In: Ellis

(ed.) Planning and task performance in a second language. Amsterdam: John Benjamins, pp. 77–109.

61.

Paige

Vande Berg

(2012) Why students are and are not learning abroad. In: Vande Berg

Paige

Lou

(eds) Student learning abroad: What our students are learning, what they’re not, and what we can do about it. Sterling, VA: Stylus Publishing LLC, pp. 29–58.

62.

Pallotti

(2009) CAF: Defining, refining and differentiating constructs. Applied Linguistics 30: 590–601.

63.

Pallotti

(2015) A simple view of linguistic complexity. Second Language Research 31: 117–34.

64.

Pang

Skehan

(2021) Performance profiles on second language speaking tasks. The Modern Language Journal 105: 371–90.

65.

Pellegrino

Coupé

Marsico

(2011) A cross-language perspective on speech information rate. Language 87: 539–58.

66.

Pickering

Garrod

(2004) Toward a mechanistic psychology of dialogue. Behavioral and Brain Sciences 27: 169–90.

67.

R Core Team (2021) R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing. Available at: https://www.r-project.org (accessed January 2023).

68.

Riazantseva

(2001) Second language proficiency and pausing: A study of Russian speakers of English. Studies in Second Language Acquisition 23: 497–526.

69.

Robinson

(2003) The cognitive hypothesis, task design, and adult task-based language learning. Second Language Studies 21: 45–105.

70.

Robinson

(2011) Task-based language learning: A review of issues. Language Learning 61: 1–36.

71.

Schmidt

(1992) Psychological mechanisms underlying second language fluency. Studies in Second Language Acquisition 14: 357–85.

72.

Segalowitz

(2010) Cognitive bases of second language fluency. New York: Routledge.

73.

Segalowitz

(2016) Second language fluency and its underlying cognitive and social determinants. International Review of Applied Linguistics in Language Teaching 54: 79–95.

74.

Selinker

(1972) Interlanguage. International Review of Applied Linguistics 10: 209–31.

75.

Serrano

Llanes

Tragant

(2011) Analyzing the effect of context of second language learning: Domestic intensive and semi-intensive courses vs. study abroad in Europe. System 39: 133–43.

76.

Skehan

(1998) A cognitive approach to language learning. Oxford: Oxford University Press.

77.

Skehan

(2003) Task-based instruction. Language Teaching 36: 1–14.

78.

Skehan

(2018). Second language task-based performance: Theory, research, assessment. New York: Routledge.

79.

Skehan

Foster

(2012) Complexity, accuracy, fluency and lexis in task-based performance. In: Housen

Kuiken

Vedder

(eds) Dimensions of L2 performance and proficiency: Complexity, accuracy and fluency in SLA. Amsterdam: John Benjamins, pp. 199–220.

80.

Skehan

Foster

Shum

(2016) Ladders and snakes in second language fluency. International Review of Applied Linguistics in Language Teaching 54: 97–111.

81.

Swerts

(1998) Filled pauses as markers of discourse structure. Journal of Pragmatics 30: 485–96.

82.

Tavakoli

(2011) Pausing patterns: Differences between L2 learners and native speakers. ELT Journal 65: 71–9.

83.

Tavakoli

(2018) L2 development in an intensive study abroad EAP context. System 72: 62–74.

84.

Tavakoli

Skehan

(2005) Strategic planning, task structure and performance testing. In: Ellis

(ed.) Planning and task performance in a second language. Amsterdam / Philadelphia, PA: John Benjamins, pp. 239–77.

85.

Towell

Hawkins

Bazergui

(1996) The development of fluency in advanced learners of French. Applied Linguistics 17: 84–119.

86.

Towell

(2012) Formulaic and proceduralised language in the initial and advanced stages of learning French. Language, Interaction and Acquisition 3: 93–113.

87.

Tree

JEF

(1995) The effects of false starts and repetitions on the processing of subsequent words in spontaneous speech. Journal of Memory and Language 34: 709–38.

88.

Vercellotti

(2017) The development of complexity, accuracy, and fluency in second language performance: A longitudinal study. Applied Linguistics 38: 90–111.

89.

Watanabe

(2003) The constituent complexity and types of fillers in Japanese. In: Proceedings of the 15th ICPhS. Adelaide: Causal Productions, pp. 2473–76

90.

West

Welch

Galecki

(2014) Linear mixed models: A practical guide using statistical software. Boca Raton, FL: CRC Press.

91.

Yuan

Ellis

(2003) The effects of pretask planning and on-line planning on fluency, complexity, and accuracy in L2 monological oral production. Applied Linguistics 24: 1–27.