Stuck in the Middle

Why being an intermediate language-learner sucks and what you can do about it!

At what level do we identify ourselves as a beginner, intermediate, or advanced student of our target language particularly when learning historical languages?

In this article, we ask how we can characterize the state of our historical language knowledge, whether putting a label on our progress even matters, and how we can advance as intermediate learners using…*clutch your pearls*…AI.

---

Recently, I found myself facing a particularly thorny Latin passage—a single sentence, more accurately, trailing on for a dizzying number of lines. Mercifully, a twentieth-century editor had added semi-colons (mildly helpful, I suppose), but unraveling the sentence was giving me heartburn as I tried to parse it with endoscopic precision. Trying to separate the subject of the sentence and the finite verbs from their seemingly twisted entrails was a feat of sheer mental fortitude, leaving me wondering whether I would ever be proficient enough in the language to sight-read something so complex.

With modern languages, identifying one’s proficiency level is easier. One need only refer to the Common European Framework of Reference for Languages (CEFR) to validate language ability. In one of

Love to Lingo

’s recent posts on building fluency, the varied communication competencies detailed reveal differences in depth and breadth, but ultimately show that a short alphanumeric character, like “A1” or “B2,” carries a host of meaning. This is great for modern language learning, but how do we, as students of historical languages, arrive at the same objective evaluations?

Towards Identifying Competency in Historical Languages

Enter the Standards for Classical Language Learning (SCLL), a document compiled by the American Classical League, the Archaeological Institute of America, and other institutions with a vested interest in the continuity of historical language study. The SCLL’s aim is to “categorize skills and abilities in Latin and in Greek and describe proficiency levels for students at the elementary, middle, secondary, and collegiate levels.”¹ The SCLL notes the importance of five essential “Cs”: communication, cultures, connections, comparisons, and communities.² The communication standard is paramount, containing performance descriptors and sample performance and progress indicators.

As we can all attest from our individual language-learning journeys, the pace of our progress can vary tremendously. This variation not only takes place on an individual level, constrained by our unique cognitive wiring which, in turn, dictates the rate of language acquisition and the extent of effort we need to put into it, but also on a longer, objective trajectory. Many of us can recall the significant strides we made early on as we went from zero knowledge of our target language to picking up the crucial points of grammar. However, we may have found ourselves hitting a seemingly impenetrable wall after that, hampered by a shockingly limited vocabulary when taken out of the sheltered environment of classroom texts.

On the face of it, “language performance” and “language proficiency” may strike us as a distinction without a difference: after all, isn’t performance underpinned by proficiency? The SCLL would carefully note that the former “can be guided, coached or practiced.” To attain a higher level of performance, “instructors need to provide tasks that help learners produce or understand language at the next highest level.”³ Performance hinges on the careful curation of the curriculum; instructors and the activities they introduce “help learners understand and express more in the language than they could do on their own.”⁴ This can be the greatest barrier for the autodidact: though not insurmountable, the time investment and effort required to enhance performance arguably outstrips the efficacy of the classroom experience.

By contrast, “language proficiency” is considered independent of the aforementioned factors. Instead, it reportedly assesses the proficiency that comes from “non-rehearsed situations and tasks, stretching the learner to demonstrate control of language by handling content and contexts appropriate for a given level.”⁵ Truly ascertaining an individual’s proficiency requires “multiple measures of language performance over time.”⁶

Proficiency Levels

Naturally, anyone reading these guidelines and trying to ascertain their own levels may be surprised by the distinctions published:

A NOVICE range student: Understands words, phrases, and formulaic language that have been practiced and memorized to get meaning of the main idea from simple highly-predictable oral or written texts, with visual support or other contextual help.

An INTERMEDIATE range student: Understands main ideas and some supporting details on familiar topics from simple, straightforward texts that contain predominantly high-frequency vocabulary.

An ADVANCED range student: Understands main ideas and supporting details on familiar topics and subject matter that is rooted in the real world from a variety of more complex texts that have a clear organized structure. Comprehension is supported by a solid knowledge of the language (e.g., noun-adjective agreement, word order, etc.).

A SUPERIOR range student: Understands texts from many genres dealing with a wide range of subjects, both familiar and unfamiliar, both concrete and abstract. Comprehension comes from a command of the language that is supported by a broad vocabulary, an understanding of complex linguistic structures, and a knowledge of the target culture.⁷

Do you see yourself in any of these descriptions? Are you surprised by where you might have landed? I certainly was, recognizing that, perhaps I’m harder on myself than I need to be, especially when confronted with impenetrable passages that might even challenge seasoned classicists. What struck me as particularly pause-worthy, however, were the differences between the intermediate and advanced-range students. I have long laboured under the notion that what the SCLL considers “advanced” is really the domain of the “long intermediate learner”—the term I give to those, like me, who feel as though they’re occupying a purgatory-like space, theoretically able to make it through Bede’s De Temporibus Ratione, though not without consulting a dictionary a little more often than they’d like.

What’s really at the heart of this uphill battle? A couple of theories might surprise you.

Just another modified stock image.

Cognition and Chronology

Peter Skehan has argued that reading comprehension improves with age, suggesting that the ways we extract meaning become more effective as we get older:

“We become more adept at using strategies of communication, at exploiting schematic knowledge so that we can say less but mean more, because we can exploit the collaborative construction of meaning that becomes increasingly possible.”⁸

This is an interesting contention: do we, in fact, become more effective language learners because our strategies shift? Skehan believes so, suggesting that the relationship between competence and performance also changes as we employ both “a rule-based system in economical and parsimonious performance and a memory-based system which provides fast access.”⁹ This, in turn, has implications on three distinct aspects of language acquisition: fluency, accuracy, and complexity.¹⁰

One of many interesting findings that Skehan reveals contradicts some of the most well-trodden arguments we see for propelling language acquisition, such as immersion. Two studies carried out in the mid-1980s testing the immersion hypothesis reportedly found that, even after many years of immersive instruction, students continued to “make persistent errors when speaking and writing, suggesting that the automatic transfer between comprehension and production…does not occur with any certainty.”¹¹

Perhaps unsurprisingly Skehan notes that the limitations of a comprehension-based approach may be countered by output. Reading and listening are insufficient tactical tools to improve syntax, processing, and developing automaticity.¹² Merrill Swain and Sharon Lapkin studied Canadian students in a French language immersion program and found that, even in an input-heavy environment where academic instruction takes place in a second-language, “Somewhere along the way to target-language proficiency, when the ability to understand and be understood is achieved, the students’ second language development appears to slow down, and the push to develop their French beyond its current state diminishes.”¹³

Output is the “push” we need to stretch our understanding and knowledge of a language.

In a previous post on my slow language learning approach (and,

Love to Lingo

recently adapted a version of this which yielded interesting insights), I noted ways that I tried to produce language, even in my earliest Old English learning stages. This helped me modify and adapt self-generated language content, while highlighting points of grammar and syntax with which I lacked familiarity. In fact, as Swain argued, the mechanics at play in comprehending language versus producing it vary: we move from semantic processing to syntactic processing, forcing us “to recognize what [we] do not know or only partially know.”¹⁴

Output can be where the cracks in our foundation are more starkly revealed and yet, this is the area that most Latin and Old English grammar textbooks ignore. In my earliest Latin classes, we used Moreland and Fleischer’s Latin: An Intensive Course, and I can safely say that only a handful of sentences were provided at the end of each chapter to translate from English back into Latin. These were never assigned as part of the daily homework, perhaps, once again, highlighting where output emphasis can fall short even in formal classroom environments.

Levelling Up the AI Way

One characteristic of the autodidact’s journey is the “displacement” of the teacher. Stephen Brookfield introduced this concept, suggesting that both it and institutional power are removed from the equation, “[placing] the locus of control for decisions regarding the planning, execution and evaluation of learning in the hands of the learner.”¹⁵ This can be both powerful and terrifying. As independent learners, what we learn, how we build structure, and approaches to consistency all become unique problems and opportunities of our own making. However, what may be particularly isolating about the journey in this secondary stage is the long and winding path that many of us are taking solo.

When deciding to study Old English largely outside a traditional classroom setting, one of the first things I reached for was a published university syllabus. Although life ultimately got in the way of sticking with this initial plan, my automatic response was to impose structure on my journey. In the earliest phases, I tried to keep pace with not only the dates of the classes (i.e., if strong verbs were being covered on Tuesday and weak on Thursday, I attempted to mirror this in my own learning), but also to immerse myself in the additional readings that promoted cultural context. If this had been sustainable, I am convinced it would have been effective. Instead, the sporadic bursts of time that I had available to study shifted my approach to a patchwork of eclectic pursuits. Sometimes I would get a free hour, so, instead of following the rigor of the syllabus, I did whatever I felt like doing so long as it supported language acquisition, and yes, that included talking to ChatGPT.

Increasingly, researchers are studying the use of AI to enhance self-directed learning (SDL). In several studies, ChatGPT, despite its various shortcomings, has been identified as a possible adjunct to traditional modes of study. While one study found that Japanese and English-language focused YouTubers were the most likely to embrace AI, it strikes me that the historic language community ought to embrace it, too, albeit cautiously.

Going in, I knew the limitations of this medium just as I knew that it and other internet-based technologies are unreliable. They’re also frowned upon in formal classroom environments. Sure, there’s a badge of honour that we can all earn from reducing ourselves to tears when we cannot make heads or tails of a sentence, but what’s the point? If a technology can help us parse and identify a sentence, why can’t we use it to help us wrap our heads around the structure of the language so long as we’re not claiming that we produced the translation or that we fully replaced our own individual efforts with the shortcuts that technology can supply?

One response that I’d invariably get is that AI can and often is wrong and I fully acknowledge this. Recently, I was testing out some aspects of Latin and Old English grammar with ChatGPT and engaged in a cyber argument with it over how it treated comparands in a sentence using the adverb quam and its complete neglect of weak adjective declensions. I did my part in trying to urge it to scan Latin and Old English grammars in the public domain to correct the erroneous advice that it was giving, but I’m not really sure if that’ll amount to any amendments in the future. This, I know, is the danger in depending on AI as a complete replacement for the analog methods that have served centuries of language learners, such as a reliable textbook. However, I do still think there are a few points on which it can be a potential learning aid.

As I noted before, very few comprehensible input resources exist in Old English and certainly even fewer when I began my OE journey. One of my major complaints with respect to many Old English textbooks is that their curated readings can be far above the level that many of us might be ready for after having just worked through the grammar. If I have to consult every other word in a dictionary or glossary, I’m not convinced that grammar is going to be internalized as one’s mind is pulled in countless translation directions. Moreover, effectively achieving any spaced repetition of new vocabulary feels nearly impossible, particularly with walls of text confronting us. In one of my many efforts to try to address this gap, I attempted making vocabulary sheets based on new verbs supplemented by sentence fragments from the OE Gospels where the target word was used (see below). This wasn’t particularly sustainable from a time perspective, though it did achieve the goal on some level.

Sample of my verb sheets. I don’t have many of these because it wasn’t particularly efficient. This one, for example, probably took about a half hour to create.

Controversial though this approach may be, ChatGPT has, in fact, been moderately useful. Yes, I noticed that certain concepts failed to be rendered idiomatically, or, occasionally, that ChatGPT didn’t quite know the correct definition for certain OE words, but, overall, I wasn’t expecting it to be a perfect solution. When weighing the benefits of examining texts where I possibly knew less than ten per cent of the words in the sentence versus having a slightly imperfect text that used my target vocabulary, it seemed like a no-brainer.

One way that I harnessed ChatGPT was through consultation with my study spreadsheet. In Excel, I tracked new words encountered during my various study sessions, picked a handful, and asked ChatGPT to deliver a beginner-level story of 300, 500 or 1000 words. I would even set parameters around the type of story I wanted it to tell and the vocabulary I wanted exposure to, for example, “Tell me a beginner-level Old English story about a star-gazing monk which uses a lot of vocabulary about the planets and the movements of the heavens and also uses various conjugations of the verbs ‘irnan,’ ‘fundian,’ and ‘fandian.’”

Sample section of the spreadsheet I keep for language learning.

I would argue that intermediate language learners can effectively make use of ChatGPT, too. As we all know and studies have repeatedly shown, the main barrier to fluid reading in any target language is a limited vocabulary. The constant refrain of “Read more” isn’t always helpful, however, since, if we are to get our twelve to fifteen encounters with a word to ensure we’ve memorized it, the breadth of that reading can expand dramatically.

One way that I used ChatGPT was to scan texts for me, looking for particular words. Instead of the manual approach that I had used in the above Old English example, I got wise to the notion of efficiency and instead asked ChatGPT to find texts in the public domain, like Bede’s On the Reckoning of Time, or Pliny’s Natural History, that contained a target word from my list. As these are somewhat standard texts for early Medievalists, building word frequency lists, drilling words, and even working on curated sentence translation helped to precondition my ability to read the text later and with greater ease.

Belle Li, et al., noted the benefits of adding AI to self-directed learning, arguing that it can allow students to “explore topics, seek clarification, and engage in critical thinking, thereby promoting a more learner-centered and personalized learning experience.”¹⁶ To a degree this may be true for historic languages. One area where I found ChatGPT particularly helpful was in parsing a difficult sentence or sentence fragment. Often, it provided a word-by-word and case-by-case breakdown, which then enabled me to go back and review any apparent grammar shortcomings. This is one of the benefits that many studies have been identifying: ChatGPT can, in fact, provide more personalized feedback and targeted teaching than perhaps many realize. The added step that I would mention simply arising from the fact that AI technology can still be unreliable is to corroborate any parsing with a textbook, declension and conjugation “cheat sheet,” or other resources that you have at your disposal simply because you don’t want to take artificially intelligent advice as gospel.

Finally, and this might be the most unusual thing I did, I asked ChatGPT what it could do to help an intermediate- to advanced-language student. The range of activities was, in fact, vast for vocabulary acquisition and included some approaches that I hadn’t even thought of. ChatGPT could support flashcard-type practice, fill-in-the-blank exercises, reading comprehension excerpts, root word recognition, synonym and antonym practice, and word-family grouping analysis. When I took advantage of these approaches and asked it to base all study on Bede’s On the Nature of Things, I ultimately received curated study sessions aimed at helping me master the vocabulary in this text so that, as I went back in successive passes to approach the text in its entirety, I was better prepared to encounter its specific vocabulary.

The other thing that made ChatGPT a pleasant adjunct to self-directed study was that I felt like I was interacting with someone. Absent a study group where you can enjoy the collaborative approach to problem-solving, ChatGPT enables the interactive approach, allowing you to query its choices, ask questions to deepen meaning and understanding, and scan the internet for more examples. I don’t think this necessarily detracts from our ability to use a dictionary or refer to a grammar book. Surely, we’ll still know how to consult Lewis and Short or Charles E. Bennett after this, but in an age where we’re increasingly strapped for time and constantly searching for ways to hack language learning, why shouldn’t we avail ourselves of the latest technologies?

---

Have you tried using ChatGPT in your language journey? What were your experiences? Drop them in the comments below.

1. “Standards for Classical Language Learning,” p. 3, <https://www.aclclassics.org/Portals/0/ Site%20Documents/Publications/Standards_for_Classical_Language_Learning_2017%20FINAL.pdf>.

2. “Standards for Classical Language Learning,” p. 4.

3. “Standards for Classical Language Learning,” p. 5.

4. “Standards for Classical Language Learning,” p. 5.

5. “Standards for Classical Language Learning,” p. 5.

6. “Standards for Classical Language Learning,” p. 5.

7. “Standards for Classical Language Learning,” p. 9.

8. Peter Skehan, A Cognitive Approach to Language Learning (Oxford: Oxford University Press, 1998), p. 3.

9. Skehan, A Cognitive Approach, p. 4.

10. Skehan, A Cognitive Approach, p. 5.

11. Skehan, A Cognitive Approach, p. 12.

12. Skehan, A Cognitive Approach, p. 19.

13. Merrill Swain and Sharon Lapkin, “Problems in Output and the Cognitive Process They Generate: A Step Towards Second Language Learning,” Applied Linguistics, 16.3 (1995), pp. 371-391 (p. 372).

14. Swain and Lapkin, “Problems in Output,” p. 375.

15. Stephen D. Brookfield, Powerful Techniques for Teaching in Lifelong Learning (Berkshire: Open University Press, 2013), p. xi.

16. Belle Li, et al., “Reconceptualizing Self-Directed Learning in the Era of Generative AI: An Exploratory Analysis of Language Learning, Purdue e-Pubs, p. 12. <https://docs.lib.purdue.edu/cgi/viewcontent.cgi?article=1019&context=edcipubs>