Twee - AI Lesson Planning Has Advanced, But It Is Not Yet Fully Ready

Twee is an AI-powered tool designed specifically to assist language teachers in developing ready-to-use materials.

There are three reasons why I highly recommend it: its user-friendly interface, streamlining processes, and suitability for both school curriculums and self-directed lessons.

Compact Interface:

While it's often said not to judge a book by its cover, Twee's neat and visualised interface left a strong first impression on me. Meanwhile, as a language teacher I found it very pleasant because the tools it contains are what I am familiar with and would use regularly in my teaching, saving the time for me to figure out which tool to choose from excessive options from different disciplines that I'm not familiar with.

Twee's homepage

Twee's toolbar

Smooth Content Creation

Process pre-existed materials

Depending on what skills and knowledge the learners are asked to apply to do the tasks, all the tools are divided into six categories, namely reading, writing, speaking, listening, vocabulary and grammar.

Twee

Under the general categories, there are tools that generate exercises coherent to the conventional type of teaching that relies heavily on drills under the guidance of behaviorism learning theory. These exercises are close-ended reading comprehension questions, filling the gaps, matching, changing the part of speech and reordering the words. Though I didn't like them as a student because they couldn't excite me and I want to avoid them as I become a teacher, I gradually realise that they are unavoidable. It is because that they are the main question types in Gaokao (i.e., the college entrance examination in China), and sufficient attention should be given to cultivating students' familiarity with them. Twee seems to be well aware of the positioning of these exercises as well (see the description in the smaller font). 

Students are not the only group who suffer from these drills. As an English teacher, I often find myself investing valuable time crafting these materials, time that could be better spent designing captivating lesson plans or enhancing my professional development. The process of gathering exercises from various sources into a single editable document, or adapting document formats to suit different exercise types, is a tedious and time-consuming endeavour. It often entails mindless and repetitive typing, as well as extensive copying and pasting, which sounds not human enough for me.

Twee

Fortunately, I now have Twee to manage these "errands" for me. It can convert image, video and audio material into text, from which target collocations or words can be extracted. With the "raw material", users are allowed to take further steps to develop additional materials or exercises around the text, facilitating a streamlined process that minimizes repetitive copying and pasting. The results generated from each process are ready-to-use, reducing the time teachers spend on small but repetitive typing or adjusting the Layouts.

Exercises generated by Twee

The Generative "Magic"

In twee, users click "do the magic" button to run the processes. While you might initially feel that the functions I described earlier lack a sense of magic, Twee offers much more. For me, the real magic of Twee lies in its power to generate coherent and contextually relevant text based on the input it receives.

Create Extended Materials

Twee

Twee can effortlessly generate sentences, dialogues, and paragraphs using provided texts, target vocabulary, scenarios, and learner levels. These materials enable students to explore how target items are used in diverse contexts while maintaining a coherent theme. Producing such examples can be extremely time-consuming for teachers who are second language learners themselves, like me. However, Twee simplifies this process with just a snap. One small but potentially helpful detail is that the Twee's use of CEFR levels instead of grades for categorizing target learners. The latter often confuses me, for proficiency levels of students of different grade vary in different contexts. For my context where the learners are EFL learners, this criteria for categorising learners can be unreliable. Accordingly, using CEFR levels can better ensure consistency in assessing students' language abilities.

Customise One's Learning

With its generative power, Twee serves as a valuable resource for preparing for the IELTS test. In terms of the writing test it offers insights into crafting compelling essays by generating topic-specific vocabulary and constructing coherent arguments. Users can effortlessly generate topic vocabulary tailored to a given prompt, facilitating the development of well-rounded responses. Moreover, Twee can assist users in formulating argumentative essays by seamlessly integrating the generated vocabulary into their compositions. The same approach applies to preparing for the speaking test. Personally, I often struggle with using active phrasal verbs when telling a story in speaking test 2. Twee can be directed to gather active phrasal verbs for a specific topic and incorporate them into a story or dialogue.

Twee's project page

Some limitations:

Having mentioned many benefits of Twee, I shall now point out some areas that can be improved upon. To start off, some functions are deceptive as they appear impressive at first glance but ultimately fail to fulfil its intended purpose. For example, when we see the upgrade-simplify function, we would assume that it can simplify long sentences or to replace complex expressions with its simpler alternatives. However, in practice, it often results in cutting the information or even changing the meaning. So this function might not be very ideal for teachers who aspire to simplify a text to fit their students' language proficiency.

In addition, the free version of Twee is somewhat stingy in terms of the number of free outputs and the length of media it can process. A free account has only 20 text tools and 10 media tools every month. For the trial I did, the 30 chances might not be enough for preparing one lesson and it can be quickly used up within one day. It is quite generous, though, as it only restricts the number of chances rather than the function for free account.

Finally, though the streamlined process saves me a lot of time by reducing repetitive copying and pasting, it costs me troubles when trying to hide a certain content or reorder the content. The lack of features such as drag-and-drop reordering or the ability to hide content with multiple checkboxes makes these little tasks time-consuming for users. Implementing more intuitive features for content management would greatly enhance the user experience and efficiency of Twee.

Further Thoughts on AI Literacy

As EFL language teachers who are second learners ourselves, we might concern about whether the language generated by AI is natural or authentic enough to be used as teaching materials. Unfortunately, sometimes we can't tell that by ourselves. In China, many English teachers have been cultivated in the circumstance where the major input were textbooks using awkward or unnatural language tailored to specific vocabulary or grammar points. The good news is, there are reasons for EFL English teachers to trust the authenticity of AI-generated language. According to an article by Assembly AI, mainstream AI models like ChatGPT and InstructGPT are trained on human labellers' input, including their responses to prompts and ratings of the quality of the model's outputs. Therefore, the language generated by AI can be expected to reflect the preferences of the human labelers and the common guidelines they follow. This offers reassurance to teachers that AI-generated language is likely to be authentic.

A slide from Nicky Hockly speech on digital literacies and AI

Still, having a preliminary understanding of AI's mechanism doesn't mean that we can trust its answers blindly. In her session on AI literacy, Nicky Hockly introduced the "draw a scientist" experiment to us, indicating how AI's response maybe reflect stereotypes or systematic constraints people face, while encouraging us to always put a critical hat on, to think deeper about the ideolog(ies) underlying the technology. The aforementioned article also points out the limitations of the training model as it is inevitably influenced by a variety of subjective factors such as human labellers, developers, and customers' biases in choosing and creating prompts and responses. Therefore, it is highly relevant in this AI era for us and the future generations to cultivate a habit of critically examining the results generated by AI and keeping ourselves updated with the technological processes underpinning it. 

 My "draw a scientist" result from Copilot

With a critical mindset, I've inputted the same AI image generator with the same prompt "draw a scientist", and I got a collections of four scientists, all female. From here we can see just how unpredictable AI is and how important it is fir us to hold ourselves from making hasty conclusions about AI.