ELTJam, machine learning, knowledge and skill

“Knowledge and skill are different. Vocabulary acquisition tools help learners improve their knowledge, which may in turn have a positive impact on skill, but it’s important to be cognisant of the differences.” [https://eltjam.com/machine-learning-summer-school-day-4/]

“We need to be careful though not to oversell the technology and be clear about what it can and can’t do. There is no silver bullet. This is especially the case when it comes to skills vs knowledge; a lot of the applications that could come from this sort of technology will help improve knowledge of English, and may contribute to accuracy” [https://eltjam.com/machine-learning-summer-school-day-5/]

The above two quotes are from a nice series of posts by ELTJam on a machine learning workshop. The first point from the first quote is indeed important to recognize. Bill VanPatten (2010) has argued that knowledge and skill are different. However what is meant by knowledge and what is meant by skill? For a nice video summary of the VanPatten paper see the video linked below.

Knowledge is mental representation which in turn is the abstract, implicit and underlying linguistic system in a speaker’s head. Abstract does not mean the rules in a pedagogical grammar rather it refers to a collection of abstract properties which can result in rule-like behaviors. Implicit means that the content of mental representation is not accessible to the learner consciously or with awareness. Underlying refers to the view that a linguistic system underlies all surface forms of language.

The actual content of mental representations include all formal features of syntax, phonology, lexicon-morphology, semantics. And a mental representation grows due to input being acted on by systems from the learners mind/brain.

Skill is the speed and accuracy with which people can do certain behaviours. For language skill this refers to reading, listening, writing, speaking, conversational interaction, turn taking. To be sure being skilled means that the person has a developed mental representation of the language. However having a developed mental representation does not entail being skilled. How skill develops depends on the tasks that people are doing. A person learns to swim by swimming. A person learns to write essays by writing essays.

It follows that the Write&Improve (W&I) tool (as the flagship example of machine learning based tool for language learning) can be seen as targeting how to be skillful in writing Cambridge English Exam texts. The claim that machine learning, and by implication the feedback by W&I, is changing the knowledge of the learner’s English does not accord with VanPatten’s description of knowledge as mental representation. His description implies that no explicit information, in the form of feedback in the case of the writing tool, can lead to changes in the mental representation of the language of writing. He states that research into writing is unclear as to whether feedback impacts writing development.

My point in this post is to briefly clarify the distinction between knowledge and skills (do read the VanPatten paper) and to suggest that the best machine learning based tools can offer are opportunities for students to practice certain skills.


W&I has never claimed that its tool has impact on language knowledge. See Diane Nicholls comment below.


Van Patten, B. (2010). The two faces of SLA: Mental representation and skill. International Journal of English Studies, 10(1), 1-18. PDF available [https://www.researchgate.net/publication/267793221_The_Two_Faces_of_SLA_Mental_Representation_and_Skill]

BlackBox Videocast 2: Mental Representation and Skill


16 thoughts on “ELTJam, machine learning, knowledge and skill

  1. Hi Mura,

    Good summary of VanPatten’s position, but we should remember that many, even within the “cognitive” framework, would disagree with VP’s view of both “knowledge” and the relationship between explicit and implicit knowledge. The claim that the feedback by W&I is changing the knowledge of the learner’s English will strike many as perfectly reasonable. As is so often the case, we must conclude “More research is needed”.

    1. hi Geoff

      which critics in the “cognitive” framework are you referring to?

      i think VanPatten has a nice argument in the case of explicit/implicit knowledge as applied to input processing (he describes this in Foundations of processing instruction – https://scholars.opb.msu.edu/en/publications/foundations-of-processing-instruction-2)

      essentially he argues that since input processing is about the underlying features of language and from his definition these underlying features are implicit then explicit-implicit debate is irrelevant : )

      1. Hi Mura,

        Yes, just as you say. He’s in the “Generative” camp, and all those who are not (e.g.s Long, Nick Ellis, Robinson, Skehan, O”Grady, Tomaselli, and, ehem, Larsen-Freeman) disagree.

  2. Interesting read, Mura. Machine learning-based tools for extra practice always come in handy. I immediately tested Write&Improve; I pretended to be a beginner describing what I can see through the window. 🙂 I’d say that using this tool can be highly motivating, particularly for less confident students. Anyway, I’ll think I’ll introduce it to my class when we are in a computer lab. As for the knowledge vs. skill dichotomy: “You’ll never know till you try”.

      1. The W&I tool provides students with opportunities to practice the skill of writing, which is clearly to the good. As for the feedback and its impact on SLA, I understand why some are doubtful. My students will make the same mistakes over and over again in writing regardless of how many times I drew their attention to them. However, based on my experience, some aspects of the skill *are* can be affected by explicit feedback. To give an example, some of my weaker students got a very good grade on the final written exam just because they fulfilled all the requirements – they got the message across, elaborated on all the points, knew when to make a new paragraph, chose the right way of addressing a person in a letter, wrote the correct number of words, etc. Although they scored low on grammar and vocabulary (which I also gave them a lot of explicit feedback on), overall they did really well. So my point is that it depends on what type of feedback the W&I tool gives. Some of it may indeed be useful and it may help students pass their exams. Correct me if I’m wrong but that’s what the tool was originally designed for.

      2. hi Hana
        yes i agree that such tools can provide ways to help with “skills”; the original quotes from the ELTjam posts were talking about it helping language “knowledge”

        this post briefly reported VanPatten’s conception of knowledge and skill

        the issue of no effect of written (corrective) feedback is an implication if you agree with VanPatten’s description of knowledge and skills; which of course is debateable

        i think if you see improvements in student’s knowledge as measured by tests from explicit teaching the debate might ask how much those tests measure explicit knowledge as opposed to implicit knowledge


      3. Thanks to my bad use of the “Comments” function, this discussion is getting rather out of synch.

        In an attempt to be clear: VanPatten’s argument rests on adopting his view of what it is we learn, and how we learn it, with reference to L2 learning. As to the first matter, he takes a Chomskian view of language competence / knowledge which leads to his (radical) definition of knowledge. It follows that what W&I offers is skills training that doesn’t affect impicit (linguistic) knowledge.

        Those who reject the generative view would allow that, in principle anyway, but depending on how and when (as Hana says) the kind of practice that W&I offers can drive forward interlanguage development.

      4. hi Geoff
        yeah WP reply function is a bit cumbersome.

        i think the skill aquisitionist bods would be amongst the non-generative lot that would say “that [what] W&I offers can drive forward interlanguage development.”

        but others e.g. the frequentist bods would say that the general cog mechanisms are implicit and so what W&I offers would not drive forward interlanguage development?


    1. Hi Mura,

      Yes and no! “Text book rules” is a loaded term and depending on how you “unpack” it, some (e.g. Schmidt with his “Noticing hypothesis”) would say that explicit knowledge can – and does – affect implicit knowledge and can affect interlanguage development. So the non-generative bods would not all agree that W&I is not changing learners’ knowledge – unless, of course, we accept VP’s definition of knowledge.

      1. Sorry, I’ve replied to this in the wrong place. May I take the chance to say that I agree with Hana.

      2. hi Geoff
        yes i think VanPatten argues that so far all the studies showing effect of explicit teaching/knowledge is due to tests measuring explicit knowledge : )

  3. hi Geoff
    yes i see, though i think for his point about knowledge being not about “text book rules” the non-generative camp would also agree that such traditional rules have no psychological basis;
    so W&I and similar tools on VanPatten’s definition of knowledge (and non-generative bods too) is not actually changing learner’s knowledge; they may help learners with writing (or other) skills

    1. This is an interesting and important discussion. I’d just like to point out that the quote under discussion was made in a presentation about a spaced-repetition-based vocabulary learning app, and not about Write & Improve. I’m not aware that anyone has claimed that W&I increases declarative knowledge. Certainly, the emphasis is on skills. It offers a safe and supportive practice environment that aims to increase exposure and confidence and demonstrate the benefits of experimentation, self-editing and reflection and revision and trying again. It is hoped it will help with increasing awareness of fossilised errors, but conclusions from longitudinal studies are some way off.

      1. Hi Diane
        thanks for clarifying what W&I claims;
        the first quote is from that vocab app report; the second is a claim made by ELTJam to machine learning in general;
        i used W&I to extend that implication, i hope readers see that that claim is mine which i did as a way to hook the post onto the knowledge-skill claim by ELTJam;
        i’ll add a note as you are worried this is not clear : )

  4. I agree with Diane: this is an important discussion. The issue of explicit / implicit knowledge is, as Long (2017) says, “central to resolving many of the long-running debates in language teaching: What can best be achieved, or achieved at all, through incidental and/ or intentional language learning? ”

    Long. M. (2017) Instructed second language acquisition (ISLA): geopolitics, methodological issues, and some major research questions. Instructed Second Language, 1.1.

Penny for your thoughts

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s