Audio Description as a Pedagogical Tool in the Foreign Language Classroom: An Analysis of Student Perceptions of Difficulty, Usefulness and Learning Progress

Audio description (AD) is an intersemiotic mode of audiovisual translation where images are translated into words to facilitate access to visual content for visually-impaired audiences. Over the last decade, it has gained prominence in foreign language research, as its formal particularities and condition as a communication-oriented and process-based activity present a fertile ground for designing innovative classroom tasks. This research was conducted with students of Spanish at the University of Manchester, and it invest igates participants’ perceptions about a classroom AD project. The exploratory analysis of quantitative and qualitative data from participants’ responses to end - of-project questionnaires leads to the formulation of a ‘triple - connection hypothesis’: a direc t relationship between perceptions of difficulty of the main challenges of AD, perceptions of usefulness of AD for developing the skills required to overcome those challenges, and perceptions of own learning progress thanks to AD. The findings show that AD is perceived by students as especially useful for developing integrated skills, communicative skills and lexical skills, to a greater or lesser extent depending on the difficulty perceived.


Audiovisual Translation and Audio Description in Foreign Language Teaching
Over the last two decades, the revitalization of translation as the "fifth skill" (Carreres et al., 2017) and as a form of mediation (Cook, 2010;Council of Europe, 2001, 2020Laviosa, 2014) has sparked interest in audiovisual translation (AVT) for foreign language teaching (FLT) (Lertola, 2019a;Talaván, 2013). Initially, AVT was introduced as a passive resource, 1 for example, using subtitles as support for aural comprehension (Gambier et al., 2014). The switch to a learner-centred pedagogical paradigm following the advent of communicative approaches has reconfigured AVT as an active resource, matching the concept of the active learner (Nunan, 2004). This has been facilitated by the increased availability of pedagogically adequate technological tools and inspired by external factors such as the fansub and fandub movements. 2 Students, like fansubbers and fandubbers, are no longer just passive consumers of subtitled or dubbed audiovisual products but active producers of the same. Pedagogically-oriented explorations of active AVT tasks have mainly focused on subtitling (among others, Ávila-Cabrera, 2021; Lertola, 2019b;Talaván, 2013;Talaván & Rodríguez Arancón, 2014), but the use of revoicing 3 has also flourished, first as dubbing activities (Ávila-Cabrera, 2022;Danan, 2010;Talaván & Costal, 2017) and, subsequently, as voice over (Talaván & Rodríguez-Arancón, 2018) or audio description-based tasks.
Audio description (AD) is a mode of accessible AVT, primarily aimed at blind or visually-impaired audiences, that facilitates access to visual content by translating visual elements into verbal commentary (Walczak & Fryer, 2017). Increased awareness of the social mediation role of AD has contributed to its incorporation into media accessibility regulations across countries (Orero, 2016), further leading to its popularisation as a classroom tool. Contrary to text-to-text translation or other forms of AVT, AD is intersemiotic: the transfer occurs from a visual system to a linguistic one, thus activating different mental processes in the translator or audio describer (Holsanova, 2016). Furthermore, describers are not constricted by pre-selected content -they select elements from the source text themselves-nor by pre-established lexical or syntactic configurations, as the source text is not word-based. The "cross-modal" nature (Ibáñez Moreno & Vermeulen, 2017a) and cognitive singularity of AD enable innovative pedagogical explorations.

Student Perceptions in Pedagogical Research
Research on the pedagogical potential of AD (or AVT in general) usually explores the impact of a didactic intervention on the development of specific skills. Emphasis is placed on measurable linguistic data from student productions (see among others, Calduch and Talaván (2018), Talaván and Costal (2017), Talaván and Rodríguez-Arancón (2014), Vermeulen and Ibáñez Moreno (forthcoming)), with extra-linguistic data from questionnaires and interviews largely supporting linguistic findings. Experimental research relying on student perceptions as evidence of learning is not the norm, 4 perhaps given the difficulty of obtaining scientifically reliable and statistically measurable data. However, exclusive dependence on scientifically provable and measurable knowledge is rarely possible in educational research (Cohen et al., 2007), so data triangulation tends to be applied (p. 468). Consequently, despite its known limitations, perception data often plays a significant role as supporting evidence of linguistic findings (pp. 461, 469).
Analysing student perceptions of AVT within FLT contexts has often helped to paint a more comprehensive picture of its pedagogical value and to explore the metacognitive processes involved in language use. Findings range from the identification of links between cultural factors and response rates (Ibáñez Moreno & Vermeulen, 2015a, 2015b to useful suggestions for improving the design of tasks and projects (Ibáñez Moreno & Vermeulen, 2013, 2016b. Furthermore, questionnaires requiring students to engage in metalinguistic and metacognitive reflection also provide additional learning opportunities and promote learning autonomy beyond the actual task, given the highly complex cognitive processes that respondents engage in when answering (Ibáñez Moreno & Vermeulen, 2017b, 2021.

Aims
This article stems from an AD Project carried out with students of Spanish at the University of Manchester (2018)(2019)(2020). It presents the main observations from an exploratory analysis of data from end-of-project questionnaires (EPQs) and focuses on three variables selected for their relevance in assessing the pedagogical value of AD as a tool in the foreign language (FL) classroom (see section 4.2). The aims are: 1. To analyse students' perceptions of the difficulty (variable 1) and usefulness (variable 2) of AD tasks as a language learning tool; 2. To analyse students' perceptions of their own learning progress throughout the experiment (variable 3); 3. To analyse the interconnections between variables 1 to 3.
A first approach to quantitative data from closed questions (CQs) and qualitative data from openended questions (OEQs) suggested a direct relationship between the three variables which we have called the 'triple-connection hypothesis'. The present analysis covers selected data to illustrate such connections and to support this hypothesis. Brief explanations of the data retrieval and selection processes are provided as needed (see section 5).

A Methodological Framework for the Use of AD in FLT
Audio description is a real-world communicative activity requiring both linguistic and non-linguistic skills. This makes AD particularly suitable for designing tasks following the principles of communicative language teaching (CLT) (Canale & Swaine, 1980;Richards, 2006) and the integrated skills approach (ISA) (Brown, 2001;Oxford, 2001).
In CLT, communication is placed at the centre of classroom tasks and students at the centre of the learning process. CLT understands students as active learners who learn by engaging in situationspecific activities with a clearly-established communicative goal. Such a role mirrors that of a professional audio describer, who aims to bridge the communication gap between visual content and visually-impaired audiences. Accomplishing this entails engaging in social mediation, 5 since audio describers act as "social intermediaries" (Ibáñez Moreno & Vermeulen, 2017a, p. 56), mediating between the visual world, the social connotations of its elements, and the audience. In an AD classroom task, students become audio describers, thus approaching communication not just as learners or interlocutors, but as linguistic and social mediators.
The integrated-skills approach (ISA) revolves around the idea that interlocutors in everyday communication constantly exchange roles, switching between sender and receiver, and use multiple skills, switching from speaking to writing and from reading to listening (Su, 2007). The ISA is linked to the notion of "whole language" (Abdrabo, 2014, p. 8), as it understands that linguistic functions and skills must work together for successful communication and that FL activities should replicate such conditions. Creating an AD script (ADS) requires combining written lexical and grammatical skills to ensure accuracy and idiomaticity as well as decision making (for selecting the relevant visual content). Reading skills are also involved, as students need to revise their own script. Additionally, they may be asked to compare it to other classmates' or to the original, to reflect on content and linguistic decisions. If the ADS is narrated, speaking skills are required, too, alongside other non-linguistic skills such as summarising and paraphrasing (to fit the information into gaps in the dialogue). In sum, an AD task approaches language use from many angles, fostering the interconnected development of multiple skills.

Context
This experiment was carried out with second-year students of Spanish at the University of Manchester in 2018-2019 (Phase I) and 2019-2020 (Phase II) and was based on previous work by Vermeulen and Ibáñez Moreno (forthcoming). It was designed as a classroom AD project and embedded in a compulsory module (Spanish Language 5) taught at a B2 level. 6 The module was divided into three components: (1) Grammar Skills, (2) Translation & Writing Skills, and (3) Oral Skills.
Components (1) and (2) were grouped in a double weekly session; component (3) had its own 50minute time slot. Project sessions took place during slots for components (2) and (3) and were delivered by the module teachers, but tasks were not part of the summative assessment. 7

Aims
The conception of the AD project was twofold: as a research experiment and as a learning activity. On the one hand, the experimental aim was to investigate the potential benefits of intersemiotic translation (AD) over interlinguistic translation (text to text) for developing idiomaticity and metalinguistic awareness. This followed Vermeulen and Ibáñez Moreno's (forthcoming) hypothesis that the absence of linguistic content in students' first language (L1) in an AD task would reduce the risk of negative transfer. 8 Their hypothesis was based on Cohen and Brooks-Carson's (2001) observations that direct writing in the FL enhanced idiomaticity more than translation, thanks to such 6 Common European Framework of Reference for Languages (Council of Europe, 2001, 2020. 7 Changes to evaluation were subjected to a long, complex process, and incorporating the AD Project would have required reweighing all assessment components. Due to this and to the fact that the effectiveness of the AD Project had not been tested when changes should have been proposed, project tasks were excluded from the summative assessment. 8 Negative transfer refers to the interference of the L1 in the learning and production of the target language (Zhao, 2019).
absence of explicit L1 content. On the other hand, the pedagogical aim was to promote the development of integrated and communicative skills through the creation and narration of an AD script.

Participants
Students enrolled in the Spanish Language 5 module who combined Spanish with either another FL or a Humanities or Science discipline were eligible to participate. 9 All participants had studied Spanish for 5 to 10 years and their ages ranged between 18 and 20. All were proficient speakers of English, which was the L1 for the vast majority (98% in Phase I; 93% in Phase II). The majority of participants were familiar with or at least aware of the existence of AD, but did not have experience watching audio described programmes and had never attempted audio describing before.

Procedures
The Project ran over 5 weeks and included an introductory workshop, two writing sessions (one for each task), and a feedback and reflection session between tasks (see Table 1). In each writing session, participants completed either an AD task (ADT) directly from the images, or a translation task (TRT) from English into Spanish. Students were divided into an Experimental Group (EG) and a Control Group (CG). Since the aim was to test whether AD promoted idiomaticity more than interlinguistic translation, the first task (pre-test) for the EG was AD, while for the CG it was translation. Once the pre-test was completed, a feedback and reflection session took place, where participants were exposed to the original Spanish AD and to certain language structures. 10 Two weeks later, the second task (post-test) was completed to test whether exposure to and familiarisation with the structures had an impact upon idiomaticity in either type of task. For the post-test, EG and CG did the opposite task, to enable comparison across task types within each experiment group, too. In Phase I, students worked collaboratively; in Phase II, they worked individually. 11 The end-of-project questionnaire (EPQ) was completed in the two weeks following the second task (post-test). Table 1. Source: Author's own elaboration.

Project Materials
For the tasks (ADT and TRT), participants had to create or translate an ADS for two consecutive clips from Woody Allen's Match Point (2005). 12 The clips were selected because of the linguistic similarity of the original Spanish ADS for both scenes (which enabled the use of the first as pre-test and the second as post-test), and their suitability for first-time, non-professional audio describers (limited dialogue, limited number of characters, self-contained narrative). For ADTs, the source texts were the clips; for the TRTs, they were English translations of the Spanish ADSs done by the researcher and proofread by L1 English-speaking colleagues. 13

Project Questionnaires
The experiment included pre-project questionnaires (PPQs) and end-of-project questionnaires (EPQs). 14 PPQs collected information about students' linguistic background and familiarity with AD (see section 4.3). EPQs collected participants' perceptions regarding the AD task, the project, and their learning progress throughout. After Phase I, a revision process 15 was carried out to identify weaknesses in questionnaire and project design, which resulted in modifications in Phase II. The main ones affecting the EPQ were: 1. Adding a data protection section; 2. Simplifying the overall structure; clarifying instructions, question numbering and removing redundant, irrelevant, and unnumbered questions; 3. Adjusting question texts and items; 4. Turning checkbox grids into rating scales (RSs) to facilitate isolation of unique variables for analysis; 5. Turning 5-point rating scales into 4-point ones with an 'I don't know' option, to minimise the middle-ground tendency effect. 16 Modifications (3), (4) and (5) affected some of the questions considered for the analysis of variables 1 to 3, so data re-grouping was necessary to enable comparison between Phase I and Phase II.
Modifications and data re-grouping processes are explained in the analysis section when applicable. A second revision process after Phase II confirmed that the modifications had facilitated both the collection of data and its linking with research aims. This is illustrated in the analysis section when relevant.

Data Analysis and Results
This exploratory analysis of EPQs focuses on the variables identified in section 2: (1) perception of difficulty of the ADT, (2) perception of usefulness of the ADT, and (3)  2. Qualitative analysis (OEQs) for variables 1 to 3.
In Phase I, 22% of eligible students completed the EPQ, while 18.6% did so in Phase II (see Table 2).
Despite the low response rates, perceptions expressed can be considered representative, as in both phases over 90% of respondents had completed at least one task and attended all sessions. 17 Given the difference in response rates, only percentages are considered in the analysis, although absolute numbers are provided for reference. The term 'respondents' and not 'students' or 'participants' is used in this section, since only data from those who completed the EPQs is considered.
Due to space limitations, when relevant questions (those related to the three variables) had multiple response items, the analysis focuses on the ones most illustrative of the 'triple connection hypothesis' (see section 2). To present the data more logically, the variables are analysed in an order different from that followed in EPQs. Responses presenting reliability issues were removed from the sample and are not considered in this analysis. 18 The small minority of respondents who had only completed the TRT had been exposed to AD either in the introductory or the feedback session, so they were still able to compare tasks to an acceptable extent.
as a 4-point scale with the same response items, Q9: Rate the following aspects of the AUDIO DESCRIPTION task (from images into words) in terms of difficulty, where 1 is "not particularly challenging" and 4 is "very challenging". If you didn't do the AD task, select "I don't know". The following re-grouping was necessary to compare EQ1 and EPQ2 responses: 1. EPQ1: total frequency of scores for each item; 2. EPQ2: total frequency of scores for ratings expressing difficulty, i.e. '3' ('challenging') and '4' ('very challenging') for each item. Figure 1 compares the grouped scores for variable 1 in EPQ1 and EPQ2. Results show a higher perception of difficulty among EPQ2 respondents across all items except for (a) 'content selection', which is considered challenging by a higher percentage of EPQ1 respondents (61.9% vs 53.8%). Item (c) 'idiomaticity' was largely perceived as the most challenging in both EPQ1 and EPQ2 (66.6% and 92.3%, respectively). Item (b) 'word choice/combination' is the only other item perceived as difficult by more than 50% of EPQ1 respondents (52.3%). It is also commonly perceived as difficult by EPQ2 respondents (84.6%). The remaining items, (d) 'condensing/summarising' and (e) 'narrating', follow the usual trend, with a larger percentage of EPQ2 respondents considering them difficult (84.6% and 69.2% respectively). However, the difference in frequency of selection in EPQ1 was more notable, with just 38.1% for (d) and much lower at 23.8% for (e).

Variable 2: Perception of Usefulness of the ADT
Variable 2 refers to perception of usefulness of the ADT for language learning. In EPQ1, the responses were gathered through a multiple-choice question, Q10: Overall, how useful do you think AUDIO DESCRIPTION is as a language learning activity? Please, tick the option that better matches your views. The possible answers considered AD in general (for example, (a) Not particularly useful) and in comparison to translation (for example, (b) Useful, but less than regular translation) (see Fig. 2). The post-Phase I revision process revealed that this format did not allow for isolating perceptions of usefulness in relation to specific learning outcomes. Thus, in EPQ2 it was re-designed as a 4-point RS, Q10: Based on your experience during this project, how much do you agree with the usefulness of AUDIO DESCRIPTION as a language learning activity?; 1 being "I do not agree at all", and 4 being "I completely agree". 19 Q10 required respondents to rate the level of usefulness for improving the following aspects: (a) General vocabulary, (b) Specific vocabulary, (c) General grammar/syntax, (d)

Specific grammar/syntax, (e) Communicative skills, (f) Intercultural skills and (g) Transferable skills.
Modifications made data re-grouping necessary. Because analysis by learning outcome was not possible in EPQ1, three 'general levels of usefulness' were established to compare EPQ1 with EPQ2 data. The three main labels are: 'limited usefulness', 'average usefulness' and 'high usefulness'. Table  3 shows how EPQ1 and EPQ2 response items are grouped under the three main labels. No EPQ2 scale values can be assigned to the 'average usefulness' label, as they belong to either the 'limited' side (values '1' and '2') or the 'high' side (values '3' and '4'). Source: Author's own elaboration.
In terms of general usefulness, a majority of EPQ1 respondents (52.3%) expressed a perception of 'Limited usefulness' of AD (see Fig. 2). Adding the 23.85% for whom AD was as useful as TR ('Average usefulness'), we ascertained that a total of 76.15% did not consider AD particularly useful for language learning.
Conversely, a majority of EPQ2 respondents expressed a perception of 'High usefulness' of AD, as evidenced by the larger percentages of 'useful' or 'very useful' ratings for all selected items (see Fig.  3). This general higher perception of usefulness (variable 2) among EPQ2 respondents matches the trend observed for perception of difficulty (variable 1).
In EPQ2, response items linked usefulness (variable 2) to specific skills. For the analysis, only the response items more narrowly connected with variable 1 (difficulty of AD) were selected: (a) Lexical skills, (b) Communicative skills, and (c) Transferable skills. Figure 3 shows that for EPQ2 respondents, AD was especially helpful for improving 'transferable skills' (92.3%). Scores for the other two selected items were similarly high: 84.7% considered AD useful for improving 'communicative skills' and 76.9% did so for 'lexical skills'. This more nuanced view of perceptions of usefulness reveals that AD was considered especially beneficial for developing the skills needed to overcome what had been identified as its main challenges, namely, 'sounding natural' and 'word-choice' (lexical skills), 'summarising' (transferable skills), 'narrating' (communicative skills) and 'decision making' (transferable and communicative skills).

Variable 3: Perception of Own Learning Progress
Variable 3 refers to participants' perception of their own learning progress in relation to different skills. This information was collected in EPQ1 through a checkbox grid, Q4: Which aspects of your learning would you say have improved throughout the AD project? You may click more than one. Response items were: (a) Lexical skills, (b) Grammar skills, (c) Syntactic skills, (d) Oral skills, (e) Extralinguistic skills, and (f) Other. During the revision process, it was noted that rethinking the selection of skills was necessary to better link responses to research and pedagogical aims. In EPQ2, this question became a 4-point RS with an 'I don't know' option, Q5: In terms of your own learning throughout the AD project, how much do you agree with the following statements?; 1 being "I do not agree at all" and 4 being "I completely agree". It asked to rate improvement of (a) Lexical skills, (b) Grammatical/Syntactic skills, (c) Oral skills, (d) Transferable skills, (e) Intercultural skills, (f) Communicative skills and (g) Awareness on accessibility issues.
To analyse variable 3, items directly connected with the trends observed for variables 1 and 2 were selected: (a) Lexical skills, (b) Oral skills, (c) Transferable skills ('Extra-linguistic skills' in EPQ1), (d) Communicative skills and (e) Accessibility awareness. Item (e) was included because developing accessibility awareness is linked to identifying the aims of AD and thus its challenges (variable 1) as well as strategies or skills to overcome them (variable 2). Furthermore, 'accessibility awareness' is linked to the development of social mediation skills and other "cross competences", 20 such as audiovisual literacy, which are of interest for the pedagogical aims. To compare data, response grouping followed the same criteria as for variable 1: 1. EPQ1: total frequency of scores for each item 2. EPQ2: total frequency of scores for ratings expressing perception of learning progress, i.e. those expressing agreement: ratings '3' ('agree') and '4' ('totally agree') for each item. Figure 4 compares the grouped data for variable 3 in EPQ1 and EPQ2. The three items selected which are applicable to EPQ1, (a), (b), and (c) were the only ones in Q4 with selection rates above 50%. 'Transferable skills' was selected by the largest percentage of respondents (61.9%), followed by 'lexical skills' (57.1%) and 'oral skills' (52.3%). This means that perception of learning among EPQ1 participants was higher for those skills linked to the ones for which AD was perceived as useful, which in turn are those needed for overcoming the aspects perceived as particularly challenging: 'decision making', 'idiomaticity' and 'word-choice'.
In EPQ2, perception of learning was higher across all skills, considerably so for the selected items: 84.6% of respondents reported improvement in 'transferable skills' and 'oral skills', and 76.9% for 'lexical skills' and 'communicative skills'. These higher selection rates can be linked to the higher perception of usefulness of AD for improving these same skills among EPQ2 respondents as well as to their higher perception of difficulty of the associated AD aspects. Lastly, 92.3% reported having improved their accessibility awareness.

Data for Variable 3: Perception of Learning Progress (EPQ1 vs EPQ2)
Source: Author's own elaboration.

Qualitative Data Analysis
EPQs included open-ended questions (OEQs) alongside the closed questions (CQs), either to allow respondents to expand on answers to certain CQs ('Expansion OEQs') or to require them to provide general feedback ('Final OEQs') (see Table 4). For the qualitative analysis, references to variables 1 to 3 (perception of difficulty, usefulness and learning) across responses to relevant OEQs are considered to see if quantitative findings are supported. - Source: Author's own elaboration.

Variable 1: Perception of Difficulty of the ADT
In EPQ1, only one reference to variable 1 was found. This seems in line with the low perception of difficulty revealed by the quantitative data. The respondent mentioned word choice, highlighting that "even simple vocabulary like parts of furniture was hard to remember", which is also consistent with the quantitative findings.
In contrast, references to the challenging nature of AD were frequent in EPQ2. The comments mainly addressed the issue of condensing language and narration-related aspects ("enunciating clearly" or "at pace"; "including intonation and expression in the voice when narrating"). Perceptions of lexical challenges were less frequent, often mentioning that "more … help with key vocabulary" would have been appreciated. One respondent further expressed that "the use of translation, communication and pronunciation together was challenging but very useful". This seems to imply a direct relationship between difficulty and usefulness, besides pointing at integrated skills development.

Variable 2: Perception of Usefulness of the ADT
In EPQ1, references to the usefulness of the ADT for skills development mainly underscore its potential to promote transferable and integrated skills ("group work"; "using translation and oral skills at once"), and "discovering a practical application of language learning", that is, improving linguistic skills in a transferable way. However, perceptions of limited usefulness of the ADT were also shared. This was attributed mainly to the lack of individual feedback or formal assessment, its timeconsuming nature, and the insufficient overt connection to course lexical or thematic contents. Others complained of the reduced explicit focus on grammatical and syntactic aspects, and requested more interlinguistic translation (which was part of the formal assessment). Nevertheless, some respondents actually advocated including more AD tasks in the curriculum. These mixed perceptions are in line with the quantitative analysis in relation to usefulness, where AD had only been seen as potentially beneficial for lexical learning and transferable skills development, and only by some respondents.
In EPQ2, no references to 'limited usefulness' were recorded. Instead, the comments were mainly enthusiastic. One respondent maintained that AD was "more advantageous [than other tasks] for language learning, particularly regarding vocabulary learning and grammatical structures", and most tended to link the idea of authentic language use to communicative skills improvement. The requirement of "having to combine a lot of skills together" in the ADT was emphasised, which is worth noting since 'integrated skills' had not been explicitly mentioned in CQs. Lastly, the chance to develop 'transferable skills' such as critical or creative thinking and summarising was also highlighted. Thus, EPQ2 qualitative data also support the quantitative findings, showing a widespread perception of usefulness among respondents.

Variable 3: Perception of Own Learning Progress
In EPQ1, contributions to OEQs related to learning progress were limited to vocabulary learning and transferable skills development, specifically decision making, summarising, and learning about AD. One response considered it an "interesting way to improve language skills", suggesting perception of integrated-skills development, while another highlighted the opportunity it provided to increase oral practice. However, instances of negative responses were also found, again linked to the lack of individual feedback or formal assessment. Others would have appreciated more guidance during the writing phase, and some considered the group work arrangement a setback. In sum, EPQ1 qualitative data reveal a perception of limited learning, restricted to some responses, as well as a perception of limited usefulness and difficulty. Exceptions to the above are lexical and transferable skills, which connect with the only AD aspects that were considered challenging: word choice and condensing language.
In EPQ2, no references to limited learning were found. Across OEQs, comments highlighted oral skills improvement (including fluency, accuracy, intonation, and the chance for self-assessment), communicative skills including awareness of the target audience ("I felt I was really building tension for the viewer") and everyday communicative needs, as well as lexical skills, with a focus on the applied nature of the task and the chance to learn "how to use [words] concisely to convey meaning", which suggests development of idiomaticity. Others valued how AD had revealed "the different ways in which [they] could think about language", and how it helped them "to understand the nuances between [English and Spanish]". This suggests that AD was perceived as valuable for developing metalinguistic awareness, another concept not explicitly mentioned in CQs. Transferable skills development was also reported, with references to accessibility awareness, decision making and critical thinking. Again, the qualitative data is in line with the quantitative findings, since these responses highlight the skills most often selected in CQs, i.e. those needed to overcome the identified AD challenges.

Phase I Data
In EPQ1, quantitative and qualitative data were consistent for variable 1, perception of difficulty. Quantitative data showed that respondents only perceived content and word selection as challenges. This was supported by the virtual lack of references to difficulty in the qualitative data, with only one response referring to those aspects. This consistency is also observed for variable 2, perception of usefulness. Most respondents perceived AD as either less useful than translation or equally useful. This was supported by the qualitative data: OEQ responses highlighted either the limited usefulness of AD or its potential for improving lexical and transferable skills, including accessibility awareness. As for variable 3, perception of learning, improvement was only commonly perceived in transferable and lexical skills. The qualitative data also included references to limitations, thus reinforcing the idea of insufficient learning. In sum, EPQ1 data support the 'triple-connection hypothesis', as they suggest a direct relationship between a moderate to low perception of difficulty and usefulness of the AD task (variables 1 and 2) as well as of learning progress (variable 3).

Phase II Data
Conversely, EPQ2 quantitative and qualitative data evidence a higher perception of difficulty of the AD task (variable 1). Idiomaticity, lexical choice, summarising, and oral narration were identified as particularly challenging both in CQs and OEQs. Similarly, EPQ2 respondents tended to express a higher perception of usefulness (variable 2). Question design in EPQ2 linked usefulness to specific skills, revealing that a majority had found AD particularly useful for transferable, communicative and lexical learning. As for variable 3, the quantitative data showed a high perception of learning progress, especially in transferable, communicative, lexical, and oral skills. The qualitative data supported this and suggested a link between communicative and idiomaticity awareness. Furthermore, they revealed a perception of usefulness for metalinguistic awareness raising, an aspect on which respondents had not been explicitly asked to reflect. Overall, EPQ2 data also support the 'triple-connection hypothesis', as they suggest a direct relationship between high perception of difficulty and usefulness of the AD task as well as of learning progress (variables 1 to 3).

Limitations
Firstly, further analysis would be necessary to determine statistical significance for the observations obtained from this exploratory analysis. It would also be worth investigating the factors behind the differences in participant perception in Phase I and Phase II, two of which may be the following: 1) that the group-work arrangement in Phase I was counterproductive both in terms of learning -some students felt it had been a drawback due to lack of application on the part of some classmates-and in terms of project organisation; and 2) that the researcher had not been directly involved in the delivery of the sessions in Phase I, 21 which may have hindered communication with students, negatively affecting their engagement.
Secondly, the minor design shortcomings identified in EPQ1 and the modifications introduced in EPQ2 did not facilitate a straightforward analysis, as data had to be regrouped for comparison. A more systematic examination of the impact of these factors would help in understanding their implications. However, these limitations have been taken into account, and comparisons have been made only to the extent to which this was possible without compromising the integrity of the analysis. The degree of consistency observed across questionnaires and data suggests that a strong hypothesis can be postulated, offering a fertile ground for further research.

Conclusions
4. Perception of learning progress is commonly observed among those sets of skills listed in (2), to a greater or lesser extent; 5. The expressed perceptions suggest a direct relationship between variables 1 to 3: the lower one is rated, the lower the others are rated (as in EPQ1), and the higher one is rated, the higher the others are (as in EPQ2). This supports the 'triple-connection hypothesis' between perception of difficulty of AD as a classroom task, perception of usefulness for skills development, and perception of own learning progress throughout.
Questionnaire responses not only provide data relevant for the research aims (perceptions of idiomaticity and metalinguistic awareness development), but also for the pedagogical aims of the AD Project (perceptions of communicative and integrated skills development). Having access to what students thought of and took from audio description provides valuable information about which learning outcomes can be expected from AD-based tasks as well as to how to minimise limitations and maximise learning potential. In sum, it may be worth reassessing the role of student perceptions in FLT research, considering them beyond their role as supporting evidence for linguistic data and seeing them as sources of valuable information in themselves.