Teenagers face particular challenges when learning languages in a classroom settings, partly due to the complex physiological, psychosocial, and cognitive developments associated with puberty and adolescence (cf. Harklau 2007: 639-640). One of the effects this can have is a high “affective filter” (Krashen 1982), the concept that various attitudinal and environmental factors can influence the process of SLA. A learner with a low affective filter is able to more easily comprehend pedagogical input and therefore become more competent in the language, as compared to a learner with a high affective filter, which prevents the same input from effectively contributing to competence. While the affective filter model may have been developed over 40 years ago at time of writing, Behney / Marsden (2020: 38) argued that it is still relevant to the field today. This section will examine three of these factors, motivation, foreign language anxiety (FLA), and willingness to communicate (WTC), the role they play in teenage SLA, and how corpus-based DMC can be used to combat them, thus reducing the affective filter.

Motivation is the “desire to achieve the goal, positive attitudes, and effort” (MacIntyre 2002: 46). It has long been positively linked with achievement in SLA (cf. Krashen 1982: 30; Dunn / Iwaniec 2022: 988), and has even been called the “strongest predictor of achievement” (Ushioda / Dörnyei 2012: 403), not just among SLA (for an examination of motivation in mathematics, see Michaelides et al. 2019). Motivation can come from both internal factors to the learner and external influences, such as the classroom, and there is significant interplay between the two (cf. Dörnyei 2009).

Foreign language anxiety is a “subjective feeling of tension, apprehension, nervousness, and worry [in] the language learning situation” (Horwitz / Horwitz / Cope 1986: 125), which impedes the ability to process information and produce language in a classroom setting (cf. Sachs / Baralt / Gurzynski-Weiss 2023), and is particularly prevalent in teenagers and younger learners of languages (cf. MacIntyre et al. 2003; Russell 2020). This is in part due to the heightened fear of exposure to ridicule from peers or the avoidance of displays of vulnerability associated with puberty (cf. Taylor 2013: 2).2 These anxieties have been reported even to lead to “students discontinuing their study of foreign languages” (Dewaele 2013: 177).

The willingness to communicate (WTC) by a learner refers to the intention to use a foreign language in a communicative context and has been subject to significant investigation since it was first proposed as a concept in the 1990s (see MacIntyre et al. 1998).3 Increasing WTC is central to SLA, not only is it relevant for the development of language fluency (cf. Myhre / Fiskum 2021: 213) and proficiency (cf. Reinders / Wattana 2014: 102), but Dörnyei (2001: 52) refers to WTC as “the most central objective” for communication in a foreign language.

While there are several ways of addressing the three factors individually, some are impractical for time-pressed educators, such as personalised psychological evaluations to reduce perfectionism and improve self-confidence and stress-resistance (see Dewaele 2013). An aspect of common ground that has been shown in previous research to combat all three factors, is the development of learner-centred materials, i.e., adapting materials to be relevant to the lived experiences of the learners, in this case teenagers, with a focus on using digital environments and digitally-native texts. This should be no surprise as young people spend a considerable amount of time engaging with DMC and digitally-mediated content; according to Beisch / Koch (2022: 461-464), over 99% of young people (14-29) in Germany used the internet daily and almost 2.5 hours were spent every day on engaging with social media and DMC.

Teenagers and young adult learners have been regularly shown to benefit from innovative learning materials and environments. Digital environments have been widely demonstrated to play an important role in increasing WTC and decreasing FLA amongst teenagers as they are often familiar with such communication situations, leading to an increase in confidence, a willingness to request help, a reduction in anxiety, and an increase in fluency (cf. Reinders / Wattana 2014: 115; Kruk 2019: 17-21; Jauregi-Ondarra / Canto / Melchor-Couto 2022). These environments include course management systems (cf. Erbaggio et al. 2012), the comments section for news articles (cf. Marchand / Akutsu 2015), video games (cf. Peterson 2016; Kruk 2019), and a mix of multiple digital platforms (cf. Lee / Lu 2021). However, simply placing learners in a digital context may not automatically promote their WTC in an additional language. It is important to combine the context with the access to learning material that contains cultural similarities to the learner (cf. Ballis 2014: 220), such as DMC.

The NottDeuYTSch corpus can prove useful as part of the development of SLA resources to combat all three factors: increasing motivation and WTC, and decreasing FLA, as a source of DMC familiar to the target learners. Integrating YouTube comments in the classroom can expand the existing materials in an educator’s arsenal, to be used alongside YouTube videos, which have also been demonstrated to increase motivation (cf. Kabooha / Elyas 2018: 78; Dizon 2022: 24). By utilising language and (digital) communication situations familiar with learners, learners may more readily engage with the material and communicate in the target language.

However, it is relevant to stress that while the motivation of teenage learners increases and FLA decreases when DMC material is introduced into classroom materials (cf. Ritchie / Black 2012), it is DMC that was authentically produced, rather than texts contrived to resemble DMC. Section 2.2 discusses the importance of using authentic materials to construct and compile SLA resources, focusing on DMC and sourcing materials from corpora.

The use of authentic language for learning materials is based on the premise that learners benefit from analysing real-world language data to discover patterns and regularities in the language and so become more aware of realities of communication in the target language, which can improve SLA in many stages of the language learning process (see Gilquin / Granger 2022). Previously, language teaching relied on educators contriving examples, such as conversations (for practice dialogues in 16th century, see Villoria-Prieto / Suso López 2018), or learning by rote in the 19th and early 20th centuries (cf. Cohen 2018). From the 1970s, the benefits of authentic materials for learners have been shown to improve cognitive engagement in the target language to solidify acquisition (cf. Craik / Lockhart 1972) and later work demonstrated further psychological and affective benefits, linking learner exposure to authentic materials with an increase in WTC, thus lowering the affective filter (cf. Gilmore 2007).

However, while contrived examples and texts do not prepare learners for the reality of language, most authentic language has been “produced in order to communicate rather than to teach”, therefore, it is recommended that “a version of an original which has been simplified to facilitate communication” (Tomlinson 2012: 161) be used for learners. For example, learners can be asked to search for patterns and regularities, to identify collocations, to analyse the use of grammar and vocabulary in context, and to use the language data to create their own texts (cf. Gilquin / Granger 2022: 434-436).

The use of DMC as a source of authentic language for SLA has significantly increased over the last 30 years, with studies employing a variety of DMC sources, including Internet Relay Chat (IRC) (cf. Smith 2005), online forums (cf. Meskill / Anthony 2005; Ritchie / Black 2012), email (cf. Vyatkina / Belz 2006), social media (cf. Chawinga 2017; Barrot 2022), online gaming (cf. Peterson 2016), watching YouTube videos (cf. Terantino 2011; Arndt / Woore 2018; Kabooha / Elyas 2018), and analysing YouTube and Facebook comments (cf. Chen 2020). While the majority of existing studies have investigated learners of English, research has also encompassed Russian (cf. Meskill / Anthony 2005), German (cf. Vyatkina / Belz 2006), and French (cf. Ritchie / Black 2012), and Terantino (2011) additionally offers a broad overview of resources for learning several languages. Studies have also demonstrated the benefits of using DMC in specific areas of SLA, for example, vocabulary learning (cf. Kabooha / Elyas 2018), writing skills (cf. Ritchie / Black 2012), modal particles in German (cf. Vyatkina / Belz 2006), and pragmatics (cf. Cunningham 2019).

Unfortunately, there are relatively few corpus-based resources of DMC for SLA. Existing SLA DMC corpora include the Telekorp corpus (cf. Vyatkina / Belz 2006) of email correspondence (roughly 1.5 million words) between learners of English and German, which serves as a useful bilingual corpus resource between the two languages. The NAIST Lang-8 corpus (cf. Mizumoto et al. 2011) of language-learning social media posts (580,549 entries/comments) contains data in many languages, including English, Japanese, and Mandarin Chinese. The deL1L2IM corpus (cf. Höhn 2015) was compiled from instant messaging dialogues between German speakers and learners (52,000 tokens). The CMC learner and reference corpora (cf. Marchand / Akutsu 2015) were created from user comments underneath news websites (1.6m words). Finally, the Uppsala WordReference corpus (cf. Berdicevskis 2020) contains roughly 170m words taken from forum messages in English, Spanish, French, and Italian. However, none of the above corpora contain language explicitly produced by younger speakers that can be used to teach other younger speakers useful context-dependent language skills. This is representative of a wider problem, identified by Goschler / Stefanowitsch (2014), that there are few suitable corpora for learners that are representative of a particular group of speakers. Additionally, they identified three further problems with the corpus landscape that affects the ability of producing effective learning materials: many learner corpora are too small, they exhibit a lack of speaker variety, and it is often not clear what prompted the language production collected for the corpus data. Section 2.3 provides an explanation of how the NottDeuYTSch corpus can tackle these four limitations.

YouTube comments have been an, as yet, untapped source of authentic language from young people but studies of either DMC or youth language have rarely analysed the linguistic features used by young people in YouTube comments, despite the platform being the one of most-accessed means of communication in this demographic for several years (cf. Saferinternet.at 2018; Bahlo et al. 2019: 80) with 77% of German 14-19 year-olds describing themselves as active users of YouTube (cf. Statista 2023a), slightly behind Instagram with 79% (cf. Statista 2023b). The NottDeuYTSch corpus is one of the first large corpora of linguistic data containing online language from YouTube comments, i.e., what people have written underneath the videos, rather than the content of the video (see Figure 1).

The NottDeuYTSch corpus can also address the four shortcomings mentioned in Section 2.2 in the corpus landscape for SLA learners in the following ways:

A statistical breakdown of the corpus can be found in Table 1 (adapted from Cotgrove 2022: 343).4

YouTube itself has long been a platform containing SLA materials, in particular, content creators uploading videos that offer to teach languages, specifically in an extramural context (cf. Chowchong 2022). While these can be helpful to learners, the videos come with a host of caveats, such as the lack of quality control and reliable feedback method (cf. Dizon 2022). Within the YouTube platform, users are reliant on recommendations from the algorithm or reading judgements from other viewers in the comments to evaluate the potential quality of a video for SLA, especially since there is no easy way to quantitatively gauge quality following the removal of a public ‘dislike’ counter under a video. YouTube language content creators often try to use the platform to advertise paid courses, also without recognised qualification (cf. Bruzos 2021).

Within the comments section of SLA videos, there are often a significant amount of comments engaging in metalinguistic discourse, i.e., opining on ‘correct’ pronunciation or spelling, among other topics. While these comments have the potential to guide users further in their language learning, as well as providing a feedback mechanism if a commenter replies to a question, similar to an online forum, again quality control proves tricky for a learner; the number of comments containing metalinguistic discussion is often too large for the effective synthesis of knowledge when faced with so much new material (cf. Benson 2015). The applications for a corpus consisting of comments written under language videos would be limited for DaF/DaZ learners as there would be a lack of authentic target language, compounded by the heightened awareness of the commenters to write what they perceive as ‘correct’ German. While the NottDeuYTSch does contain metalinguistic discourse, it is produced in an environment where the viewers are not primed to think about adherence to lexicogrammatical rules.

Dizon (2022: 22) reported of concerns by learners that “[watching] YouTube [videos] was not an effective means to improve English ability when compared to more formal learning environments”, despite finding YouTube “an interesting and flexible language learning method”. I would argue that teenage learners stand to significantly benefit from the combination of YouTube content, in this case comments, with the structure and feedback offered by teacher-led learning. Section 3 discusses how educators can create learning resources containing YouTube comments, providing several worked examples.

In the NottDeuYTSch corpus there are many comments containing examples where spelling, word order, punctuation, and other grammatical aspects would be considered as ‘non-standard’. Busch (2021) convincingly argued that the “binary paradigm of standard and non-standard” is outdated, but often within an institutionalised setting, such as a classroom, any deviation from a narrowly prescribed ‘standard’ language is classified as an error (cf. Milroy / Milroy 1999: 83), with the educator or learner unable or unwilling to discuss the nuances of linguistic variation. Therefore, a more pragmatic approach is necessary, where educators can use YouTube comments to demonstrate how young people communicate online, and how learners can identify and comprehend DMC-specific linguistic features, and how learners might expect the comment to be written in standard German. While non-standard language in DMC is not an exclusively youth-specific phenomenon, it occurs significantly more frequently and more varied in youth language than in other communicative situations, such as informal adult communication (for corpus-based comparisons of non-standard lexis and syntax, see Cotgrove 2022). Generally, non-standard language usage can be utilised by educators for a number of classroom activities including:

The following examples taken from the corpus exhibit a wide range of non-standard features, which are discussed underneath:

Example (1) contains orthographical variation in ir (ihr, which should also be capitalised at the beginning of the sentence), seit (seid), and dum (dumm), as well as a missing space following the comma between peinlich and unlustig. The comment would work well in discussing spellings and avoiding common mistakes based on phonology, such as ihr seit. The comment can also be used within material discussing interaction (see Section 3.2) and impoliteness strategies (see Lorenzo-Dus / Blitvich / Bou-Franch 2011).

Example (2) contains the phrase echt mega, an example of an increasingly common way for young people to express joy in informal situations, similar to geil in the last decades of the 20th century. The adverb echt serves as an opportunity to discuss intensification, such as alternatives to sehr when providing opinions. Furthermore, mega could serve as a springboard for discussion on language change, having originally been used as an intensifier itself, e.g., megageil, before first becoming an interjection (mega!), and then an adjective in its own right, productive in adjectival phrases, and able to be intensified itself.

Example (3) also contains lexis specific to young people in krank, although in this comment used to express negative emotions. However, krank, similar to the (British) English sick, can also be used to express positivity (cf. Palacios Martínez 2018: 381), which can lead to discussions on meaning change. The comment also contains an example of a subordinate clause containing verb-second word order (weil sonst wäre das echt krank) instead of verb-last word order (weil sonst das echt krank wäre). The use of verb-second word order following subordinating conjunctions has dramatically increased over the past decades within youth language, roughly doubling from 2009 to 2018 (cf. Cotgrove 2022: 180). The phenomenon is not just restricted to weil; clauses starting with obwohl, dass, and wenn have all been found in the NottDeuYTSch corpus alone. This feature is useful for lessons on word order, which is often a sticking point for learners of German, especially for speakers of subject-verb-object languages, such as English. Other deviations from the standard include common ‘mistakes’, such as the confusion of das and dass, where dass is the subordinating conjunction and das the definite article, and the lack of commas separating the main clauses from the subordinating clauses.

The lack of commas separating clauses is also present in Example (4). The comment contains a good example in wenn es gegongt hat for discussing the differences between the uses of the conjunctions wenn (for conditional sentences), als (for temporal sentences), and wann (for questions) (even if the reality is not so clear cut, e.g., wenn in temporal case, see Breindl / Volodina / Waßner 2014: C1). The comment could be interpreted as particularly conceptually oral, i.e., it represents spoken language in written form, for example the pronoun-dropping (weißt, instead of weißt du). While a discussion on the applications of the Nähe/Distanz model (see Koch / Österreicher 2007) might exceed the capacity of a final-lesson-on-a-Friday-class of 15-16 year-olds, it might aid learners in exercises where they should write dialogue (also refer to Section 3.2).

Authentic interaction between conversants is an often overlooked aspect of SLA, especially as there is a strong correlation between the quality of the input and the quality of a learner’s language output, with exposure to authentic language leading to effective production of the target language (see Li 2017: 58). Example (5) is a thread containing a short multi-party conversation between six participants.5 The video under which the comment thread was written, produced by comedy duo DieAussenseiter, has a deliberately low-budget aesthetic and a provocative style of humour that invites sarcastic or otherwise sarcastic comments.

This exchange can be used as a useful starting point for developing resources on a range of topics, such as orthography, interaction, and communication issues within DMC. Commenter A (all names have been anonymised) jokingly insults the video using extreme repetition of a representation of laughter (Muhahaha), to which Commenters D and E jokingly respond, calling A an opfa (Opfer, victim/loser) - a reference to a part of the video, and mirror the representations of laughter. However, Commenter F does not seem to understand the context and writes a comment that could be interpreted as a rejection of A (in particular, the ‘middle finger’ emoji) (for more on the role of emoji in YouTube comments, see Cotgrove 2022: 242). In the exchange, we see a demonstration of important pragmatic aspects of conversation between young people, such as the metacommunicative functions of laughter and repetition (see ibid.: 230). Here the exaggerated length of the representations of laughter (e.g., hahahahahahahahaha) underlines the irony of the statements made by commenters A, D, and E, i.e., they are trying to ensure that other readers of their comments understand that they do not really mean the video is bad or another commenter is stupid (for more on irony in DMC, see Thompson et al. 2016). Learners can hope to also understand or identify with the role of insults and impoliteness, e.g., ritualised insults, that are particularly relevant for teenagers (see Kehily / Nayak 1997; for more on impoliteness, see Blitvich / Lorenzo-Dus / Bou-Franch 2013), or how to avoid misunderstandings when communicating online (see Bou-Franch / Lorenzo-Dus / Blitvich 2012), which can reduce FLA and increase WTC.

However, the thread, such as Example (5), does not necessarily need to be used verbatim for the development of teaching resources. Educators can adapt such materials to the needs of the class, as long as the basis is authentic material, as mentioned in Section 2.2. Here educators could, for example, ask learners to continue the conversation, either written or orally, in the same style, to practice turn-taking, conversation repair, and other aspects of interaction.

The linguistic construction of youth identities has been a well-researched topic since the 1990s, often looking at constructions of youth ethnic identities (cf. e.g., Cutler 1999; Androutsopoulos 2005; Keim 2007; Kerswill 2013), but has tended to focus on lexis and phraseology, although more recent work has investigated grapheme-based variation (cf. Androutsopoulos 2018; Busch 2021), including emoji (cf. Cotgrove 2022: 251-252). Therefore, it is important to develop resources to increase WTC and motivation, and reduce FLA in teenage learners by focusing on how teenagers can use language to sound like an authentic teenager in the target language in DMC (and other youth-specific situations), in this case helping learners identify linguistic features in YouTube comments that index youth identities (cf. Silverstein 2003).

Examples (6) to (8) contain several features that, when combined, index a youth identity. In Example 6 we see the term of address junge (mate/lad, lit. boy, see also the use of alter in Gysin 2014), the use of modal particles (mal), and the dialectal spelling of nicht (net) used in southern and southeastern areas of the DACH region. Example (7) contains the popular youth intensifier echt (also present in Examples 2 and 3 in Section 3.1), verbs related to youth culture, e.g., zocken (to game) and labern (to chat, see Hee 2018), and phraseology, e.g., jemanden ganz doll lieb haben (see Nowotny 2005) or übertreib nicht. Example (7) and (8) also contain linguistic features specific to youth DMC, such as the representations of laughter (hahah), graphicons (, :D), and the hashtag #glockeaktiviert, which refers to clicking the bell icon underneath a YouTube video or on a channel to be notified of when that channel publishes a new video (cf. Cotgrove 2022: 229-230). These comments can be incorporated into lessons and activities on register and sociolinguistics, and be used to demonstrate when certain words or structures are appropriate, developing learners’ language awareness (see Krumm / Jenkins 2001; Dannerer 2014). For example, learners can be set homework to write comments under YouTube videos, building on what was observed in the lesson, an exercise has been demonstrated to improve writing skills in language learners (cf. Ritchie / Black 2012). Such comments can also be used by learners to write short pieces or dialogues about themselves or peers, providing authentic ways of expressing themselves instead of defaulting to the almost rote answers of enjoying going to the cinema or playing football in the park with friends, thus empowering and motivating the learner to communicate (cf. Reinders / Hubbard 2013: 371).

There are, however, several limitations to using the NottDeuYTSch corpus to develop SLA materials, not least of all the restriction to produce material for intermediate teenage learners. The corpus requires that educators are familiar with YouTube and DMC, as well as text types and language forms that are found within YouTube comments. YouTube comments may not be as effective for teaching older learners or in societies where online videos are not consumed (although YouTube has a significant user base throughout the world). It might be difficult for educators to scaffold the material to make the comprehension of YouTube comments accessible for lower-ability learners (cf. ibid.; Dizon 2022), particularly if orthography is weak amongst the group, as Dizon (2022: 23) demonstrated that the incorporation of YouTube in learning materials is not beneficial for spelling, although it was shown to improve vocabulary.

There is no guarantee that lessons based on YouTube would enthuse a group of teenagers; it might be seen as an attempt by the educator to ‘look cool’, which reduces, rather than increases WTC. The corpus might also not enthuse a wide range of educators, due to the linguistic variation in the comments, as both DMC-specific and youth-specific aspects may be considered to contravene the prescriptivist standard language ideologies present in many classrooms (cf. Milroy / Milroy 1999: 83), and as such is believed that it should not be taught to learners.

Furthermore, the NottDeuYTSch corpus was not created specifically be a learner corpus (for an overview of learner corpora, see Fernández / Davis 2021), although the texts were written by young people, learning German mostly as a first language (for a summary of the reasoning behind this, despite the lack of ethnographic data, see Cotgrove 2022: 62-64). This might require some extra annotation or markup from researchers or educators but there is the opportunity to use the NottDeuYTSch corpus as the basis for the creation of a learner corpus, e.g., for the analysis of parallels between learners of German as a first language and an additional language, or between DMC styles, such as with other existing corpora of online data, e.g., the MoCoDa2 corpus of WhatsApp messages (cf. Beißwenger et al. 2020), the DiDi corpus of Facebook texts (cf. Glaznieks / Frey 2020), and the Internetbasierte Kommunikation (Internet-based communication, IBK) corpus, which contains blogs, emails, and internet relay chats (IRC) (cf. Lüngen / Kupietz 2020).

The biggest restriction to the use of the corpus in learning material is linked to the rapidly changing nature of youth language and DMC. The earliest data in the corpus come from 2008, which means that it contains several linguistic features that are already ‘out of date’ for young people, e.g., the preference to use emoticons rather than emoji,6 the prevalence of geil, or pop culture references. That being said, the rate of change for syntactic elements is significantly slower than for lexis, meaning that the corpus will still be useful, for example in 10 years’ time for studies of word order, if not as a way of learning the latest slang. Furthermore, the production of such corpora take a considerable amount of time, from the research of the source material, to balancing, data extraction, and cleaning, meaning that no similar corpus can truly be current. The NottDeuYTSch corpus, on the other hand, is ready to be deployed.