Potential and future features of the prototype

User data potential

There are numerous ways in which the data generated by the user's learning behavior could be further used to represent his learning profile and allow the app to adapt with more accuracy in ways that improve language acquisition. Examples of data usage could be values (as well as their averages and standard deviations) such as :

• Time spent on vocab mode.

• Rate of new words (per hour, day, month, session).

• Number of words typed.

• Number of characters typed.

They can be compared by:

• Input speed.

• Error rate.

• Change error rate.

• Change in learn score.

Change in total learn score by day divided by numbers of characters typed could be a key performance indicator (KPI) for efficiency. If this KPI would decrease, that might be because the mental energy of the user is used up (learning in small chunks, not big ones, is a solution to this). There might be a user specific optimum for this KPI, which might even change over time. The app could calculate this, base its adaptation decision on it and also show it to the user.

Moreover, users could be compared by their behaviors and characteristics, which might reveal clustering around learning behavior profiles. In another example, learning (eg. retention) rates for a word could be compared across users, revealing their difficulty.

Comparing user learn data could be used to adjust the app features and adaptation algorithms. Design decisions for teaching methods could be evaluated based on data and experiments (such as A / B split tests). The scalability and accuracy of such an infrastructure would be far higher than conventional methods and accelerate the development of ever more optimal learning and teaching methods. Future works thus should focus on finding meaningful relationships in user data.

A crucial feature for future work: Virtual classrooms

Virtual Classrooms would enable teachers, or anyone, to create classrooms (which might be better named "learning groups"). The teacher could create a classroom for a group of students. He could then give the class vocab lists that would be converted into lessons and he could check and monitor the learning behavior of the students. This would make it easier to see if the class is too easy for some students and if others struggle. This also would allow for an optimal way to synchronize the material of the offline course with that of the online app. However, at the same time the users should be able to maintain their desired degree of privacy by controlling how much the teacher can see about them and selectively hiding some behaviors. In turn, it might be good for the teacher to have the ability to require to see certain behaviors so that the teacher can do his job. Finding the balance here is not trivial and it might be necessary to find it by a case by case basis. While sharing of learning data between teacher and user is vital, the user needs to be able to practice without having external pressure in mind. In such a system, tests might become redundant as, over time, the app can accurately determine the user's language competency by analyzing their learning data. Moreover, a traditional class might not even be as much required as it is now, because users could simply browse for online classes. These online classes might have skill requirements that the user needs to achieve in the app by himself or in other classes within the app. Once the student has joined, a teacher can monitor and help the user without ever meeting him in person. While the social component of a physical classroom offers benefits, this outlook is not mutually exclusive and at the same time is much more scalable and reduces barriers.

Implications for traditional teaching and language classrooms

This degree of supplementation for the early stage of L2 acquisition has three benefits: First, such a system is much more scalable as students are independent from their access to a language class or teacher and can decide when to start and continue their language education. Second, teachers could focus on what they are best at: That is, once the student has reached an intermediate level, to provide him with a social component, and offer exercises based on communication and meaning such as role playing games. Teachers, freed from explaining basics, could focus more on taking on the role of a pedagogical coach. In other words, teachers can focus more on the application of language. The third benefit is that the rich learning data created by the software usage allows for classes to be more flexible and tailored to their attendees. Users could be grouped into skill groups, based on which words they know how well. Another possibility is to require a certain skill level: Users would simply have to learn a required vocab list with the software and this would give them access to take a slot in a class that they want to attend. This would mean that the attendees of a class have the sufficient skill level to participate and that the class would also have a higher probability to be of use. This would also allow for class sizes that suited to their needs and appropriate for their levels of advancement. Finally, teachers and classes could be evaluated with the corresponding student data. This would not only show what works best, but it could also protect the users from suboptimal teaching methods. While data would enable all this, it should not dictate everything by itself. Instead, experts should use it as an aid for decisions. Such experts might make mistakes, so transparency, where it would be appropriate, should be provided.