Preface
In this blogpost I outline what I believe to be a novel use of existing technologies to dramatically increase the useful content available to learners of a second language. I introduce myself then follow with an outline of some relevant developmental and educational principles. From there I analyse the system requirements and present the system, and finally take a look at the opportunities for future research and development.
For a more detailed look on the philosophical differences between my proposed system and existing tools, see my blogpost here.
Introduction
For my 40th birthday I decided to take a sabbatical year and tick off one of the big remaining items on my bucket list - learn Chinese. I had started learning Chinese during my Linguistics degree 20 years ago but fate took me to Europe and life as a software engineer, and not only did I not get very far with Chinese, I completely forgot what I had learnt all those years ago. The only character I could actively identify when I started learning again was "中". I was starting again from zero.Technology in general has advanced remarkably over the last 20 years, so this time I was going to make sure I was using all the tools I could find to make the task easier. Alas, with the exception of "connected" spaced repetition software and certain "gamified" smartphone/tablet apps (which I find horribly tacky), there is still nothing much available. Those tools that are available are very difficult to find, and I only found those I did after reading journal articles and conference proceedings! Large amounts of content are now available at the click of a mouse but finding interesting material at my level (a few hundred characters/words) is all but impossible. I spent a great deal of time trying to find content but of the little I could find, almost none of it appealed to me. There is an entire world of readily available Chinese language content out there that I am very interested in (technology, history, scifi/fantasy, food, travel, hiking,...) but almost all of it inaccessible to me.
I am currently enrolled in full-time study at Yuxi Normal University in Yuxi, Yunnan, China. In addition to my classes, I have been using the open source Anki spaced repetition software for up to several hours per day to learn characters. I have chosen to focus on learning characters because I am a technophile, and China is a leader in all things digital. Getting properly "online" means being able to read well though, so the quicker I could read, the quicker I could "start livng" again. Unfortunately, my memory has degraded very significantly since I was a university student and after a while I realised that I was not going to be able to remember enough simply using Anki. Words that have been memorised with very little (class textbooks) or no context at all (words written on the board during class) are very quickly forgotten. In order to properly internalise vocabulary and grammar, they need to be seen multiple times in real-world content.
As I struggled along with hours of dubiously useful memorisation and very frustrating partial understanding of real-world content, a realisation came to me about the technology that is available to us and how to put it together to solve this problem.
Developmental and educational concerns
Learning preferences
I have always been of the opinion that humans have different learning preferences, and that is Ok. Some scholars reject the idea of "different learning styles" but no one can deny that there are markedly different learning preferences. Some prefer learning a second language through chatting with new friends, some prefer listening to the radio or watching television, some prefer reading novels or comics, some learning songs, some reading news on the web... There are almost as many preferences as there are language learners, and apart from very basic vocabulary and grammar, there is no "right way" or "right order" to go about learning. You need to build up vocabulary and confidence using a range of grammatical constructions but whether you do that watching the latest (Chinese) scifi blockbuster or reading about the daily lives of Mary and Mike at the Beijing Language and Culture University is immaterial. You do need to consume significant quantities of real material though, and that requires significant amounts of time and effort.Motivation
A great deal has been written about the importance of motivation in education in general and motivation in L2 contexts in particular. Zolt'n Dörnyei (1998) provides a useful, if slightly dated, overview of the many complexities involved in attempting to understand motivation in the various common L2 contexts. While there may be no universal agreement on what motivation is, there is virtually universal agreement on its importance for learning a second language. Many learners fail to achieve their (or their parents') goals because the effort required is simply too much. Learning language in traditional contexts can be very dry and learners interests change over time. Reading about how Anna would like Mary to accompany her to buy a new feather down jacket might have interested you at 13 but no longer does at 16, or 40.One of my grandmothers wrote a successful series of books for learning to read (English) for primary schoolers and she was always very vocal about her opinion on the role of teachers. A teacher is not there to "teach" but to guide learners as they follow their passions and discover the world. If you force them to focus on material they are not interested in or that is not at an appropriate level, then learners will quickly lose interest and educational outcomes suffer. The "motivation problem" becomes irrelevant if learners are able to follow their passions - they are intrinsically interested in what they are passionate about!
L2 Motivation
In the context of L2, the fact that there is a "motivation problem" at all today is rather strange. At least for all the major languages (which probably corresponds to 90% of formal L2), there is such a wealth of material available digitally that it is very unlikely that there is *nothing* available on a subject a learner is passionate about. Again, I argue it doesn't matter what it is, particularly if you are consuming enough of it. If someone is deeply passionate about and expert in Chicago "gangsta rap", then they will be learning a lot of non-standard grammar and idioms. Native speakers from New York to New Zealand will still understand, and using someone's passion as a starting point for learning more standard forms is easy. Using examples from the non-standard form to illustrate how it differs from the standard can also give a much deeper understanding of the acceptable standard forms. And standard forms (newspaper and magazine articles, etc.) are also produced about rappers. They can be read and analysed in a much more meaningful way by a passionate learner than reading about how Mary's friends came over to her dorm room on her birthday and gave her a toy dog.The knowledge gap
One of the main issues a learner immediately faces when they attempt to consume content they are actually interested in is an often significant gap in what they know and what they require to understand/enjoy the content. With a dictionary and possibly a translated version of the text, most of the meanings can be obtained. The problem being that when the percentage of unknown words goes over a certain threshold, the time taken to achieve any sort of understanding reduces pleasure (and so motivation) to zero very quickly. Applications exist to get much faster translations when consuming digital media, such as dictionary add-ons for e-readers, and translation plugins for the web. Google translate can translate an entire page to use as a reference, at the cost of changing tabs and finding the sentence you were reading in the translation. But over a certain threshold these too become too cumbersome to use for a prolonged period. Learners can and do use these tools, and a good deal of learning happens in just this way. It is far from optimised for maintaining maximum motivation though.The ZPD
In the early part of the 20th century the Soviet developmental psychology scholar Lev Vygotsky put forward the idea of the Zone of Proximal Development (ZPD). The basic idea is that the optimal environment for learners is when they are in a zone where they are unable to perform a learning task on their own without help but with a little guidance from a "helper" (usually an older peer, teacher, etc.), they can perform the task themselves. They are challenged but not too much. Vygotsky and many subsequent scholars focus a great deal on the role and importance of the "helper" but I believe the idea can be generalised, particularly in today's technology-rich environment. The "scaffolding" a language learner needs is enough help to be able to understand and continue consuming content in a fluid and enjoyable manner. What needs to be provided is sufficient help to narrow the knowledge gap to a point where the learner can be challenged but not so much that the task becomes too onerous.The System
Requirements
So what would a system need to do to "solve" these problems? It must be:- Easy to use. It must integrate naturally into the normal daily activities of the learner and be trivial to use and understand.
- Omnipresent. With the significant amounts of time required to properly internalise grammar and vocabulary, it should be everywhere the learner is consuming content. With smartphones and other connected devices, that is now everywhere, at all times and on all devices. The content I am consuming for pleasure should also be connected into the tools I am using in more formal learning contexts in a natural way - there should be no boundary between the learning contexts.
- Digital. Digital media consumption is rising rapidly and an increasing proportion of real content is now only available digitally. This tendency will only increase as the content consumption patterns of today's younger learners suggest.
- Adaptive. In order to keep the user optimally engaged the system should evolve as the user does transparently and effortlessly. As the user learns new words, the system should be continually removing the help it provides to make sure the learner is getting maximal learning value from all contact with the second language.
- Universal. Because peoples' interests and passions are legion, the system needs to be available for all content available in the second language.
As a nerd, I do what nerds often do when they are faced with difficult real-life problems - I think of a technical solution. Sometimes those solutions turn out to require a super-computing cluster or things that haven't been invented yet, sometimes they are very simple. I was very confident that not only did all the bits exist but also that I could relatively quickly and easily put them together to make something useful. So I gave it a try, and now 3 weeks later, I have a working prototype.
Presentation
The idea came to me as I was watching a Chinese fantasy cartoon series (画江湖之侠岚). Because the speech and subtitles of the series has too many new words to consume in real time, I was constantly pausing the video to read the Chinese-only subtitles and then look up words when I couldn't get enough help from the context of the video. It was very, very hard going but I did it for hours on end because I still enjoyed it. But I needed far too many translations to get any sort of flow. I didn't want or need for the subtitles to be in English, what I *really* needed were translations (glosses) just *for the words I didn't know*. Now the state of my memory meant that if I were to try and learn a new word/character without putting it in Anki, I would soon forget it. That meant (means) that I have a pretty accurate digital representation of the entirety of my (lexical) knowledge of Chinese. And then it clicked - Anki knows everything that I do, so the remaining words are words I don't know.The system is actually very "simple", considering the technologies widely available today. The following four images contain all of the technologies required.
Ankidroid
From my class textbook
Bing translate (the programming API is much richer than Google's)
A browser plugin translation tool (+add-to-Anki), Chrome ODH
Because at heart Anki is just a database, and databases can be put on servers with APIs in front of them, this information can be made available on any connected device. It can also be updated from any connected device. All that I needed to be able to get good learning value from *all* content in Chinese was to have that content enriched with help for just the words I don't know. I can get help from dictionaries when reading but if that help is embedded directly into the texts without any action required on my part then the quantity and speed with which I consume content would explode. All I needed was to find the best context-based translation for the words I don't know and gloss those words directly in the text.
This almost totally incomprehensible Xinhua news article
Becomes this
Sometimes the translations aren't that good or the grammatical construction is too difficult. I already needed to parse the sentences to identify the words, and find translations for the sentences and individual words, so it was obvious to make all that information available with a single click. The system's goal is simply to reduce the knowledge gap between what I know now and what I need to know to enjoy the content. It is still challenging but I get precisely the help I need in order to be able to succeed by myself. It is adaptive because I can easily tell the system - with one more click - that in fact I do know this word. Or a single click to indicate I have forgotten it and it needs to be bumped to the top of the queue for learning in Anki.
So what exactly have I built?
I have a server component that receives a piece of text, parses the text into words with parts of speech information, gets translations for the whole text and dictionary definitions for the individual words (both "known" and unknown) and then sends an enriched representation back. In order to be able to actually use it, I also developed a browser extension/plugin that means I can simply click on a button in the browser and the whole page I'm looking at will be enriched. Because not everyone has an Anki database with the entirety of their knowledge, I started with a much reduced Anki database (HSK2) for testing. I indicate I know a word by clicking on the word (or gloss) and then clicking on the "I know this word button" that appears with the definition. The gloss is then removed from the rest of the text immediately and the next time I see that word on a new page, it won't be glossed. It is added to Anki with the pinyin and definition from the translation provider for later revision (normal Anki). On the same principle I can indicate I want to actively learn a word I think is interesting or required (for whatever reason) - if I know it in the context of this page it can be added as a new word to Anki but removed from the current text, or programmed to appear for revision later.This is already a useful tool for enabling me to consume more content. Research (see Laufer, 1989; Hu and Nation, 2000; Schmitt et al., 2011 - cited in Lee, Lam and Jiang 2016) suggests more than 2-5% unknown words means learners will not usually understand a text - I can now consume texts with well over 50% unknown words and still get some flow of enjoyment. That doesn't correspond to all Chinese content yet, but it will very soon. Because of the way language works - the most common words appear very often - almost any text will never have less than about 50% of the most common 1000-2000 words. It is very far from perfect but it is fit for purpose - it means I can get learning value from reading virtually any Chinese text available digitally.
For certain technical reasons I won't go into here the server part is currently only running on my laptop, so I can't use on all my devices yet. It is also still pretty inefficient and will likely need to be completely rewritten for better translations, security and scalability. If anyone is interested, I'll write a post on the technical details and add a link here, with my plans for the server side going forward.
Opportunities, drawbacks and future research
I never studied Natural Language Processing and only have experience processing text using basic regular expressions from previous jobs. Before 3 weeks ago I didn't even know whether it was possible to find "the best translation for a given word in the context of a phrase/sentence" using current technology. It is over 10 years since I last did any javascript development and my knowledge of how it works ended with Internet Explorer 6. A web page is also one of the most challenging environments to do this - interacting javascript and styles on pages that were not meant to be interferred with is a much more difficult task than enriching normal text in, say, an e-reader. It is a hacky prototype but I have something of great value to me with less than 3 weeks (of admittedly *very* long days :-)) development time. What could be achieved with more time and resources? A lot of my ideas for further features might not turn turn to be good ones, or only be good for advanced learners, but I have so many of them that surely some will be good!I developed a browser extension that could be used on both laptops/desktops and mobile browsers (using Firefox at least). I really want to see whether I am able to watch Chinese movies with glossed Chinese subtitles so I will develop an enriched subtitles plugin for a media app (probably VLC) next. The smartphone streaming music platforms here in China have a sort of "karaoke" screen where the lyrics are highlighted as they are sung - that will need to be attempted at some point too. Developing a plugin for an e-reader app will probably be one of the easier plugins to develop and my Chinese will hopefully have developed enough over the next few months for me to start reading novels with the help of this.
Once the basic principle of constituting an evolving database of what is known by the learner and then making that information available via the network is set as the foundation, the opportunities this presents are almost endless. When we add that to the realisation that what is most important for bringing the content learners are most engaged with to them is not what they know but what they *don't know*, then we can create a powerful tool for all learners by bridging that knowledge gap.
If the system is available everywhere where learners consume content, then it will likely be used everywhere. It is a tool to allow the learner to consume any content on their own terms, with exactly the help they need at this precise moment in time. While I'm obviously developing this for my own learning of Chinese, there is no reason this can't be used to great effect for learning *any* language. Some studies suggest there are over a billion people learning English. That's a lot.
What I find most exciting about this idea is that if we take this information and make it available not only to the user but also to others, then I believe we can begin to start changing the way second languages are learnt. If we have databases with the words that individuals know, and what they are required to know in the context of a formal course, then we can also plug into search engines to go out and find content that the user is interested in that also aligns with formal learning goals. Integrating grammatical constructions and somehow using those is also definitely possible, though how is best achieved will likely need some formal research.
If the system is used by thousands or even millions of users then it will be highly economical to develop more active learning devices, such as questions on what is going on in a movie or book, particularly for the most popular content. A single teacher could develop a set of questions for say, the latest Star Wars movie, which could then be used by the 10s of millions of language learners that will watch it. All integrated into the applications they would use to watch it anyway via plugins, extensions or native integration, as they are watching it. The system could also auto-generate questions tailored to a given level. While this would be a little dangerous to present directly to language learners, language teachers could very quickly validate/reject hundreds of auto-generated questions. These could then be available for every learner.
The principle can also be expanded to those who are not actively learning a language but are regularly consuming content in a language that is not their strongest written language. Hundreds of millions of people around the world regularly consume content in English - for both work and pleasure - and many do not understand everything they want or need to. As the better you know a language the less of an impact this system will have on the presentation of the content, it could also be integrated directly into email applications, PDF viewers - everywhere. If a manager or client knows the manuals or emails will be better understood because the reader will get exactly the help they need with zero extra effort, that is significant value for a lot of people.
Reading is only part of learning a language, and I haven't mentioned the rest yet. A truly revolutionary system would integrate everything - speech recognition and pronunciation help, virtual classrooms with learners with common interests, automated examinations using the learner's preferred content, fully personalised search engines,... This list is endless. I strongly believe that because language is everywhere, the tool should be too. And this is not a strange idea - the biggest tech companies are all rushing to be present at every step of our lives. Today that is to better advertise to us but *exactly* the same technology could be used to help us learn and live in second language.
Existing tools
Conclusion
An idea that has deeply affected my thinking around technology that I first discovered in 2002 and was later reminded of in late 2009. The first occasion was presented by a researcher from a US university working with their Department of Defense on digitally-enhanced helmets for their fighter pilots - the goal was to increase the "bandwidth to the brain". If a pilot can get sensor information directly in their helmet exactly when they need it, then they are able to fly much better. Pranav Mistry's "Sixth Sense" is another version of the idea - more generally we can provide huge value to individuals by using technology to augment their experiences with information useful at that precise moment in time.I believe the system described above does this in a simple yet intuitive way for L2 learners with textual content. Because the user can interact with the system wherever they consume content that can be augmented with text, any effort they put in to training the system is paid back everywhere they consume content.
I have always intended to do a PhD around Linguistics (the other big checkbox remaining to be filled on my bucket list!) and am very interested in looking for opportunities to see just how far we can go with this. I believe the basic principle is solid but how we turn it into an amazing learning tool will require a lot of research and testing. How do we accelerate the constitution of the knowledge database without requiring new advanced users to click many thousands of times on words they already know? The best context for answering these and many other questions is probably a research programme at a university.
Commentaires
Enregistrer un commentaire