1.1. Cognitive task complexity and attentional resources

In the last few decades, the use of tasks in L2 classrooms has been widely proposed as an optimal tool to promote language development and acquisition. Several studies have investigated the impact of the cognitive complexity of a task on the way learners allocate their attention towards language while performing the task. In the literature, there are two competing models of attention allocation: the Limited Attention Capacity (LAC) approach (Skehan, 1998, 2009, 2014, 2015) and the Cognition Hypothesis (CH) (Robinson, 2001, 2003, 2005, 2011, 2015). The former framework claims that different performance areas (complexity, accuracy, lexis and fluency) might compete for resources when a task’s cognitive complexity increases, due to limited attentional capacity. According to the LAC approach, if production is linguistically more complex, accuracy and fluency may not also be elevated; thus, increased complexity ‘might be associated with lower fluency, or raised accuracy with lower complexity’ (Skehan, 2015, p. 125). Following this perspective, fluency and complexity often go together, as do accuracy and fluency, but the least likely association is elevated complexity and accuracy (Skehan, 2009, 2014, 2015). The purpose of LAC research is to explore how task characteristics and conditions can mitigate avoidable trade-off effects between complexity and accuracy, as these performance areas tend to compete for attentional resources (Wang & Skehan, 2014). In contrast, the CH proposes a multiple resources attentional model and claims that both linguistic complexity and accuracy increase during task performance. Within the CH, Robinson (2010, 2015) presented a taxonomy of task characteristics—the Triadic Componential Framework (TCF)—and the stabilize, simplify, automatize, restructure and complexity (SSARC) model for pedagogic task sequencing, which is based on the premise that learners should perform simple tasks on relevant parameters first, and then the cognitive demands of the task should be increased on subsequent versions for cumulative learning. The TCF classifies task characteristics according to three different factors: cognitive, interactive and learner factors. By making use of cognitive factors, task complexity can be manipulated by syllabus task designers to push the learner’s output. In his model, Robinson distinguished between resource-directing and resource-dispersing factors. According to his framework, increasing the complexity of a task along the resource-directing dimensions (±here and now; ±few elements; ±spatial reasoning; ±causal reasoning; ±intentional reasoning; ±perspective taking) will direct the attention of the learner to the form of the target language and consequently will lead to a more complex and accurate output. In contrast, increasing the demands of a task along resource-dispersing variables (±planning time; ±single task; ±task structure; ±few steps; ±independency of steps; ±prior knowledge) will disperse the attentional and memory resources of the L2 learner with negative consequences for production, in respect to all components (linguistic complexity, accuracy and fluency). Robinson’s (2010, 2015) SSARC pedagogic model for task design and sequencing is based on the premise that the cognitive complexity of the task is to be increased first along the resource-dispersing dimensions and only afterwards along the resource-directing dimensions.

This study aims to analyse the effects of manipulating the cognitive complexity of a task along the resource-directing factor [±few elements] and along the resource-dispersing variable [±planning time] on the oral production of Chinese learners of European PFL, an under-researched population in CALF literature. In the next section, a brief review of previous studies investigating the factor ±planning time and studies investigating the variable ±few elements is presented.

2.1. Studies of ± planning time

Several studies have investigated the effects of pre-task planning time on learners’ oral performance. Presenting a descriptive synthesis of planning time studies, Ellis (2009) outlined some conclusions: (i) overall, strategic planning has positive effects on fluency; (ii) in respect to accuracy and complexity the effects of planning are more variable (although clearer for complexity); (iii) strategic planning has a stronger impact on fluency and complexity, suggesting that learners prioritise what they want to say, that is, the conceptualisation, rather than the formulation of a specific linguistic plan.

Skehan and Foster (1997), Mehnert (1998), Ortega (1999) and Wang (2014) confirmed that pre-task planning time had positive effects on fluency and syntactic complexity. Although in the first two studies the hypothesis of a competition for attentional resources between accuracy and complexity was supported, Ortega’s (1999) accuracy results were mixed. Yuan and Ellis (2003) found that strategic planning positively affected syntactic complexity, and Guará-Tavares (2008, 2011) reported gains in respect to accuracy. Along the same lines, Mochizuki and Ortega (2008) showed that under guided planning conditions, learners’ output was more accurate. Gilabert (2005) found that planners produced more fluent and lexically diverse speech. It should be noted that the participants in all these studies were learners of EFL, except for Mehnert (1998) and Ortega (1999), who investigated the oral performance of learners of German and Spanish, respectively. The informants were mostly from different L1 backgrounds, but Wang (2014) and Yuan and Ellis (2003) studied the oral L2 performance of Chinese native speakers. In most of these experiments, the learners were given ten minutes for pre-task planning time, but Mochizuki and Ortega (2008) gave only five minutes, and in Mehnert (1998) three experimental groups were given one, five and ten minutes, respectively. Questioning Robinson’s distinction between resource-directing and resource-dispersing characteristics, Skehan (2009, 2014, 2015) claimed that recent LAC research had shown that two resource-dispersing features, i.e., planning time and task structure, had been mis-analysed, as their impact could result in a joint increase of complexity and accuracy, as Tavakoli and Skehan’s (2005) study demonstrated. Skehan (2015) considered that planning was not a monolithic category (as sometimes it could meet the criteria of a resource-directing variable). This controversy and the lack of clear supportive evidence for the CH predictions justifies this study of planning, as some clarification is needed. Although previous planning studies yielded consistent results, some ambiguities remain concerning the actual impact of this task complexity feature on L2 learners’ oral production.

2.2. Studies of ±number of elements

Citing Sasayama, Malicka and Norris’s (2013, 2014, 2015) unpublished research synthesis, Sasayama (2015) found that the ±few elements factor was the most common operationalisation within Robinson’s resource-directing variables. However, the studies included in this synthesis yielded inconsistent findings on the impact of raising the elements of a task on CALF measures (namely in respect to the first three dimensions), because according to Sasayama (2015) the increased task complexity results varied: (i) only some studies found that accuracy was increased; (ii) syntactic complexity was positive, null or negative across studies; and (iii) two studies showed a joint increase in accuracy and lexical diversity. The impact on fluency has been more consistent, as with a greater number of elements fluency decreased in all studies. Bearing these differences in mind, this section reviews studies in which the variable ± elements was manipulated as one of the independent variables.

Robinson (2001), Michel, Kuiken and Vedder (2007, 2012), Michel (2011), Révész (2011) and Oh and Lee (2012) investigated the effects of task complexity (±few elements) and interaction (±monologic). Robinson (2001) measured the output of 44 Japanese L1 students while performing two city map tasks. The results confirmed effects on lexical variety and on fluency. Michel et al. (2007, 2012) and Michel (2011) increased the cognitive demands of two argumentative tasks. In Michel et al. (2007), the participants (Turkish and Moroccan learners of Dutch) were given a full-colour leaflet with two (simple task) or six (complex task) electronic devices. Learners were more accurate and less fluent on complex tasks, but there were no significant main effects on syntactic complexity. Michel (2011) reported different findings: increasing the number of elements affected lexical diversity, but accuracy and fluency were not affected. Révész (2011) and Oh and Lee (2012) increased the cognitive complexity of a task along the number of elements in different ways: the former used an argumentative task–the simple and complex version diverged with respect to the amount of economic resources and the number of projects to be allocated (three vs. six)–and the latter chose a narrative task, with more complex storylines and characters in the complex version of the task. Forty-three learners of English, with different L1 backgrounds, participated in Révész’s (2011) classroom-based study. Regarding general production measures, their speech was lexically more diverse and accurate, but syntactically less complex. The specific measure used showed that participants were more likely to use complex conjoined clauses while performing the complex task. Oh and Lee (2012) examined the oral production of 40 Korean university learners of English. There were no overall significant effects on linguistic complexity, accuracy or fluency. Kuiken and Vedder (2011, 2012) studied the effects of increasing the number of elements and proficiency level in a written vs. oral argumentative task performed by 44 Dutch students of Italian L2. The results contradicted the CH, as in the oral mode the manipulation of the factor ±elements led to gains in accuracy but less syntactic complexity. Sasayama (2015) researched the effects of task complexity along ±few elements; Malicka (2014) did likewise and also investigated the impact of ±spatial reasoning dimensions. Instead of using a dichotomous approach, these two studies used tasks with multiple levels of cognitive complexity: the former had four tasks and the latter three. Sasayama (2015) found that the most complex task (with more elements: nine characters in a narrative) did not elicit the best L2 performance (in terms of accuracy and linguistic complexity); the best results were elicited by the second-simplest task according to measurements (although it was designed to be the third-simplest). In contrast to Robinson’s predictions, Sasayama (2015, 2016) found that the resource-directing feature (number of elements) could have deleterious effects. The findings showed that the four tasks posed a mix of both extraneous and facilitative load, though to clearly different degrees. Malicka (2014) found that increasing task complexity along the number of elements promoted more accurate and lexically diverse production, so the CH was only partially confirmed.

Levkina (2008), Levkina and Gilabert (2012) and Sasayama and Izumi (2012) manipulated task complexity along the resource-directing ±few elements variable and the resource-dispersing ± planning time factor, based on Robinson’s CH, as in this study. The participants were all learners of EFL: in Levkina (2008), they had different L1 backgrounds; in Levkina and Gilabert (2012), the learners were Spanish and Russian L1 speakers; and Sasayama and Izumi (2012) investigated the performance of Japanese L1 high school students. Levkina (2008) and Levkina and Gilabert (2012) used a Latin Square design, and the participants were asked to perform four tasks under four different conditions. Participants in both studies received four full-colour leaflets with two (simple task) or six (complex task) holiday destinations or apartment descriptions, but Sasayama and Izumi (2012) manipulated two monologic narrative tasks. In the three experiments, participants had five minutes for the planned condition. According to Sasayama and Izumi (2012), planners’ performance was less fluent. Planning had positive results: on lexical diversity and syntactic complexity in Levkina (2008), on lexical diversity and fluency in Levkina and Gilabert (2012) and on syntactic complexity in Sasayama and Izumi (2012). Regarding the manipulation of the factor ‘±few elements’, the three studies confirmed a decrease in fluency when learners performed a task with more elements. Levkina (2008) also showed a negative impact on syntactic complexity and Sasayama and Izumi (2012) found a negative impact on accuracy. The latter study reported that a greater number of elements in a task positively affected the specific measure of syntactic complexity, and Levkina and Gilabert (2012) showed positive effects on lexical diversity. Levkina (2008) found combined effects of ‘±planning time’ and ‘±few elements’ on lexical diversity and Levkina and Gilabert (2012) reported significant overall combined effects for not only lexical diversity but also fluency. As is demonstrated by this brief literature review, research findings are somewhat mixed and not conclusive; neither Robinson’s nor Skehan’s framework has been confirmed or disconfirmed. Note that many researchers have chosen an argumentative task type, although the operationalisation of the variable ±few elements has varied.

3.1. Research questions and hypotheses

• RQ1: What are the effects of manipulating the cognitive task complexity along the resource-dispersing factor (±planning time) on the oral production of Chinese learners of PFL?
• RQ2: What are the effects of increasing the cognitive task complexity along the resource-directing variable (±few elements)?
• RQ3: To what extent does the simultaneous manipulation of planning time and number of elements affect the oral performance of these learners?

Based on the claims of the CH, the following hypotheses (H) were formulated for each of the research questions.

• H1: Increasing the cognitive task complexity along the ±planning time factor will have negative effects on all dimensions of L2 oral production.
• H2: Increasing the number of elements of a task will result in a less fluent but more accurate, complex and lexically diverse speech.
• H3: The attention-directing effect of performing a complex task along the number of elements will be boosted by decreased complexity along the resource-dispersing dimension (+planning).

4.1. Participants

Thirty-nine Chinese university learners of PFL participated voluntarily in the study. Twenty-three students were Cantonese native speakers, twelve students were Mandarin native speakers and four students were bilingual (Cantonese and Mandarin). Their mean age was 20.59; 71.8% were female and 28.2% were male. They were undergraduate students majoring in Portuguese. Concerning Portuguese language learning, they had the same educational background: 840 hours of formal language learning in PFL. The level of proficiency was between A2 and B1, as the results of the standardised exam DEPLE for PFL for the B1 CEFR level of proficiency ranged between 44% and 74%. Participants were evenly assigned to the experimental conditions.

4.2. Tasks and procedures

Two sets of monologic information-giving tasks (appendix 1) were designed on the same topic: travelling and holidays. Learners received a full-colour leaflet with two holiday destinations (countries or cities) for the simple task or six for the complex task. The input was mainly visual, to avoid lexical support in L2. The simple version of the task included the name of each country/city, three pictures, the price, the airline company and the number of travelling days. The complex task offered six destinations, four of which were in Asia, as students were more familiar with Asian cultures. Each destination had five pictures, the number of travelling days and dates, the price and four hotel icons. After examining the visual input, learners were asked to leave a message in a friend’s WeChat giving information about all possible holiday destinations. All participants performed one simple and one complex task. The order of the tasks was counterbalanced to avoid possible carryover effects from one experimental condition to another. Half of the learners (n = 19) had five minutes to plan the task, and the other half (n = 20) were only given 30 seconds to look briefly at the leaflet. Learners performed the tasks in a language laboratory environment. To validate the construct of task complexity, students were asked to rate their performance on a seven-point Likert scale affective variables questionnaire (AVQ) after completing each task.

4.3. Dependent variables – CAF measures

Bearing in mind that CAF measures are multidimensional constructs (Housen, Kuiken & Vedder, 2012; Michel, 2017; Norris & Ortega, 2009), both general and specific measures were chosen to quantify learners’ oral production. Linguistic complexity combined five measures: syntactic complexity was measured by words per clause (clause length), clauses per AS-unit and coordinate clauses per AS-unit; lexical diversity was measured by Guiraud’s Index and VOCD, computed by CLAN. For accuracy, two general measures (percentage of error-free clauses per total clauses and errors per 100 words) and four specific measures were chosen (lexical errors per AS-unit, omissions per AS-unit, morphosyntactic errors per AS-unit and the percentage of self-repairs per total errors). For fluency, three measures were calculated: two measures of speech rate (rate A, i.e., the ratio of words per minute in unpruned speech, and rate B, i.e., the ratio of words per minute in pruned speech) and one measure of fluency repair (number of repetitions, self-repairs, reformulations and false starts per minute).

4.4. Transcription and coding

To transcribe and code the data, the CLAN program (MacWhinney, 2000) was used. The selected basic unit for analysis was the AS-unit (Foster, Tonkyn & Wigglesworth, 2000). The transcription of the speech samples was carried out by the researcher and a research assistant, who were native speakers of European Portuguese. The researcher checked all of the transcripts and coding. Interrater reliability was assessed by means of percentage agreement on 10% of the data (randomly selected), and reached 97.7%.

4.5. Statistical analyses

The statistical analyses were carried out with SPSS 24.0 for Windows. To address the research questions, a repeated measures analysis of variance (ANOVA) was performed with ±task complexity as the within-subjects factor and ±planning time as the between-subjects variable. The results reported here should be interpreted with caution as multiple ANOVAs were computed in this study. The alpha-level was set at p < 0.05. Partial eta square (η_p²) effect sizes are reported for reference, as they were computed by the SPSS; Cohen’s d values were also calculated, as they are commonly used in SLA research and can help avoid bias problems reported by some studies in the use of partial eta square with small sample sizes (Larson-Hall, 2016). Cohen’s d effect sizes equal or greater than 0.2, 0.5 and 0.8 were considered small, medium and large, respectively (Cohen, 1988). A Pearson correlational analysis (appendix 2) was computed to explore the hypothetical relationship between the dependent variables, namely between complexity and accuracy performance areas, as Skehan (2009, 2014, 2015) suggested. To determine the strength of the correlation, Cohen’s (1988) guidelines were followed: r = 0.10 to 0.29, r = 0.30 to 0.49 and r = 0.50 to 1.0 were considered small, medium and large, respectively.

5.1. Complexity

5.2. Accuracy

5.3. Fluency

5.4. Perception of task difficulty

A subjective self-rating questionnaire (AVQ) was used to assess learners’ perceived difficulty of task complexity, stress, confidence, interest and motivation. The results validated the manipulation of task complexity along the number of elements, F(1, 37) = 7.42, p = 0.015, d = 0.55, as the complex task was perceived as more difficult. The perception of difficulty was not significant along the ±planning time operationalisation. Concerning the amount of time given to plan the task (‘I had time’/‘did not have time to plan the task’), learners’ perception was significant along the number of elements, F(1, 37) = 6.29, p = 0.017, d = 0.34, and along the planning condition, F(1, 37) = 7.75, p = 0.008, d = 0.79. Learners’ perception of confidence and interest reached statistical significance: the level of confidence decreased along the complex task (more elements), F(1, 37) = 7.74, p = 0.008, d = 0.53, and learners showed more interest when performing tasks under non-planning conditions, F(1, 37) = 7.04, p = 0.012, d = 0.72. Perceived stress and motivation did not yield significant results.