1To support intelligent learning, there is a growing trend of using information technologies to build intelligent, data-based foreign language teaching platforms (Miller and Wu, 2022). Meanwhile, although information technology is commonly integrated with English language education in China, there has been little research on the technology-enhanced teaching of other languages such as French, which is an official language in numerous countries and international organisations and a popular subject in Chinese universities (Lei and Bel, 2015; Xu, 2014). Hence, there is a need to fill this gap.
2Computer-assisted learning tools can provide new approaches to French language teaching. Given that one major challenge for Chinese learners of French is its phonological features (e.g. their intertwining with verb conjugations, gender–number agreements, and orthography), computer-exploitable learner corpora can help teachers capture generalisations of learners’ performance.
- 1 See Stansfield (1985) for a discussion of the history of dictation and Nieberding (1997) for a case (...)
- 2 Leaving teachers’ feedback aside, learners’ poor performance and inefficiency at self-correction in (...)
3This study focused on computer-assisted French dictation, which is a popular teaching and/or testing device in FFL.1 This is due to its holistic reflection of learners’ language proficiency and its ability to help teachers simultaneously assess learners’ listening comprehension and written production (Oual and Abadi, 2022), the proficiency of which is often obstructed by the difficulties mentioned above. Despite its importance, conventional paper-based dictations have obvious shortcomings, such as, for teachers, the time-consuming nature of corrections and the tedious task of systematically analysing errors and, for learners, the inefficiency of self-correction due to a lack of acuity of the French language. These impediments hinder teachers’ ability to measure learning progress and weaken the efficacy of both learners’ practice and teachers’ feedback, which may explain learners’ typically poor performance in dictations.2
4In this research, we adopted a design-based approach (Sandoval and Bell, 2004) to design, develop, and apply an intelligent dictation platform for FFL learners in China. Through pilot experiments, we aimed to answer the following questions:
-
Is dictation prized by learners, and why? How do students feel about using our dictation platform?
-
In what ways can computerised dictation be useful for teachers?
-
What frequent errors are made by Chinese learners in French dictations? How can the errors be described in a way that is compatible with both traditional classroom-based language teaching and the requirements of automatised error classification?
-
To what extent can a computer-assisted dictation platform be further exploited?
5The theoretical framework underpinning this study was based on error analysis (Corder, 1967, 1981), which views errors as systematic deviations by learners who have not yet internalised the rules of a target language. These errors reflect the learners’ current stages of language acquisition or underlying competence and are indicators of active engagement in hypothesis testing about language rules (Larsen-Freeman and Long, 2014). Integrating the interlanguage theory (Selinker, 1972; Yates and Kenkel, 2002), we understand these errors as part of an evolving interlanguage system, unique to each learner and shaped by influences from both their native and target languages. This perspective redefines errors as essential elements in the dynamic process of language learning, revealing common difficulties and the extent of language mastery.
6Studies have proposed a series of procedures for conducting error analysis (Corder, 1975), which involves identifying errors in a sample of learners’ productions, describing and explaining those errors, and evaluating them to enhance teaching strategies. Unlike essay writing, dictation exercises require a word-for-word evaluation against a reference text, where learners listen to a text (either previously recorded or dictated on site) and produce as faithful a replication as possible. Therefore, information technologies can be ideal for automatically identifying errors in dictations—our platform allows teachers to bypass the laborious manual corrections associated with paper-based dictations. Instead, they can directly focus on describing, explaining, and evaluating errors. Based on the data collected via the platform, we identified some frequent errors made by Chinese learners in French dictations and aimed to describe them in a formalised way.
- 3 For a recent study about the implementation of dictation in the classroom environment, see Cohen (2 (...)
7Another issue that needs to be addressed, and which has been the subject of a long-running debate, is the efficiency of dictation as a language teaching and/or testing tool. Although the perception of dictation has evolved over time along with the evolution of foreign language teaching theory (Stansfield, 1985), from our perspective, dictation is a useful device in that it combines pronunciation and orthography (‘la complémentarité de la prononciation et de l’orthographe’; Nieberding, 1997, p. 144). Accordingly, the question is how, rather than whether, dictation should be conducted, and this is the line of work that our research follows.3
8This study follows a design-based approach (Sandoval and Bell, 2004) through which, in order to solve real-life educational problems, researchers continuously improve the design based on users’ feedback from practice in a real and natural context until all flaws are eliminated and a maximally reliable and effective design is achieved. Following Easterday et al. (2014), design-based research processes consist of the following six iterative phases: focus on the problem, understand the problem, define goals, conceive the outline of a solution, build the solution, and test the solution. The next section details how the research was conducted, demonstrating the application of these phases to iteratively refine the educational tool and ensure its practical utility and efficacy in addressing the identified issues.
9We conducted the study in two phases, 1) needs analysis, design, and development and 2) pilot experiments, the details of which are given in the following sections.
- 4 This simply corresponded to the level of the textbook the learners used.
10Fifty Chinese FFL learners and two teachers from Beijing Foreign Studies University (BFSU) participated in this study. Group 1 comprised 27 undergraduate students (22 females and 5 males), and Group 2 comprised 23 undergraduate students (22 females and 1 male). The learners’ average age was 20 years. These students were majoring in international business, law, computer science, etc., rather than foreign languages, and therefore were learning French as a second foreign language as a beginner. They had been learning English for 10 to 13 years. Their French level corresponded approximately to CEFR levels A1 or A2.4 Group 1 started to learn French one year earlier and began to use the platform seven months earlier than Group 2. The two teachers were Chinese, and one of them taught both groups.
- 5 The platform was developed in collaboration with Beijing Waiyan Online Digital Technology Co., Ltd. (...)
11Through classroom observation and communication with both learners and teachers, we recognised the inconvenience of practising dictation in a traditional teaching environment. After designing a prototype that considered both learners’ and teachers’ needs, we fully developed an internet-based dictation system,5 which could later be integrated into a platform that provides a much wider range of online learning material.
12The pilot experiments commenced immediately after the finalisation of the development. Learners performed dictations on the platform, either following teachers’ assignment instructions or of their own accord.
13We gathered user feedback throughout the experiments to ensure iterative optimisation of the platform. One year after launching the platform, we distributed questionnaires to gather the learners’ perspectives and suggestions. Since only two teachers participated in our experiments, their feedback was collected through daily communication instead of a survey, which would have been statistically insignificant.
14At the time of writing, the first 50 learners continue to use the platform, and more FFL learners and teachers at BFSU have gained access to it.
15Before designing the prototype, we analysed several existing dictation platforms, such as Hujiang,6 Aboboo,7 Shanbay,8 and TV5 Monde Dictée.9 Hujiang supports dictations in multiple languages, including English, Japanese, French, and German, Aboboo and Shanbay are designed solely for English dictations, and TV5 Monde Dictée provides French dictations on various themes. These platforms all provide immediate, automatic corrections based on learners’ electronic inputs; however, they lack functionality that would enable human teachers to evaluate and analyse errors.
16It is in this context that our platform came into play, which considered both learners’ and teachers’ needs to improve dictation practice for the former and error analysis for the latter. The internet-based platform can be used on laptops or tablet PCs by both learners and teachers. The main functions of the platform are described in the following sections.
17For learners, the most important functions include (1) choosing exercises, (2) playing audio and entering text online, (3) checking feedback upon submission, and (4) receiving scores and comments from teachers. After logging in, learners see a dictation syllabus comprising 126 exercises from Progresser en dictée (niveau élémentaire) (Li, 2009)—a reference widely used in China. Another syllabus contains exercises provided by teachers from BFSU’s Faculty of French and Francophone Studies (10 exercises at present) that correspond to the learning progress of their classes (see Figure 1).
Figure 1. Interface for Choosing a Syllabus
18By clicking on an exercise, the user can access the appropriate interface. To simulate dictation in real-time classes, the audio text is played only once (although the text is read four times), and the play rate is not adjustable. However, learners can click the play button to pause the recording. Learners must finish each exercise within a limited time span determined by the teacher (e.g. 30 minutes). They can repeat the exercise, but there is a maximum limit (usually 10 times). Since we could not locate a French handwriting Optical Character Recognition (OCR) tool that met our expectations, the platform presently cannot analyse learners’ handwritten manuscripts. Instead, learners must enter text using the keyboard or an Apple Pencil connected to an iPad. The iPad operating system enables the French alphabet, written with the Apple Pencil, to be instantly recognised and converted into electronic text. To facilitate input, an image of a physical French keyboard is provided as a visual reference, along with virtual buttons for quickly entering letters with diacritics (à, é, ï, etc.).
19Upon submission of the dictation, the platform automatically checks learners’ inputs against the source text and immediately displays the results. This process relies on a language-independent, open-source text comparison algorithm10 and web technologies such as HTML, CSS, and JavaScript. To make the comparison more intuitively understandable, erroneous inputs and source text are highlighted in different colours (red and green, respectively) and displayed side by side (erroneous input comes before the source text), as shown in Figure 2. At this stage, learners can freely replay the audio.
Figure 2. Platform’s Immediate Feedback
20This process renders teachers’ manual corrections and learners’ self-corrections unnecessary, significantly improving feedback efficiency for learners. After a teacher’s review, the page shows the final score and remarks. Learners can review this page at any time, whereas traditional dictation sheets are easily lost.
21The platform’s teacher functions enable teachers to manage dictation syllabi, review learners’ exercises, and retrieve errors from the error database for systematic error analysis. After logging in, teachers can see the previously created dictation syllabi and modify the associated exercises. To create new exercises, teachers must provide an audio file and source text, along with specifications such as time limit, total score, points to be deducted per error, maximum number of allowed retries for an exercise, etc.
22During a review, exercises can be filtered according to the exercise name and/or learner name. The review interface highlights each discrepancy between the learner’s replication and the source text, which allows teachers to rapidly identify errors. Based on the number of errors, the platform can also calculate scores, which teachers can revise if necessary. In addition, comments can be added at this stage (see Figure 3, where the teacher gives explanations).
Figure 3. Teacher’s Review Interface
23Most importantly, the error database is updated following each submission, and the system stores each ‘incorrect form–correct form’ pair in the database (e.g. ‘ça-sa, six-ses, autre-autres’ in Figure 2). Teachers can filter the data by syllabus, exercise, and/or learner name. The data can be exported to an Excel file, allowing teachers to conduct detailed error analyses. For example, teachers can sort Excel data using the column named ‘correct text’. This operation shows, for every single word in a chosen source text, all erroneous forms. In Figure 4, for the word ‘chômage’, the database has identified ‘(void), chomage, chaumage, choumage’, and so on.
Figure 4. Error Sorting in Excel
24We started pilot experiments on 7 May 2021—the day the platform was launched—to investigate learners’ usage experiences, opinions, and suggestions, as well as to summarise and explain learners’ frequent errors based on the automatically collected data.
25By that time, the learners were following a conventional syllabus as scheduled (six hours of French per week in class). They were instructed to practise dictation on the platform after class, following teachers’ weekly assignments or at their own pace. We provided detailed usage instructions for all participants. During the experiments, the two groups’ dictation exercises gradually shifted from class-based to online exercises. By 24 May 2022, the 50 learners had submitted 610 dictations on the platform (mean = 12, standard deviation [STD] = 14). The highest number of submissions by a single learner was 87, and the lowest was 0.
26During the experiments, we gathered feedback through regular communication with users to iteratively optimise the platform. The main updates included 1) fixing bugs reported by users, 2) adding a French keyboard image to the text-entering interface, and 3) enabling text entry using an Apple Pencil for iPad users who preferred handwriting.
27One year after its first use, we distributed questionnaires (see Appendix 1) to collect learners’ feedback and received 31 completed questionnaires (18 from Group 1 and 13 from Group 2). The questionnaire contained 13 open-ended questions, and the responses totalled 6,479 Chinese characters. We conducted a thematic analysis of the survey data using inductive coding with NVivo software (Bazeley and Jackson, 2013). The steps included preparing and organising the data, reading through the data, data coding, theme mining, results collation, and data interpretation (Creswell, 2013). The qualitative analysis results are presented below.
- 11 In the earlier version of this article, we used the term ‘motivation’ instead of ‘efficiency’. We t (...)
- 12 Readers may wonder how dictation differs from listening comprehension in improving listening accura (...)
28First, we were interested in the efficiency of dictation from the learner’s point of view.11 All survey respondents recognised the importance of dictation for learning French as follows: 1) dictation helps to ‘improve listening comprehension ability’, which is of vital importance in communication;12 and 2) dictation comprehensively ‘combines listening comprehension and written production’, helping to sharpen learners’ attention to phonological phenomena and spelling in the French language. In terms of practice frequency, most respondents reported practising once or twice per week.
29Second, we investigated how learners entered text on the platform and whether they adapted effectively to this change. For those who frequently used the platform on a laptop, 16 learners (52%) directly typed text while listening to the audio, and 15 learners (48%) first wrote on paper before typing their text when the audio stopped playing. For tablet PC users, 8 learners (26%) wrote with an Apple Pencil, 8 learners (26%) first wrote on paper before typing the text, and 8 learners (26%) typed text as the audio played. Concerning expertise in using the French keyboard, only 2 learners (6%) reported that they typed French as quickly as they typed English, which enabled them to type while listening to the audio. Other learners reported lower proficiency when typing in French, and half of the respondents stated that they were willing to practise more frequently to achieve full expertise. Thirteen learners (42%) explicitly expressed their preference for handwriting because they found writing quicker and more fluid than typing and, hence, more comfortable. However, 4 learners (13%) claimed that it was easier to modify electronic text than handwriting. Almost all the respondents agreed that they adapted easily to entering text using a keyboard or an Apple Pencil, except for two learners in Group 2, who found it inconvenient.
30Third, we analysed learners’ opinions about the advantages and disadvantages of the platform and whether it helped them improve their dictation skills. The three most prominent advantages were 1) immediate correction upon submission, 2) the opportunity to repeat the exercise, and 3) greater flexibility compared to the class-based approach in terms of time and space. Learners also indicated that ‘they could hear the audio more clearly and that they felt less nervous when practising online’. They also reported that ‘the platform saved precious teaching time in class and provided access to more dictation materials’. In contrast, the five most prominent disadvantages included 1) the challenges of typing rather than writing, 2) the lack of collective learning, 3) the stress of being evaluated in a real-time situation, 4) slackness due to the lack of supervision, and 5) the inability to ask questions as they would do in class. All the respondents agreed that the platform contributed to improving their dictation skills, and the main reasons were as follows: 1) the platform allows for more practice after class, which guarantees a steady input and output ratio in language learning; 2) increased practice improves learners’ familiarity with French pronunciation, grammar, and vocabulary; and 3) immediate feedback helps learners quickly identify their errors.
- 13 Since the platform was originally designed to replicate class-based dictation in a real-time situat (...)
31Finally, we solicited learners’ suggestions for developing a more advanced dictation platform. The most common advice, suggested by seven learners (23%), centred on removing constraints on audio playback (i.e. permitting rewinding and play-speed adjustment). We adopted this suggestion because it favoured dictation exercise efficiency.13 Second, learners would have preferred the integration of more diverse materials (e.g. dialogue, interviews, and news items) and specific grammar components (e.g. past participles and noun–adjective agreements). Three learners (10%) mentioned that video-based materials would be more interesting than simple audio materials. Finally, learners wished for errors to be automatically classified and for frequent error types to be presented with associated distribution graphs. We are currently focusing on this idea, which was echoed by the two teacher participants, with the aim of developing an automatic toolkit supported by a rudimentary error categorisation model, as presented in the next section.
32Our investigation of the difficulties faced by Chinese learners in French dictation was grounded in the platform’s error database.
- 14 Considering that each learner participating in this study completed a different range of exercises, (...)
33At the time of writing, the database comprises 610 submitted dictations corresponding to 69 different exercises (words per exercise: min = 35, max = 169, mean = 127). For the error analysis, we chose samples from 22 learners, each of whom submitted more than 10 dictations (sum = 470, mean = 21, STD = 17). Due to staffing constraints, we selected a subset of these 470 samples (covering all 69 exercises) for manual error analysis. For each exercise, we analysed a certain number of samples until no new error types emerged, and we finally obtained 820 ‘incorrect form–correct form’ pairs.14 To sort these pairs, we developed an error categorisation model based on phoneme and grapheme pairs, as explained below.
34A phoneme pair denotes a pair of two phonemes, one of which is misheard and mistaken (i.e. relative to the ‘correct sound form’ as shown in parentheses in Table 1 below) and is therefore the ‘incorrect sound form’. A phoneme pair can be minimal or nonminimal. In the former, the two phonemes share an articulatory-acoustic resemblance to a considerable extent. Some minimal pairs exist in both French and Chinese (e.g. close-mid versus open-mid front unrounded vowels: [e] versus [ε]), while others only occur in French (e.g. voiced versus voiceless consonants). In the latter, the two phonemes normally do not share prominent phonetic features.
35The definition given above is provided to ensure that the two phonemes in all pairs can be formally opposed, but with enough flexibility to account for differences between L1 and L2 phonological systems. For example, following a strict minimal pair definition in the phonological sense, [e] contrasts with [ε] despite both being unrounded vowels, and the former is close-mid, but the latter is open-mid. These kinds of minimal pairs develop only in established and stable phonological systems, which is not the case in beginner foreign language acquisition.
36Our minimal pair system is proposed in an acquisitional sense because it is based on, in addition to phonological minimal pairs, an opposition we define as [+acquired] versus [-acquired]. Taking a real-life example to demonstrate how this system works: if a student mishears [œ], [o], and [ɔ], it is because they have acquired the feature [roundedness] and know, in the case of these three vowels, that its value is ‘+’, but have not acquired the features [posteriority] and [closedness] (nor their values), which are necessary for making clearer distinctions between the three. In other words, the feature X (i.e. [roundedness]) minimally contrasts with the feature Y (i.e. any feature that is not X, such as [posteriority] and [closedness]), in that the former is [+acquired] and the latter is [-acquired].
- 15 Again, some malleability in the definition is required. While an orthographic feature can be define (...)
37In the same way, a grapheme pair consists of a pair of two graphemes (or a bigram/trigram, even words in a looser sense), one of which is miswritten and mistaken as the other. The two components share identical pronunciations according to spelling rules. A grapheme pair can again be minimal or nonminimal. In the former, the two graphemes differ from each other only by one orthographic or lexico-syntactic feature,15 whereas in the latter, the two graphemes differ from each other by more than one orthographic or lexico-syntactic feature.
38Furthermore, the phoneme and grapheme pairs may or may not be lexico-syntactically (L-S) related. By ‘L-S related’, we mean that learners could have used lexical or syntactical knowledge to exclude erroneous forms.
Table 1. Phoneme Pairs
|
|
Minimal
|
Nonminimal
|
|
L-S related
|
1. il joué (→jouait) au foot
2. prendre une touche (→douche)
3. le goût (→coût) de la vie y est plus élevé
4. me laver et me baigner (→peigner) en dix minutes
5. mon travail est (de) téléphoner aux clients
6. vivent dans villes (→en ville)
7. elle (l’)a remercié avec
8. pour (l’)inviter
|
12. pour leur (→les) remercier
13. je le (→lui) promet de
14. au bord de (→du) fleuve
15. devant la (→le) téléviseur
16. je suis certainement (→certain) que
|
|
L-S unrelated
|
9. vivre en Provence (→province)
10. achter (→acheter)
11. spetacle (→spectacle)
|
|
Table 2. Grapheme Pairs
|
|
Minimal
|
Nonminimal
|
|
L-S related
|
17. je ne les (→l’ai) pas encore visité
18. elle accueil (→accueille) les clients
19. pas assez d’espace vert (→espaces verts)
20. peu de passant (→passants)
21. faire de bon (→bons) choix
22. cette (→cet) été
23. elle s’est séchée (→séché) les cheveux
24. comment s’est passé (→passée) ton inscription
25. goûter les biscuits que sa mère a préparé (→préparés)
26. les prix sont attribué (→attribués)
27. les plats ne coûte (→coûtent) pas cher
28. nous nous reverront (→reverrons)
29. je ne les voie (→vois) pas
|
34. vous n’avez pas apprendre (→à prendre) l’autobus
35. j’espère quel (→qu’elle) sera tout heureuse
36. une armoire pour mettre mes à faire (→affaires)
37. la distance apart courir (→à parcourir)
|
|
L-S unrelated
|
30. project (→projet)
31. dialect (→dialecte), frai (→frais), environs (→environ)
32. pardoner (→pardonner)
33. restaurent (→restaurant), chaumage (→chômage)
|
|
39Tables 1 and 2 show the error classification matrices obtained by deploying our model. From our perspective, this rudimentary formalised categorisation system captures several generalisations observed in FFL dictation among Chinese learners.
40First, most of the errors are L-S related. Given that learners have sufficient time to examine their outputs after the audio finishes playing, this suggests that the underlying problem in poorly performed dictation is (in many cases) a lack of expertise in vocabulary and grammar.
41Consider minimal L-S related phoneme and grapheme pairs: on the one hand, the abundance of open syllables in word formation (samples 1, 3–8) and in particular the monosyllabicity of functional words (samples 5–8) considerably diminish the recognisability of the words in question. Moreover, the mishearing is aggravated by the fact that many minimal phoneme pairs do not exist in Chinese, such as the systematic opposition between voiced and non-voiced consonants [d] versus [t], [g] versus [k], and [b] versus [p] (samples 2, 3, and 4). Additionally, grammatical features such as verb endings (samples 17, 18, and 27–29) and gender–number agreements (samples 19–26) are often silent, meaning that misspelling is often related to inaccurate acquisition of these features. In the case of nonminimal L-S related phoneme and grapheme pairs (samples 12–16 and 34–37), the errors were somewhat less expected because the divergence between the incorrect and correct forms was not minimal.
- 16 By ‘idiosyncrasy’, we mean the near-impossibility for beginners to predict where to put a silent le (...)
42Regarding L-S unrelated phoneme and grapheme pairs, errors are always minimally biased relative to the correct form. Various factors are responsible, such as a non-existent minimal phoneme pair in L1 (i.e. opposition between [ɛ̃] and [ɑ̃] in sample 9, which does not exist in many Chinese dialects), the elusive schwa (sample 10), consonant dissimilation (sample 11), or orthographic idiosyncrasies (samples 30–33).16 Additionally, L-S unrelated nonminimal errors are statistically the most improbable (zero occurrence in Tables 1 and 2), probably because learners have enough vocabulary or grammar knowledge to exclude such obvious errors. According to our predictions, although such errors are possible, they are unlikely to occur.
- 17 The learner in question acoustically parsed the phonemes as ‘ils-son-tun-peule-santé’ (instead of ‘ (...)
43The descriptive power of our model is not limited to the examples presented in Tables 1 and 2, because pairs can coexist to form clusters. One fascinating error of this type is ‘ils sont un peu le (→ en bonne) santé’ which involves four minimal pairs: [ɛ̃] versus [ɑ̃], [p] versus [b], [œ] versus [ɔ], and [l] versus [n].17 Finally, nonminimal phoneme pair clusters can represent word-order errors (e.g. ‘tu n’as pas m’appelé → ne m’as-tu pas appelé’).
44The methodology and findings in the present study bring further clarity and perspective, to Swanson’s remarks (2017):
Many studies do support the notion that dictation may be one suitable way to assess general proficiency […] But the fact that dictation activities may be able to help assess proficiency does not also mean that dictation is a means for developing language skills […] Future research should attempt to determine the effect of dictation on the various aspects of language learning: grammar, spelling, listening (both general comprehension and sound-spelling correspondence), etc. (p. 181).
45These remarks echo our research questions in the following ways: first, learners clearly noted in their feedback an insightful understanding of the goal of the dictation task, namely ‘combines listening comprehension and written production’, which is proof of its efficiency per se. All participants concurred that the platform played a positive role in enhancing their dictation abilities. The question, though, is whether dictation is efficient as a training or testing device in terms of language skill acquisition. Given the context in the present research, it is reasonable to think that dictation, at least when deployed in the traditional way, is more of a ‘flaw detector’ than a ‘skill trainer’. This is our understanding of the learners’ comments, which employed phrases such as ‘improve listening comprehension ability’. This correlates to error analysis that, in the sense of Larsen-Freeman and Long (2014), views errors as engagements in hypothesis testing within the process of grammatical rule formation.
46Second, the borderline between testing and training is subtle because for a given grammatical feature to be acquired, one needs to realise, and to be repeatedly reminded of, its absence in one’s own language system. Accordingly, the identification of a flaw is the onset of training. However, if we can deliver these reminders on a massive scale, testing becomes training. This is one of the greatest advantages of a computer-assisted dictation platform: since exercises are sortable by learner name, the teacher can track one learner to see all the errors they have made and discern a pattern. Or, when filtered by exercise name, the errors show a summary of what the class has (and has not) mastered, which is effective feedback for the teacher. Crucially, the more the platform gathers data, the more it can depict these patterns in a principled way, hence its ‘intelligence’.
47Third, the various aspects of language learning mentioned in Swanson (2017) are, thanks to these findings, now more fathomable. The first deployment of our error categorisation model shows that the postulation of minimal and nonminimal phoneme and grapheme pairs can lead to a system endowed with significant descriptive power. The finding that most errors are minimal and L-S related is revealing because it suggests that although dictation resorts to listening abilities for input, the underlying language skill that guarantees the correct outcome is learners’ quick access to a well-established lexico-syntactic system.
48Fourth, we need to further refine our error classification model by adding fine-grained tags, such as SPELL, PRON, VERB-FORM, NOUN-NUM, and PHONETICS, in order to describe the lexico-syntactic system in question. As a follow-up to the present study, we have adapted an automatic English grammatical error annotation toolkit (Bryant et al., 2017) to classify French dictation errors, which is rule-based and relies solely on part-of-speech tagging, dependency parsing, lemmatisation, and stemming information (Qin et al., 2023). Once this system is fully developed, we will integrate it into the platform in the hope of providing more teacher-like tutoring.
49We also expect to develop other extensions for the platform, especially the possibility for learners to play the audio more than once in each dictation exercise. Specifically, the learners would have the choice, after finishing the first listen of the dictation (and rather than directly accessing the instant correction), to highlight words that they are unsure about and listen to the audio again. During the second playback, learners would be able to modify their initial inputs before moving on to the third (and final) playback, where the second input is automatically corrected and visualised, and learners could focus solely on the marked errors and eventually correct them. The platform would track all changes made throughout the three attempts and feed these data into a database, which would help us better analyse learners’ errors and listening strategies in dictation.
50Our study demonstrates the efficacy of the intelligent French dictation platform, highlighting its potential to enhance the learning and teaching of French as a foreign language. The platform’s ability to provide immediate feedback, facilitate error analysis, and offer a flexible learning environment is particularly beneficial for language acquisition.
51Key findings from our research include the platform’s positive impact on learners’ dictation skills, the significant time and effort saved for teachers in error correction and analysis, and the identification of common errors made by Chinese learners. These outcomes underscore the value of integrating technology into language learning practices.
52Additionally, we aim to incorporate user-centred design improvements based on the feedback collected from our users. This iterative process of enhancement will ensure that the platform remains a cutting-edge tool that meets the evolving needs of language learners and teachers.
Flux RSS
