TL;DR Almost none of what follows is new. It is presented here to bring together various questions and suggestions in a single place.
Learning, teaching and problem solving
Learning and teaching, including their online versions, sit within the larger area of problem solving. At the most basic level, humans, like quite a few other species, have the ability to solve problems, including deciding among competing potential solutions. So, for example, faced with a stretch of water, a person can choose to walk around it, swim across, find or make a boat, turn back, or do something else. Problem-solving can be done by a single individual, but it frequently also occurs when a group works together to solve a problem, trying out strategies and refining them. Humans can also remember previous solutions and, importantly, share them with others or consult others for help.
Over time, this role of acquiring, retaining and transmitting knowledge has become specialized and in many societies assigned to humans known as teachers. And the role of acquiring knowledge has been assigned to humans known as learners. In the simplest case, a teacher puts one or more learners before a problem and either shows them a solution or guides them to find one themselves. Teaching and learning often happen synchronously, with teacher and learners all present, at the same time, at least virtually, in the case of online classes. However, constraints on resources, such as a limited number of teachers, have led to use of devices for sharing knowledge with learners asynchronously, with some device replacing the teacher. For centuries, and still today, the textbook has played this role. What is presented here represents, in part, an extension of the textbook model that takes advantage of computing technologies.
A textbook or other teaching material, including in-class teaching, can adopt any one of a number of approaches, of which we will focus on two here:
1) a concept is presented and illustrated with examples (the deductive approach);
2) examples or problems are presented and the learner or learners must work out the solution or find the pattern (the inductive approach).
Of course, the two can be combined, as in starting with a problem, showing the solution, giving more problems of increasing complexity and variety, having the learner or learners find the solutions, and perhaps reinforcing the concept again.
The teacher's role can be more or less robust. The teacher may:
1) guide the learner at every step of the way, as in showing a child how to write the letter 'a' by holding his or her hand;
2) let the learner do the work but intervene when he or she is going off the rails, as in showing how to add in arithmetic and gradually introducing the notion of how to carry 10's;
3) let the learner do the work and only evaluate the final product, pointing out then in more or less detail how well or badly the student has done (think essay writing and correction);
4) be entirely hands-off, allowing the learner or group of learners to both propose and evaluate solutions, debating perhaps within the group to come to consensus or perhaps disagreement (think in-class debates).
Both of these scales (presentation model and degree of intervention) are available in asynchronous online teaching and learning, and of course different approaches may be interspersed or offered in parallel to cater to different learning styles.
Humans are naturally curious, and sometimes the simple existence of a puzzle or problem is sufficient to spur them into action. For example, many people do a daily crossword puzzle for the simple pleasure of trying to solve it. On the other hand, some people are motivated by an external score, or by a ranking against others. Online learning environments should cater to both approaches. At the same time, it is important to recognize that different learners have different attention spans and online teaching materials should be explicit about their expectations. For example, a wordgame might be playable in five minutes, while a longer chapter might take several hours.
Sharing and privacy
Humans are social creatures and like to share their accomplishments. Online learning environments do not need to be synchronous, but they should allow for easy sharing of results as the learner sees fit. On the other hand, any familiarity with online sites makes it clear that a disturbing quantity of information about the user is often harvested, shared and sometimes sold. Online learning software should always respect privacy. In addition, access to the internet is not universally robust, so online materials should be light on the ground in terms of bandwidth and ideally downloadable so that a learner can work offline. Offline work can also support enhanced privacy.
A typical textbook has a built-in sequencing. Learners begin at the beginning and work through levels of increasing complexity. Of course, this is a gross oversimplification. Consider a former COBOL manual. At its beginning it had a table composed of rows (different subject areas) and columns (increasing complexity within a subject area). Learners can adopt a strategy of studying each subject area in complete detail, or of skimming all subject areas first. Similarly, online learning materials should provide a map which allows a learner to see what is needed to deal with a particular set of materials, but not a rigid constraint on what can be seen.
Materials creation and evaluation
Any experienced teacher will recognize that the two most time-consuming elements of teaching are the creation of teaching materials and marking. In the case of the former, the teacher needs to consider what will be interesting, how to scaffold it to ensure learning, and how to produce enough materials to avoid repetition. This last area is one where computational resources can be of assistance. For example, some of the materials on this site have been generated by a computer (see the description of the ivi/Vinci software).
At the other end of the spectrum, it is useful to distinguish evaluation of student work that is essentially mechanical, requiring little to no judgment on the part of the teacher, and evaluation (like reading an essay or thesis) that requires a complex of teacher judgments. At least in the former case, software can play a useful role. For example, the ivi/vinci software described elsewhere on this site has been used in online university courses to create quizzes and, using the error analysis mechanisms described below, mark them for classes of sometimes over a hundred students. In both cases, creation and evaluation, the underpinning is provided by the use of patterns.
Rules vs lists
As any computer scientist knows, there exists a correspondence between a rule and its output on the one hand and a list on the other. For example, the rule x => a + b + c generates the sequence abc, which also corresponds to the ordered list [a, b, c]. The same is true for language. For example, in English, we form the plural of many nouns by adding a final -s, as in cats. So an English speaker could either store all forms with plural -s, or a rule for making the plural. A famous test called the Wug test (based on the created word wug) shows that the rule is at least potientially available. If a speaker is given the sequence one wug, two ___, most will say wugs, even though the word wug is new to them.
The problem with rules though is that there may be competing rules. So, in English, alongside the rule that makes plurals in -s, there exists another rule whereby some words (like those ending in y) take a plural in -ies, as in daisies, ladies, parties and so on. So a learner must know both a rule for the forms in -s and another for the forms in -ies, or as language teachers are inclined to say, a list of exceptions. The boundary between rules and exceptions depends to some extent on the relative efficiency of the two approaches. Consider the French words matinée, fée, musée, mausolée ('morning', 'fairy', 'museum', 'mausoleum'). The first two are feminine (la matinée, la fée), while the second two are masculine (le musée, le mausolée). One can either treat all words in -eé as a list, with a gender for each word (memory-intensive), or one can note that the second two correspond to English words of Greek origin ending in -um (museum, mausoleum) (two rules). When modelling learning and teaching, one is faced with a similar choice. Does one define sets of rules for different forms, or does one simply use a list? In many cases, we have chosen to use sets of rules on the logic that learners can turn rules into lists, but that the reverse is harder.
Generation and filtering
Rules provide a powerful tool for modelling, but overgeneration must be avoided to prevent difficulties (like the production of daisys). To deal with this, a two-step approach is often useful. Generation is used to create long lists of potential examples, but these are then presented to a teacher to filter, so that only those judged to be both correct and useful are included in what is shown to learners. This is analogous to what many teachers do, sometimes on the fly, when they select some examples and reject others. Similarly, it is possible to produce a list of examples using many different rules, with results of different rules interspersed in the output, requiring learners to decide which rule applies. Alternatively, one can define many small grammars, each of which produces one set of outputs, showing the results of each to learners in increasing order of complexity. Of course, the boundary between the two approaches is porous and depends on the teacher and on the learners.
Consider the cats, daisies examples earlier. It is clear that the -s rule is more general than the -ies rule: the second applies only to words ending in y, while the first applies to words ending in many different letters. It is not surprising that learners will tend to replace a less general rule by a more general one. So we see mistaken forms like daisys and familys but not usually forms like caties. Let us use this to postulate what has been called a malrule to generate regular but mistaken forms like daisys (where the cat rule is applied to daisy). This approach makes it possible to produce a starting form: daisy, request a plural, and if the answer is daisies, know that it corresponds to the correct rule, but that if the answer is daisys, know that the learner has used a too-general rule.
Presentation of materials
A generative environment, when coupled with a modern web browser, permits a wide range of materials to be presented to a learner, including text, spoken language, images and configurations, and combinations of these. Which one is chosen for any particular area of study depends on the teacher's and learners' preferences and potentially on the level of difficulty. For example, in an online textbook on business French found elsewhere on this site, one of the exercises involves dragging the parts of a business letter into the correct order. If an item is dropped in the proper place, it remains; otherwise it reverts to its initial position. This allows the learner to focus on the high-level question of textual structure without having to worry about issues of grammar and spelling.
An online system allows for a variety of learner inputs of increasing cognitive complexity. At the simplest end of the spectrum we find flashcards, where a learner sees a question then the answer. Slightly more complex is the use of buttons to select one of a set of multiple choices. In a variant of this, the computer will speak the selected answer. At an even more complex level, a learner may reorder items or retype them (see Feedback below). An ongoing area of weakness is in the capture of learner speech. Because of issues of learner accent, software often fails to properly identify what a learner has said, at least in the case of complex utterances. On the other hand, we have had some success in exercises where a learner is asked to repeat an utterance and sees the intonation of the model and their attempt, with the potential to retry their speech to more closely approach the model.
We saw earlier that a teacher may provide a range of feedbacks ranging from immediate and constant to intermittent, to delayed or even absent. A generative system permits the same range of options. A learner can be shown no feedback, immediate feedback, including error analysis, pointing out the specific nature of the problem, or, sometimes more interestingly, a pointer to the area of difficulty with a request to resolve the problem. For example, in one of our exercises, a learner sees the phrase Elle ne pas voit la beau phrase. When he or she clicks on the red word, they are asked to drag it to the proper position. Some exploration will show them that this is after voit. When they click on the orange word, they are asked to retype it correctly, here belle to agree with phrase. The result is a correct sentence.
Over the course of a session, a learner will succeed with some things but not with others. Some learners wish to focus on their previous errors while some prefer to ignore them. It is important to provide mechanisms for both. In some of the systems on this site, the learner can choose to replay only those questions they have got wrong until all have been dealt with correctly. Logfiles provide an alternative overview of correct and incorrect responses. Both are optional, so those wishing to simply forge ahead can do so. Another alternative worth exploring is a dosage of previous errors within new forms.
Learners vary significantly in their preference for how materials are presented. Some prefer to see an initial concept while others prefer to explore first. In some of our online textbooks that have been used over a number of years now, the approach that seems to work best involves providing a very general overview followed by web-based exploration using mouseover, clicking, drag and drop, then more detailed discussion and finally exercises and quizzes.
On the other hand, granularity also depends on the subject. Some subjects, like learning how to read French, are large collections of interdependent terminologies and language skills, while others, like Morse code, are much more compact, involving mastering first the code for letters, then words, then sentences, either interpreting heard sequences or producing them.
It is also important to remember that humans like to play and that this is often an ancillary source of learning. Simple standalone activities like wordgames can often be useful way of taking advantage of this while surreptitiously teaching.
Tools and tool use
If one wants to use online language learning tools, the simplest solution is probably to purchase something readymade among the many commercial products available. The downside, as noted to us by one teacher who was involved in that approach, is that one has no control over the content or trajectory of what is available.
An alternative is to use course development tools as is done in many university remote learning departments. These permit the non-computer-savvy to enter, following instructions, the materials for simple questions, answers, quizzes and so on. Most provide as well a course management feature to keep track of grades, enrollments and so on. They provide more freedom than off-the-shelf software, but still limited options for exploration. In addition, most require a consistent good internet connection, often a luxury for remote or less-well-off users.
What we are proposing here is more complex. It lets users both use and create a variety of materials. This offers much freedom but it depends on the user (teacher or learner) knowing what he or she wants and deciding on the level of complexity they are willing to engage with. In its current form, this website includes the following elements:
2) sets of online games to encourage learners to explore the limits of their knowledge. For example, the ponderwords game presents, in English or French, a 12 letter word and asks a user to find within it two, three and up to eight letter words composed from the letters of the longer word.
3) larger full online textbooks including concepts, exploration, and so on.
The site also provides access to a natural language generation tool called ivi/Vinci. This is a piece of software, written in C, that runs on a computer (Mac, Windows, Linux). It is free for download. Once installed, the user enters a grammar, including a lexicon, grammar rules, parts of speech, question and answer formats, and the computer generates one or many products based on these rules. These may then be uploaded to a website for use. See the manual for details. The software is designed to be relatively easy to use, but a good background in linguistics is useful.
Form and meaning and metalanguage
There exist many approaches to the teaching of language, especially second languages, ranging from those which focus on form, attempting to make the elements of the language as salient as possible, to those which focus on communication, backgrounding form in favour of spontaneity and authenticity. Behind this spectrum lies an important distinction between two views of language: those which take meaning as the primary dimension and look at how it can be represented, and those which take form as their focus and ask how forms are used in a particular language.
Of the two, the latter has to date received the most attention. We would argue that there is value in exploring the former. Each language has its own way of slicing up the representation of reality, and grasping this is an important first step. For example, in French, events may be represented as backgrounded or secondary. At the same time, they may be seen as completed or not. And finally, they may be seen as conjectured or not. The combination of these three dimensions underlies the French tense system with its use of imparfait, passé composé, futur as well as the conditionnel. Learning how to use these forms is traditionally a challenge for English-speaking learners. Yet our experience teaching suggests that bringing learners to understand the dimensions of meaning behind their use gives them a much better chance of achieving more native-like proficiency.
Representing meaning, though, especially in the case of second language learning, raises the challenge of the metalanguage: how to represent meaning without using the form that carries it. One approach is to use another language. If this is the learner's native language, this brings the advantage that quite subtle distinctions can be made since learners can be supposed to master the language well. However, a disadvantage is that the second language to be learned becomes an object and less culturally real.
An alternative is to insist that everything happen in the second language. This is used in many so-called immersion contexts. An advantage of this approach is more authenticity, but the price is that learners are learning both language and metalanguage at the same time.
Observation of language teachers using this approach will often show that they complement their use of language with gestures, images and other embodiments. This provides a clue to an approach possible in online learning. We have found that images, sounds, graphical representations of meaning and simplified cartoon-like sequences can often help learners to see events without the filter of their first language. This approach has, in our opinion, enormous potential, especially in an online context where manipulation of multimedia is now easy to attain.
All of what has been written here provides an invitation for learners and teachers to explore the potential of tools that they themselves have made or adapted and that they control. This is not an easy task, but when complemented by sharing of resources, we believe that it gives greater power, especially to learners and communities that traditionally have not had it, as in the case of indigenous languages or less-studied dialects or linguistic phenomena.