Task Complexity, Intentional Reasoning Demands, L2 Speech ...

Download PDF
3 downloads 46 Views 363KB Size Report
Comment
In the theory I will be describing increasing task complexity by requiring ...... complexity of tasks along resource-directing dimensions can recapitulate the effects ...
Peter Robinson Task Complexity, Intentional Reasoning Demands, L2 Speech Production, Learning and Syllabus Design

Series A: General & Theoretical Papers ISSN 1435-6473 Essen: LAUD 2008 Paper No. 721

Universität Duisburg-Essen

Peter Robinson Aoyama Gakuin University; Tokyo (Japan) Task Complexity, Intentional Reasoning Demands, L2 Speech Production, Learning and Syllabus Design

Copyright by the author 2008 Series A General and Theoretical Paper No. 721

Reproduced by LAUD Linguistic Agency University of Duisburg-Essen FB Geisteswissenschaften Universitätsstr. 12 D- 45117 Essen

Order LAUD-papers online: http://www.linse.uni-due.de/linse/laud/index.html Or contact: [email protected]

Peter Robinson

Task Complexity, Intentional Reasoning Demands, L2 Speech Production, Learning and Syllabus Design Introduction: An Overview of a Theory, and the SARC Model, of Task-based Syllabus Design Syllabus design is essentially a decision about what units should be used to organize the events constituting classroom second language (L2) instruction, and a specification of the order or sequence in which the units are to be presented to learners. Task-based approaches to language instruction assume tasks are the units of syllabus design (see e.g., Long 1985, 2006; Long and Crookes 1992; Long and Robinson 1998; Robinson 2001a, in press; Skehan 1998; Van den Branden 2006). This paper describes a theory of how such units - classroom pedagogic tasks - should be classified and sequenced so as to promote successful L2 task performance and further L2 learning, and following this it presents the results of some recent research into the issues the theory raises, and describes their implications for pedagogy and future empirical research. In the approach I will be describing, task-based syllabus design requires an initial specification, and taxonomy, of the categories and components involved in L2 pedagogic task performance. This specification is not exhaustive, but selective of those categories and components that are; a) most effective in promoting learning and performance (i.e., they are selected following some theoretical SLA motivation, and I will describe this); b) of most utility value to task designers attempting to promote the abilities needed for real-world task performance in classrooms (i.e., they are selected because they map pedagogic task demands to needed real-world task performances in a coherent way); c) and are operationally feasible for task and materials designers (i.e., the categories and components can be used in a consistent way by a wide variety of materials designers, thereby ensuring comparability of design decisions in one context, and program, with those in others). We can call the resulting taxonomy of categories and task components, using a term from dynamic/complex systems theory (see De Bot, Lowie and Verspoor 2007; Ellis and Larson Freeman 2006), the ‘phase-space’ for task design, representing all the possible ways in which pedagogic tasks can differ. Within this phase-space, actual pedagogic task design is the result of mapping the coordinates of specific real-world tasks identified by a needs analysis (say, the need to explain to a superior what caused a conflict in the workplace) to all the parameters of tasks specified by the classification system and taxonomy as available for systematic 1

manipulation. So individual pedagogic tasks will have their own ‘parameter-space’. For example - in ways which I will describe below - the parameter space for one sequence of pedagogic tasks could be +/- planning time, and +/- intentional reasoning demands. Task sequencing is the outcome of designing tasks simple on all the relevant parameters, and then gradually increasing their cognitive complexity on subsequent versions. In the theory I will be describing increasing task complexity by requiring reasoning about the intentional states that cause people to perform actions (+ intentional reasoning) versus simply describing their actions (- intentional reasoning) directs learner attention to those aspects of linguistic code which can be used to meet these complex task demands (e.g., cognitive state terms such as ‘think’ and ‘wonder’, ‘doubt’, and the complex syntactic complementation that accompanies their use, (‘X wonders if Y’, ‘X doubts that Y believes Z’ etc.). On the other hand, removing planning time (- planning time) increases cognitive complexity but simply disperses attentional resources over many dimensions of tasks with no particular linguistic correlates. I argue increasing task complexity of the former resource-directing dimensions promotes interlanguage development, and on the latter resource-dispersing dimensions it promotes increasing automatic access to current linguistic resources. Both are important. In the theory I will be describing there are only two principles for task sequencing. Principle 1: Only the cognitive demands of tasks are sequenced - the interactive demands of Task Conditions are replicated each time pedagogic task versions are performed so as to help ensure elaboration and successful transfer of the particular ‘schema’ for interactive or monologic task performance to real-world contexts of use. The theory is thus parsimonious, placing the sole emphasis for task sequencing on the cognitive factors I describe as contributing to Task Complexity. Principle 2: Increase resource-dispersing dimensions of complexity first (e.g., from + to - planning time), and then increase resource-directing dimensions (e.g., from - to + intentional reasoning). The rationale for this can be described in the following way. First, tasks simple on all dimensions are performed (e.g. + planning, - intentional reasoning). Simple task performance thus draws on the stable (S) ‘attractor state’ of current interlanguage. Next, increase complexity on resource-dispersing dimensions (e.g., - planning, - intentional reasoning). This promotes access to, and so automatization (A) of, the current interlanguage system. Third increase complexity on resource-directing dimensions (e.g., + planning, + intentional reasoning). This promotes restructuring (R) of the current interlanguage system, and the development of new form-function/concept mappings. Finally, increase complexity on both dimensions (e.g., - planning, + intentional reasoning). This approximates the demands of real-world task performance, introducing maximum complexity (C) and destabilizing the current interlanguage system - itself a phase space, where all possible states of the system are represented. Increasing task complexity by

2

sequencing shifts in task demands induces (in the theory proposed here) similar shifts in the structure of interlanguage resources used to accomplish them. Let us call the steps just described the 'SARC model' for increasing L2 pedagogic task complexity, and represent it in the following way, where i = current interlanguage state; e = mental effort; 's' = simple task demands; 'c' = complex task demands; rdisp = resource dispersing dimensions of tasks; rdir = resource directing dimensions of tasks; and n = potential number of practice opportunities on tasks, which are determined in situ by teachers observing pedagogic task performance by individuals, dyads and groups: Step 1. S = i X Step 2. A = i X Step 3. R = i X Step 4. C = i X

e e e e

('s'rdisp) + ('s'rdir) n ('c'rdisp) + ('s'rdir) n ('s'rdisp) + ('c'rdir) n ('c'rdisp) + ('c'rdir) n

1. The Cognition Hypothesis of Adult Task-Based Language Learning: The Fundamental Pedagogic Claim The fundamental pedagogic claim of the Cognition Hypothesis, underpinning the SARC model just described, is that pedagogic tasks should be designed, and then sequenced for learners on the basis of increases in their cognitive complexity alone (Robinson 1996a, 2001, 2003b, 2005a, 2007a, in press; Robinson and Gilabert 2007; Robinson, Ting and Urwin 1995). The aim of such pedagogic task sequences is to gradually approximate, in classroom settings, the full complexity of real-world target task demands, so prompting learners to shift from program-entry levels of L2 use and task performance to program-exit targeted levels along a manageable, but continuously extending, developmental and performative route. These increasingly complex sequences of pedagogic tasks should constitute the task-based syllabus and provide the basis for program-exit task-based assessment (see Harper 1986; Prabhu 1987; Long, 1985, 2006; Long & Crookes 1992; Long and Norris 2000; Norris, Brown, Hudson and Yoshioka 1998; Robinson 1996b; Robinson and Ross 1996; Schank, Berman and MacPherson 1999 for complementary pedagogic rationales). Research into the Cognition Hypothesis (Robinson 2001a, 2001b, 2003b, 2005a, 2007a) has so far addressed a number of concomitant, theoretically motivated claims that motivate (in part) the pedagogic proposal for syllabus design. These are that increasing the cognitive demands of tasks contributing to their relative complexity along certain dimensions will; a) push learners to greater accuracy and complexity of L2 production in order to meet the greater functional and conceptual communicative demands they place on the learner - so promoting interlanguage development and restructuring of the current L2 system; b) promote interaction, and heightened attention to and memory for input, so increasing learning from the input, and incorporation of lexical items and more elaborate constructional forms made salient in the input; as well as c) longer term retention of input; that d) performing simple to complex sequences will also lead to automaticity and efficient 3

scheduling of the components of complex L2 task performance; and that e) individual differences between learners in task-relevant abilities will increasingly differentiatelearning and performance4 on tasks as they increase in complexity. The study reported later in this paper specifically addresses claims a), b) and e), finding evidence to support each of them. 1.1. A Taxonomic System for Task Classification and Sequencing: The Triadic Componential Framework Gradually approximating target-task demands on increasingly complex pedagogic task versions requires both; i) an operational taxonomy for classifying features of target tasks which can be simulated by task designers, and performed and practiced by L2 learners in pedagogic settings (at first separately, and then increasingly in combination) and also; ii) principles for sequencing these features, and combinations of them, in an order which approximates target-task demands. A taxonomic system (see Robinson 2007a; Sokal 1974) for pedagogic task classification which can accommodate the fundamental pedagogic claim of the Cognition Hypothesis should therefore be both; i) descriptively adequate, “listing” in its taxonomy important features of target-task performance which are susceptible to pedagogic design and simulation in classroom settings, and also; ii) theoretically motivated, capturing in its taxonomic “system-structure” categories of the design features of pedagogic tasks which can be simulated and sequenced to promote both automatization of, and fluent access to the current L2 interlanguage knowledge needed to accomplish task demands, and also categories of the design features of tasks that can be simulated and sequenced to promote further analysis and development of existing interlanguage knowledge in line with the target L2. Figure 1 describes a taxonomic system for task classification that aims to meet these descriptive adequacy and theoretically motivated charges. The taxonomic “listing” of pedagogic task features in the Triadic Componential Framework, illustrated in Figure 1, falls under three broad classificatory categories: features of tasks contributing to their intrinsic cognitive complexity; features of tasks determined by the situational setting, and conditions in which they take place; and learner factors which contribute to the extent of the difficulty faced in attempting to successfully accomplish cognitively complex tasks, in a variety of situational settings. Task conditions concern the participation (e.g., 1-way versus 2-way) and participant (e.g., familiar versus unfamiliar) factors that can be identified in any target-task context of use. Task difficulty concerns the learners’ perceptions of task demands, as these are variably affected by the affective and ability factors listed in Figure 1. Task difficulty will contribute to between learner variation in performing any one (simple or more complex) task, in the same way differences in aptitude for Math will distinguish the speed and success of those solving calculus or geometry problems. Task complexity (the proposed basis of sequencing tasks for learners) contributes to intra-learner variation in performing any two tasks, such as doing simple

4

addition versus calculus, or the various simple and complex tasks studied in the third section of this paper. Task Complexity (Cognitive factors) (Classification criteria: cognitive demands) (Classification procedure: information-theoretic analyses) a) Resource-directing variables making cognitive/conceptual demands, e.g. +/- here and now +/- few elements -/+ spatial reasoning -/+ causal reasoning -/+ intentional reasoning -/+ perspective-taking b) Resource-dispersing variables making performative/procedural demands, e.g. +/- planning time +/- single task +/- task structure +/- few steps +/- independency of steps

Task Condition (Interactive factors) (Classification criteria: interactional demands) (Classification procedure: behavior-descriptive analyses) a) Participation variables making interactional demands, e.g.

Task Difficulty (Learner factors (Classification criteria: ability requirements) (Classification procedure: ability assessment analyses)

+/- open solution +/- one-way flow +/- convergent solution +/- few participants +/- few contributions needed +/- negotiation not needed b) Participant variables making interactant demands, e.g.

h/l working memory h/l reasoning h/l task-switching h/l aptitude h/l field independence

+/- same proficiency +/- same gender +/- familiar +/- shared content knowledge +/- equal status and role

h/l openness to experience h/l control of emotion h/l task motivation h/l processing anxiety

+/- prior knowledge

a) Ability variables and task-relevant resource differentials, e.g.

h/l mind/intention-reading b) Affective variables and task-relevant state-trait differentials, e.g.

h/l willingness to communicate h/l self-efficacy

+/- shared cultural knowledge Figure 1. The Triadic Componential Framework for task classification - categories, criteria, analytic procedures, and design characteristics (from Robinson 2007a).

1.2. Resource-Directing (Conceptual) and Resource-Dispersing (Procedural) Dimensions of Task Complexity In the Triadic Componential Framework an important theoretical distinction is made between resource-directing and resource dispersing dimensions of complexity (Robinson 2003a). The first subcategory distinguishes task characteristics on the basis of the concepts that the task requires to be expressed and understood (e.g., relative time, spatial location, causal relationships, and intentionality). Clearly, conceptualization is more, or less demanding of cognitive resources. This is evidenced, for example, by the staged emergence 5

of conceptual abilities - and their linguistic expression - in childhood (see e.g., Bartsch and Wellman 1995; Cromer 1974; Mandler 2004), and by similar stages in the ability to mark and code them linguistically in the L2 during adult naturalistic second language acquisition (e.g., Becker and Carroll 1997; Dietrich, Klien and Noyau 1995; Perdue 1993). Expending the mental effort (see Robinson 2003a; Wickens 2007) needed to make more demanding cognitive/conceptual distinctions in language should therefore prime learners - and direct their attentional and memory resources - to aspects of the L2 system required to accurately understand and convey them, thereby facilitating selective attention to, and “noticing” of these (Robinson 1995a; Schmidt 2001) and so promoting interlanguage development by speeding up L2 grammaticization in conceptual domains, and where necessary promoting the ‘re-thinking-for-speaking’ that may be needed when mapping conceptualisation to linguistic expression in the L2 (Robinson and Ellis 2008a, 2008b; cf. Slobin 1996; 2003). Grammaticization of the L2 means to mark conceptual distinctions in language follows, in many cases, a similar trajectory in adulthood to the one apparent in child L1 development and recapitulating these sequences of conceptual/linguistic development by presenting L2 tasks in an order of increasing conceptual complexity has been proposed as a ‘natural order’ for scheduling such task demands with likely beneficial effects for learners (Robinson 2003b, 2005a; Robinson and Ellis 2008a, 2008b). Discussing the parallels between child and adult language development in the emergence of prepositions for marking first topological relations of neighborhood, and containment, and later, axis-based projective relations of between, front/backness, in the European Science Foundation (ESF) project data Slobin (1993) comments as follows: “The parallels, though, cannot be attributed to the same underlying factors. In the case of FLA (first language acquisition) one appeals to cognitive development: the projective notions simply are not available to very young children. But in the case of ALA (adult language acquisition) all of the relevant cognitive machinery is in place. Why, then, should learners have difficulty in discovering the necessary prepositions for spatial relations that they already command in the L1? There are at least two possibilities: (1) adult learners retain a scale of conceptual complexity, based on their own cognitive development, and at first search the TL (target language) for the grammatical marking of those notions which represent some primordial core of basicness or simplicity; and/or (2) these most basic notions are also used with relatively greater frequency in the TL…but learners “gate them out” due to their complexity. In this case cognitive factors play a role in both FLA and ALA, but for different reasons: the complex notions are not available to very young children, while they are available but not accessed in early stages of ALA.” (Slobin 1993: 243) With both of the possibilities Slobin describes in mind, then sequencing tasks making conceptual/linguistic demands in the order these are met in child and adult L2 development would be complementary to adult learners’ own initial dispositions to make form-meaning mappings, and also helpful in prompting them to move beyond them and grammaticize L2

6

speech in increasingly target-like ways. Proposed resource-directing variables (see Figure 1) distinguishing task characteristics in terms of these conceptual/linguistic demands include: 1) whether the task requires reference to events happening now, in a mutually shared context (Here-and-Now) versus to events that occurred in the past, elsewhere (There-andThen); 2) reference to few, easily distinguished, versus many similar elements; 3) reference to spatial location, where easily identifiable and mutually known landmarks can be used, versus reference to location without this support; 4) simple information transmission, versus reasoning about causal events and relationships between them; 5) simple information transmission, versus reasoning about other people intentions, beliefs and desires and relationships between them; 6) and whether the task requires the speaker/listener to take just one first-person perspective on an event, or multiple second, and third person perspectives. In contrast to resource-directing variables are those that make increased performative/procedural demands on participants' attentional and memory resources, but do not direct them to any aspect of the linguistic system which can be of communicative value in performing a task. Meeting these demands during pedagogic task performance therefore should facilitate not analysis, and development of new L2 form-concept mappings but rather automatic access to, and control of, an already established interlanguage system (cf. Bialystok 1994). These resource-dispersing variables include those that distinguish task characteristics on the basis of: 1) giving planning time (and so increasing resource availability) versus not giving it; 2) providing background knowledge needed for task performance, versus not giving it; 3) tasks requiring only one thing to be done, versus those requiring two (dual) or many (multiple) things to be done simultaneously; 4) tasks where there is a clear structure available to help on deciding which steps are needed to complete it, versus those without one; 5) tasks where one or few steps are needed to complete it, versus those requiring many steps; 6) and tasks where there is no necessary sequence or 'chain' in which steps are followed, versus those which require participants to follow a strictly chained sequence, in which one step must be performed before another. 1.3. Predictions for the Effects of Task Complexity on Learning and Performance 1.3.1. Effects of Task Complexity on Language Production Most studies of the effects of task demands on speech production have employed general measures of accuracy fluency and complexity, such as percentage of error free C-units, or clauses per C-unit (see Ellis and Barkhuizen 2005; Ortega 2000; Wolfe-Quintero, Inagaki and Kim 1998 for review), enabling comparability of findings across a wide variety of task demands. Following arguments by Givon (1985, 1995; cf. Sato 1988, 1990) that structural complexity tends to accompany functional complexity in discourse, and that demanding, formal communicative tasks and contexts elicit a syntactic mode of production (characterized by greater use of morphology, greater syntactic subordination, and a higher noun to verb ratio) in contrast to a simpler pragmatic mode, the Cognition Hypothesis 7

predicts greater accuracy and complexity using such general measures of production, on complex versus simpler tasks along all resource-directing dimensions of tasks. These general measures, however, will need to be supplemented by specific measures of the accuracy and complexity of production, as these are relevant to particular resource-directing measures making conceptual/linguistic demands. For example, tasks requiring complex spatial reasoning, event construal, and reference to motion, can be expected to lead learners to attempt to use developmentally later acquired lexicalization patterns for describing motion events (Berman and Slobin 1994; Cadierno 2004, 2008; Robinson and Ellis 2008a, 2008b). Similarly tasks requiring increasingly complex reasoning about, and reference to the intentional states of others causing them to perform actions can be expected to involve greater use of psychological and cognitive state terms such as ‘think’, ‘expect’, ‘know’, and the complex syntactic predication use of these terms requires (see Astington and Baird 2005; Lohman and Tomasello 2003). In contrast to these predictions, along resource-dispersing dimensions of tasks which divide but do not direct attention to features of linguistic code, such as taking away planning time, or making dual or multiple simultaneous task demands, then accuracy and complexity of production can be expected to decrease on complex tasks. Skehan’s (1998) Limited Capacity Hypothesis makes the same predictions for the effects of planning time, and other resource-dispersing dimensions, such as removing supporting task structure. Where the Cognition Hypothesis differs from the Limited Capacity Hypothesis is over the claims described above for the beneficial effects on accuracy and complexity of increasing the resource-directing dimensions of tasks. The resource-directing/dispersing distinction is one that Skehan does not make, leading him to claim complex task performance, along any dimensions, degrades accuracy, fluency and complexity simultaneously. The Cognition Hypothesis, importantly, also claims that there are likely to be synergetic effects on speech production when tasks are made complex along both resourcedirecting and resource-dispersing dimensions simultaneously (as is often the case in real world task performance, such as impromptu reasoning about and explanations of the causes of a multi-party social conflict without the benefit of planning time). In such cases the beneficial effects on speech production of increasing complexity along the resourcedirecting dimension are likely to be weakened or negated by increasing complexity along the resource-dispersing dimension, when compared to the same task made simpler along a resource-dispersing dimension (e.g., where planning time is available).

8

1.3.2. Effects of Task Complexity on Interaction and Learning Opportunities The Cognition Hypothesis also connects input and interaction to the cognitive and conceptual demands of tasks that lead to differential amounts of interaction, or uptake of forms made salient in the input to tasks (see Mackey 1999). It predicts that along resourcedirecting dimensions, and in general too along resource-dispersing dimensions, that more interactive complex tasks will result in greater amounts of interaction, and negotiation for meaning. It also claims, following Long (1996), that such negotiation provides a context for attending to problematic forms in the input and output, and additionally that on complex versions of tasks there will be greater attention to, and uptake of forms made salient during provision of reactive Focus on Form techniques such as recasts (see Doughty 2001; Long 2006; Long and Robinson 1998). Alternatively, where proactive Focus on Form is provided, for example in the form of premodified input to the task, then it similarly claims there will be greater use of this on complex, versus simpler task versions. 1.3.3. Effects of Task Complexity on Individual Difference-Task Performance and Learning Interactions Finally, the Cognition Hypothesis acknowledges that learner factors (contributing to perceived difficulty) interact with task factors (contributing to their complexity) in determining the extent of the above predicted effects. When the ability and affective factors drawn on in meeting complex task demands are high in any group of learners, then the effects will be found most clearly, in contrast to learners low in the ability and affective variables implicated in successful complex task performance. An example of this interaction of task difficulty and task complexity with language production is found in the study to be reported below, where only those learners low in output anxiety responded to complex reasoning task demands by producing the predicted increasingly complex speech. Learners high in output anxiety were not induced by task demands to ‘push’ or ‘stretch’ production in this way. Individual differences in task-relevant abilities and affective factors can be expected to increasingly differentiate task-based learning and performance as tasks increase in complexity—an interaction effect well established outside the field of SLA (see e.g., Snow, Kyllonen and Marshalek 1984), and in much need of empirical, task-based SLA research (see Robinson 2002, 2005b, 2007b).

2. Intentional Reasoning Demands, Theory of Mind and Task Production As just described, the Triadic Componential Framework for addressing the claims of the Cognition Hypothesis specifies six dimensions along which task complexity can be increased while simultaneously directing cognitive resources to the available, and specific L2 linguistic means to meet task demands. Of these six dimensions, the intentional reasoning dimension is operationalised as the basis for increasing the complexity of task demands in the present study.

9

Reasoning about, and successfully understanding (intention-reading) the motives, beliefs and thoughts which cause people to perform actions has been a much studied subject in both developmental (Schneider, Schuman-Hengsteler and Sodian 2005) and differential (Baron-Cohen 1995) cognitive psychology, and in theories of the relationship between language and thought in child development (Astingdon and Baird 2005). The ability to represent, conceptualize and reason about psychological, mental states has been called a person’s “theory of mind” (Shatz, Wellman and Silber 1983) which “frames and interprets perceptions of human behavior in a particular way - as perceptions of agents who can act intentionally and who have feelings, desires and beliefs that guide their actions” (Malle 2005: 22-227). This cognitive/conceptual ability develops gradually in early childhood, as is evident in the child’s linguistic development (Bartsch and Wellman 1995). Language used to refer to psychological states emerges in a predictable developmental sequence (Lee and Rescorla 2002), with physiological terms (e.g., sleepy), desire words (e.g., want) and emotion words (e.g., happy) being the earliest and most common psychological state terms to occur in children’s speech. Around the age of three years children begin also to make reference to cognitive states of others using such expressions as “she/he thinks” and “she/he knows”, demonstrating the emergence of the child’s ability to theorize other minds, and reason about the mental states, and intentions guiding others’ actions. This ability to reason about mental states is impaired in children with autism who nonetheless demonstrate average or above average abilities in causal reasoning about physical events (Baron-Cohen 1995). The use of cognitive state terms such as “think” and “know” at around age three also ushers in a spurt in syntactic development, since these verbs take complement clauses “he thinks that he is...etc”, and the development of complex syntactic complementation is thought to be related to the developing ability to conceptualize and reason about other minds (Lohmann and Tomasello 2003). Lee and Rescorla (2002) demonstrated that the use of such psychological, cognitive state terms correlated significantly and positively with the use of complex syntax in child development using measures from Scarborough’s (1990) Index of Productive Syntax (IPSYN). With these issues in mind the present study examines whether tasks which require complex reasoning about the intentions of others will also result in more syntactically complex L2 speech production (as the Cognition Hypothesis claims should happen along such resource-directing dimensions of complexity). Two kinds of measure of complex speech production were used; i) general measures of complex syntax commonly adopted in previous studies of L2 task production, i.e., clauses per C-unit; and ii) specific measures of the use of cognitive and other psychological state terms (as in Lee and Rescorla 2002) appropriate to the particular cognitive/conceptual demands of reasoning about others intentions in performing actions, and also IPSYN measures of Wh-clauses, infinitives and conjoined clauses, as used by Lee and Rescorla (2002).

10

2.1. Task Complexity, Interaction and Uptake The second claim of the Cognition Hypothesis addressed in the present study has been less empirically studied than the claim about the effects of complex tasks on the accuracy, complexity and fluency of speech production. As described above, this is that complex tasks lead to more interaction and uptake of linguistic forms that are relevant to the cognitive/conceptual demands of the task when these are made salient in the input, compared to simpler task versions, and so to more “learning opportunities”. Robinson (2001b) showed that more complex direction-giving map tasks led to significantly more confirmation checks, but not clarification requests, on complex versions, and Hardy and Moore (2004) showed that more complex tasks led to greater quantities of conversational negotiation, but no study to date has examined the effects of task complexity on uptake. Uptake can be operationalised as incorporation of negative feedback in the form of recasts provided on-line, reactively, during task performance (Doughty 2001; Long 1996, 2007) and also as use of premodified input provided proactively in the materials used to support task performance. The present study operationalised uptake in the latter way, using measures that will be described in detail below, and assessed the effects of task complexity on interaction using measures of the number of turns taken on task by participants, as well as the number of confirmation checks and clarification requests task performance generated. 2.2. Task Complexity, Task Difficulty and Individual Differences in Anxiety A third claim of the Cognition Hypothesis, also addressed in the present study, concerns the relationship between the cognitive factors contributing to task complexity, and the learner factors contributing to perceptions of task difficulty. This claim is that more cognitively complex tasks will be perceived by all learners to be more difficult than less complex counterparts, and also that individual differences in ability and affective factors relevant to the cognitive demands of tasks will increasingly differentiate learners’ speech production, and interaction and uptake, as tasks increase in complexity. In the present study all participants were asked to complete a five item task difficulty questionnaire (as in Robinson 2001b, and as described below) following performance on each of the three tasks they took part in, to assess overall perceptions of the difficulty of cognitively complex, and cognitively simpler task versions. In addition, individual differences in input, processing and output anxiety (IPOAS) were assessed prior to performance of all tasks using a questionnaire developed by MacIntyre and Gardner (1994). Six items on this questionnaire assess the extent of learners’ anxiety about hearing and decoding L2 input successfully; six items assess their anxiety about understanding and mentally processing the input in order to respond to it; and six items assess their anxiety about speaking and producing the L2. Each of these three sources of potential anxiety about hearing, processing and producing the L2 were then examined for their influence on speech production, interaction and uptake.

11

2.3 Research Hypotheses In line with the above rationale, and the claims of the Cognition Hypothesis there were six hypotheses for the present study. Hypothesis 1. Using general measures of speech production, along the resource-directing dimension operationalised there will be greater accuracy and complexity, but less fluency on complex relative to simpler versions. Hypothesis 2. Using specific, theoretically motivated measures of speech production appropriate to the conceptual/communicative demands of tasks, along the resource-directing dimension operationalised there will be greater lexical and syntactic complexity on complex relative to simpler versions. Hypothesis 3. The general and specific measures of syntactic complexity will be significantly and positively correlated with each other. Hypothesis 4. There will be more interaction, and uptake, on complex tasks, compared to simpler versions. Hypothesis 5. Complex task versions will be rated more difficult, stressful, and confidence in participants’ ability to successfully perform them will be rated lower than on simpler versions. Hypothesis 6. Individual differences in input, processing and output anxiety will differentiate speech production, interaction and uptake on complex versions of tasks more than on simpler versions.

3. The Present Study 3.1. Participants and Procedure Participants were 42 Japanese L1 university students, aged between 20 and 23 years who had each had seven years of high school level education in English, and from three to four years university level education in English. Participants were randomly assigned to the roles of speaker/storyteller and listener/sequencer. Each of the 21 dyads performed three narrative tasks at simple, medium , and complex levels of intentional reasoning demand. Order of task performance was counterbalanced, with seven dyads performing tasks in the sequence simple-medium-complex, seven in the sequence medium-complex-simple, and seven in the sequence complex-simple-medium. 3.1.1. Tasks Used The dimension of Task Complexity addressed was increases in intentional reasoning, as it was required in order to perform narrative tasks at three levels of conceptual/communicative complexity making resource-directing demands on L2 oral task performance. In each of the 21 dyads one participant (the speaker/storyteller) decided on the correct sequence for pictures which had to be sequenced to complete a story, and then narrated this story to a 12

participant (the listener/sequencer) who had to sequence the pictures in the order the narration he/she heard described. These pictures were those at the first (simple), fifth (mid) and ninth (most complex) level of complexity in the Picture Arrangement (PA) subtest of the Wechsler Adult Intelligence Scale-Revised (WAIS-R), Japanese version (Shinagawa, Kobayashi, Fujita and Maekawa 1990). The PA subtest is a performance subtest of the WAIS-R, drawing on fluid intelligence (G. - see e.g., Carroll 1993; Cattell 1971) particularly as it is drawn on when interpreting, understanding and responding to novel situations involving others. The PA subtest is “primarily a test of the ability to plan, interpret, and accurately anticipate social events...each of the items requires a person to respond to some practical interpersonal interaction...Persons who score high on Picture Arrangement are usually sophisticated, have a high level of social intelligence, and demonstrate an ability to quickly anticipate the consequences of initial acts...Low scorers may have...difficulty in interpersonal relationships, and poor rapport” (Groth-Marnat 2003: 174-176). In the simplest version of the PA subtest, a single character must be understood to have the intention to build a house, and the sequence of narrative events (building the house) follows from this simple cause. In the more complex versions the main characters’ intentions to perform actions are motivated by their perceptions of other people’s thoughts and beliefs, and their actions follow in response to them. There is only one correct sequence for each set of pictures. Identifying this sequence consequently requires successfully inferring and attributing those thoughts, intentions and psychological states to a main character appearing in each picture which cause them to perform a series of actions, with respect to other characters, in a particular picture-sequence order. On complex versions of the PA task, conceptualizing, framing and then communicating this sequence in narratives should entail complex verbalizations of events and the psychological states of characters which cause them, such as , e.g., he does X because he thought person Y felt, believed, wanted, desired, Z, but then he realizes ...etc. Narratives based on picture strip sequences, and the inferential decision-making thought, reasoning and subsequent discourse verbalization processes they entail have been used in crosslinguistic-developmental research into narrative ability (e.g., Berman and Slobin 1994; Cadierno 2004); in child-developmental research into L1 acquisition and cognitive maturation (e.g., Bamberg 1997; Baron-Cohen 1995; Nelson 1996); and in L2 elicitation of effects of task complexity on L2 production (Gilabert 2007; Robinson 1995b) and were thus judged to be a suitable elicitation procedure for the research questions of interest. Additionally, this elicitation measure is extensively available (as a WAIS-R subtest), and so replicable, and the results of performance on it as a psychological test have been thoroughly researched and reported for test-retest, population, and other criteria for establishing task-construct validity, and reliably normed with regard to other measures of psychological function (see e.g., Groth-Marnat 2003; Kline 2000; Shinagawa et al. 1990).

13

In each of the 21 dyads both participants were given the pictures in a random order, though neither could see the other person’s pictures. One (the speaker/story teller) was instructed to look at them (for one minute) and decide on the correct order, and narrate the story it illustrated to a partner who had to sequence the pictures in the order his/her partner described. The listener/sequencer was told that they could ask the speaker/storyteller questions if they needed to. In terms of task conditions and participation factors, as classified in the Triadic Componential Framework (Robinson, 2001a, 2005a, 2007a; Robinson and Gilabert 2007), this was therefore a one-way, closed task. In terms of resource-dispersing dimensions of task complexity, each narrative task was a - planning time, and a + dual task (i.e., speaker/storytellers had to both think of the correct sequence, and tell the story this sequence illustrated). Only resource-directing demands on reasoning about characters’ intentions, and how they cause them to perform actions with respect to others was gradually increased on each task version making this the only factor differentiating task complexity. 3.1.2. Premodified Task-Relevant Input Task-relevant input was provided for the storyteller in the form of six phrases written in English (with a Japanese translation) below each of the three randomly ordered sets of pictures. Three phrases were of the construction type, “verb-present continuous, object” (e.g., is carrying a plank), and three were of the construction type “subject (proform, he), verb-3rd person-s morpheme, adverb” (e.g., he walks tiredly). The six phrases, conforming to these two constructional types, were lexicalized in ways relevant to the events described by each of the three sets of pictures. Each constructional type was given different token lexicalization. That is, in the nine “verb-ing, object” type constructions provided (three for each of the three sets of pictures) and the nine “verb-3rd person-s, adverb” type constructions (three for each of the three sets of pictures) no tokens of lexical words (e.g., verbs, nouns and adjectives, such as carrying, plank, tiredly) occurred in more than one of the task-relevant input phrases. Storytellers were told they could use these phrases if they thought they were useful, though they need not use them, and could use their own words to tell each story. 3.1.3. Post-Task Difficulty Questionnaire After completing each narrative task both the speaker/storyteller and the listener/sequencer were asked to complete a short questionnaire (as in Robinson 2001b) in which they rated on a 9 point scale; whether they though the task was difficult (1 not difficult, 9 very difficult); whether they felt stress performing the task (1 no stress, 9 a lot of stress); whether they felt confident they were able to do the task well (1 not confident, 9 very confident); whether they thought the task was interesting (1 not interesting, 9 very interesting); and whether they wanted to do more tasks like this (1 do not want to, 9 would like to).

14

3.2 Measures Taken 3.2.1 General Production Measures - Speaker/Storyteller To address Hypothesis 1 general measures of speech production, by the speaker/storyteller, adopted in a number of previous studies (e.g., Robinson 2001b; Skehan and Foster 2001) were used. These included measures of lexical complexity, type-token ratio (TTR), and of syntactic complexity, clauses per C-unit (CPC), and a measure of the complexity of turns taken, words per turn (WPT). Accuracy was measured using percentage of error free C-units (%EFC) and fluency by measures of syllables per second (SPS) and words per C-unit (WPC). Means and standard deviations for these measures of production on the three reasoning tasks are given in Table 1. Three of these measures, TTR, %EFC, and SPS were normally distributed, with no excessive kurtosis or skewness. To address the research Hypotheses for these units of analysis repeated measure ANOVAs, and subsequent paired ttest comparisons were used. Three other measures, CPC, WPT and WPC were not normally distributed and nonparametric Friedman repeated measure tests were followed by pairwise Wilcoxon Signed Ranks tests for these data. 3.2.2. Specific Production Measures - Speaker/Storyteller To address Hypothesis 2 specific measures of production by the speaker/storyteller, motivated by the conceptual demands of the reasoning tasks and their linguistic demands were used. These were the use of psychological state terms, as adopted in Lee and Rescorla (2002) and discussed above. A total measure of all psychological state terms used (PSY) and of the separate use of physiological (PHY), emotion (EMO), desire (DES) and cognitive state terms (COG) were used. Additionally ratio measures of the number of total psychological state terms per turn (PSYPT) and of cognitive state terms per turn (COGPT) were used. Only the use of desire terms (DES) was normally distributed, permitting the use of repeated measure ANOVA for this data. For all other data in Table 2, as described above, nonparametric Friedman and Wilcoxon Signed Ranks tests were used in the analyses. In addition to the use of psychological state terms, as in Lee and Rescorla (2002) three measures of complex syntax from the index of productive syntax (IPSYN, Scarborough 1990) were used, to examine whether, as Lee and Rescorla (2002) found, use of cognitive state terms correlated strongly and positively with these indices of complex speech, as Hypothesis 3 claims should be the case. The three measures were of the use of infinitival phrases (e.g., he likes to eat ice cream), conjoined clauses (e.g., close the door so he can’t get out), and Wh-clauses (e.g., this is where you go). Due to its nonnormal distribution nonparametric Friedman and Wilcoxon Signed Ranks analyses were also performed on the conjoined clause data, and parametric ANOVAs on the normally distributed infinitival phrase and Wh-clause data (see Table 3).

15

3.2.3. Measures of Interaction - Listener/Sequencer and Uptake - Speaker/Storyteller To assess the extent to which increasing the complexity of the intentional reasoning demands of the narrative tasks affected the extent of the listener/sequencer’s need to interact and negotiate meaning, by asking for clarification and making confirmation checks, the number of turns taken (Turns), clarification requests (CR) and confirmation checks (CC) were counted for listener/sequencers on each task version. To assess the extent to which the speaker/storyteller used the premodified input, in the form of six phrases, of two types described above when performing each task, four measures were used. Uptake exact (UE) was considered to be any use of an exact, unaltered phrase provided in the input, so with respect to the examples given above, both he is carrying a plank, and he walks tiredly, if used in that exact form by the speaker were considered exact uptake. If the speaker used the phrase but omitted elements (e.g., is carrying a plank, carrying a plank, a plank, carrying, plank) or substituted or added to them (e.g., is carrying a big plank, was carrying a plank) or used individual lexical words in the form given in the phrases alone (e.g., carrying, plank, walks, tiredly) (not individual grammatical words such as he, is, a, etc.) it was considered partial uptake (UP). Additionally ratio measures of the amount of uptake exact per turn (UEPT) and uptake partial per turn (UPPT) were calculated. Turns, and all measures of uptake were normally distributed, and analysed across task versions via repeated measure ANOVAs. Only the CR and CC data were not normally distributed, and so, as described above, for analyses of these measures nonparametric Friedman and Wilcoxon Signed Ranks tests were used (see Table 5). Interrater agreement between two raters of all measures described above was at over 90%.

4. Results Hypothesis 1: General Measures of Speaker/Storyteller Production Hypothesis 1 is not confirmed. As Table 1 shows, there are no significant differences between CPC (df, 2 , 2 = 2.16, p = .29), WPC (df, 2, 2 = 2.66, p = .26), EFC (df, 2/20, F = 1.48, p = .24), or SPS (df, 2/20, F = 1.72, p = .3) in the speaker/storytellers’ speech across task versions. The repeated measure ANOVA for TTR (df, 2/20, F = 7.57, p < .01) is significant, but in the reverse direction to the hypothesis, with TTR on the most complex version significantly lower than the simple (t = -3.32, p < .01) and the medium reasoning tasks (t = -2.95, p < .01). Only the complexity of turn length, WPT is significantly different in the predicted direction (df, 2, 2 = 9.32, p < .01), and this is attributable to a significant difference between the simple and medium tasks (Wilcoxon z = -2.72, p < .01).

16

Table 1. General measures of speaker/storyteller production on tasks differing in intentional reasoning demands TTR M/SD Simple .58/.1 Mid .57/.12 Complex .49/.14 p