Skip to content

Learning Environment Analysis: Duolingo

As an avid Duolingo user, it seemed like a natural subject on which I could perform a learning environment analysis. However, Duolingo is constantly running experiments (Aprameya, 2020) to find ways of improving their content, user retention, interface, etc., so doing an analysis of it is to analyze something that’s always changing in one way or another.

To start the analysis, I looked up Duolingo in Wikipedia (“Duolingo,” 2020) to learn general information about the company. Duolingo is a language learning software platform with over 300 million users worldwide (Lardinois, 2018). Tabulating the numbers of courses and the users for each course, as reported on the company’s website, I found 111 language courses available in 28 different languages, inclusive of 13 language courses currently in development (“Hatching” in Duolingo parlance) (Duolingo, 2020). English language Duolingo has the largest number of language courses available with 36. In contrast, English is also the most widely studied language on the platform, with over 95 million users studying it in 27 different languages (See Appendix).

Duolingo is available on the web, and as apps for tablets and smartphones using the iOS and Android platforms. Depending on how it is used, it can be either Computer-Assisted Language Learning (CALL) or Mobile-Assisted Language Learning (MALL), but regardless of how it’s used, Duolingo is primarily a very sophisticated and elaborate type of Spaced-Repetition Software (SRS). Its courses are available for free, though supported by ads, and they have a premium tier (Duolingo Plus) which removes the ads and allows some additional features, such as being able to download lessons.

Duolingo is also available to schools (Duolingo, 2020) where it describes itself as “…the perfect blended learning companion…” for classrooms. A blended learning environment is one that includes in-person instruction with asynchronous instruction activities (Schukei, 2020). For purposes of this paper, Duolingo for Schools was not analyzed in order to concentrate on the standard version.

Learning environment description

Since all of the courses use the same software, there are a lot of similarities between them, however there are important differences that should also be noted. Courses that have been on the platform a long time (e.g., Spanish for English speakers, French for English Speakers) are much more sophisticated than courses that are relatively new on the platform (e.g., Finnish for English Speakers). The older courses have more tools available for learners, such as Stories and podcasts, while newer courses typically lack these things.

All Duolingo courses consist of a skill tree, which is a hierarchical collection of skills. In Duolingo, a skill is a collection of lessons (Duolingo Wiki, 2020), and each lesson is a collection of exercises. The lessons in each skill are also organized into levels, where the subsequent levels are meant to reinforce the content by having the learner repeat the exercises. There are five levels for each skill, and after completing all lessons on all five levels, a skill is considered to be mastered, though Duolingo also gives learners the opportunity to review skills that are getting stale. Learners who have some familiarity with a certain language, can take a placement test rather than starting at the very beginning of a skill tree. Based on how well they do in the test, Duolingo’s software will place them where it feels is most appropriate.

The skill tree is a scaffold, as learners have to complete one skill before moving onto another. In some spots on the skill tree, there is a certain amount of flexibility where two or three skills are placed alongside each other. The learners will have to complete each of the skills before moving on, but they are able to choose the order in which they learn those skills.

Duolingo can also be considered a gamified learning environment, as learners can earn crowns, lingots/gems, and Experience Points (XP) by completing lessons. Duolingo also tracks the number of consecutive days a learner has completed lessons, and encourages them to continue their “streak”. Learners can earn achievements by doing lessons late at night, or early in the morning, by learning 2000 new words in a single course, by completing 100 lessons with no mistakes, and so on.

Learners on Duolingo are largely self-motivated and self-paced. They can set daily goals in their profile settings, raging from a Basic goal of 1 XP per day to an Intense goal of 50 XP per day. However, they can add “friends” in Duolingo and follow their progress and compete with them. Duolingo also tries to encourage learners by placing them into leagues, where they compete with 49 other people to score the most XP on a weekly basis. The ten highest scoring learners will advance to the next highest league, while the ten lowest scoring learners will be relegated to the next lowest league.

Duolingo is targeted at language learners of all ages, with no consideration apparently given to age, race, sex, gender identity, religion, national origin, or socioeconomic background. It is directed at people who are literate in one of the 28 different languages used in its interface.

Task/activity description

While most Duolingo courses start with exercises that teach common words and simple phrases (e.g., “I am a teacher,” “The dog is big,” etc.), courses teaching languages that don’t use the Latin alphabet (i.e., Arabic, Chinese, Greek, Hindi, etc.) start by teaching the language’s alphabet equivalent, or it’s most common characters.

In courses like Russian and Greek, which have alphabets of their own, the first lessons are devoted to teaching these characters, how they are pronounced, and their English language equivalent (when the course is being taught in the English language Duolingo). In Japanese, learners are initially taught the Hiragana writing system, which is a syllabary. In the Chinese language course, where the characters can represent a sound and/or an idea, the early lessons teach commonly used single characters, which are gradually increased into two- or three-character compounds.

Most exercises have a form of hard-scaffolding, where the learner can tap on or hover their mouse cursor over a word to reveal its English language equivalent. Some exercises take this a step further and provides translations for short, common phrases (e.g., “I am…”, “You have…”, “They like to…”, etc.)

Additionally, most exercises are meant to test the learners ability to read, write, or listen to the target language, but not all language courses have the speech recognition ability that would enable learners to work on their speaking skills.

A typical reading/listening exercise in Duolingo will provide the learner with a written sentence with accompanying audio clip, and the user has to translate it from the target language into English using a keyboard. A typical listening exercise may include just an audio clip, and the learner is expected to provide the textual version in the target language. A typical reading exercise may be a cloze sentence in the target language where the learner has to choose the correct word from a multiple choice list.

In Duolingo, every lesson is an assessment as every exercise could be considered test question. Advancing to subsequent skills depends on successful completion of prior skills.

Since Duolingo is rather large and prone to experimentation, it’s hard to state what instructional strategy it uses, since it probably uses many. In response to a question in a forum post (latinofluency, 2013), one of its engineers wrote about the language acquisition theories they use in its development, namely: active recall, spaced repetition, core vocabulary, and multi-modal learning. However, I haven’t been able to find anything explicitly stated about what instructional strategies are employed. Overall, Duolingo seems to be a mix of learner-centered and knowledge-centered approaches. Learners choose what language they want to learn, when they practice, and how often they practice, but they are limited to whatever topics are covered by the lessons. Duolingo does provide encouragement to learners, such as sending emails and push notifications to remind people to do their lessons (Duolingo, 2019). Despite this, all lessons are designed to convey knowledge, and if a student is unable to acquire this knowledge, then they will be unable to progress in the course.

Learning environment analysis

Duolingo supports collaboration to a certain extent. Every exercise lets learners comment on it, so they can ask for more information, such as why their answer was incorrect, or to get more in-depth information from native speakers. On the Duolingo website, some languages have forums, where learners can interact with each other and with native speakers, though this feature is missing from the smartphone versions of Duolingo.

The original purpose of Duolingo was to take documents in one language, and use language learners to provide crowd-sourced translation (“Duolingo,” 2020), but this was eventually removed. Since then, Duolingo has tried adding other ways of providing authentic content, such as the Stories feature.

Stories are only available in some language courses, and when they are available, they are set apart from the skill tree. The stories themselves involve dialogue between two or more people conversing in the target language. Each story is provided as text and as audio. The language is more colloquial than those in the lessons, and at the end of the story, the learner is asked a series of questions to see if they understood. The stories are very engaging and cover a variety of situations, but they are sort of generic. Having gone through Stories in the French and Spanish courses, I noticed the stories use the same storylines, but are just translated into different languages. Since the stories are therefore generic, it’s hard to argue that they are culturally authentic.

As an example of CALL and MALL, Duolingo’s resources are mostly aligned to their audience of language learners, who want the ability to learn a language in a method that’s convenient. That said, not all resources in a language course are available to all learners. Some resources, such as forums and vocabulary lists, depend on using the web-based version.

It doesn’t support the learner in defining meaning because its exercises only accept a limited number of translated possibilities. If learners don’t understand the meaning from the exercises, they are unlikely to progress. At this time, it’s impossible to add soft-scaffolding into Duolingo’s lessons due to the fact that they are self-paced and taught online instead of a classroom. However, Duolingo uses artificial intelligence (AI) for different things, so it’s possible they may develop some sort of AI-driven, real-time coaching for language learners in the future.

Duolingo Italian for English speakers course: Le mie scarpe sono elettriche? translates as Are my shoes electric?
Figure 1: Duolingo Italian for English sacrificing authenticity for humor.

Task/activity analysis

The lessons and exercises are mostly appropriate for a wide range of language learners, though it sometimes includes grammatically-correct sentences that are meant to be funny as opposed to authentic (See Figure 1). While these sentences can liven up an otherwise banal course, they can’t be considered as authentic.

Using the example provided by the image at left, if an English-speaking Australian visited Venice, Italy and asked, “Le mie scarpe sono elettriche?” they would likely be met with confused looks from the native Italian speakers.

Despite these novelty sentences, most of the lessons are appropriate for the skill level of the learner. Once a set of lessons are completed on a given level, the learners are automatically advanced to the next level.

Learners can collaborate with each other by commenting on individual exercises, and if they are using the web-based version, they can use the forums to interact with other learners, so the opportunities for collaboration and/or reflection exist, but are limited.

Courses in Duolingo can be created by the company’s staff or by the contributor community, so each course will have numerous people working on it, presumably looking at the course from different points-of-view. Even after the courses are created, they are occasionally updated to include new content.

Grounded design

It’s difficult to say whether Duolingo uses grounded theory in creating courses. While they are open about using data to help them do this, with millions of users, one would assume they are using
quantitative data, whereas grounded theory considers qualitative data (Schroth, 2019). However, as some have observed, “Quantity has a quality all its own.” (Barth, 2015)

One part of the Duolingo website includes a corporate blog where they write about different topics, and some of these relate to the process and thought that go into creating a course. However, I suspect they do a lot of research that’s proprietary, and isn’t shared with anyone outside the company. Given more time, it would be interesting to do a literature review for papers written by people known to work at Duolingo to see if the company contributes to the understanding of language acquisition, or language education in general.

Although it wasn’t analyzed for this paper, Duolingo for Schools, as part of a blended-learning environment, would have opportunities for gather qualitative data required in grounded design.

References

Aprameya, L. (2020, January 10). Improving Duolingo, one experiment at a time [Online learning]. Duolingo Blog. https://blog.duolingo.com/improving-duolingo-one-experiment-at-a-time/

Barth, N. (2015, September 11). Who said, “Quantity has a quality all its own”? – Quora. Quora. https://www.quora.com/Who-said-Quantity-has-a-quality-all-its-own?share=1

Duolingo. (2020). In Wikipedia. https://en.wikipedia.org/w/index.php?title=Duolingo&oldid=983994596

Duolingo. (2020, November 1). Duolingo for Schools [Online learning]. Duolingo for Schools. https://schools.duolingo.com/

Duolingo. (2019, April 1). Duolingo Push [Online learning]. Duolingo. https://push.duolingo.com

Duolingo. (2020, October 31). Free Language Courses—Duolingo [Online learning]. Duolingo. Retrieved October 31, 2020, from https://www.duolingo.com/courses/all

Duolingo Wiki. (2020, November 1). Skill [Fan created wiki]. Duolingo Wiki. https://duolingo.fandom.com/wiki/Skill

Lardinois, F. (2018, August 1). Duolingo hires its first chief marketing officer as active user numbers stagnate but revenue grows. TechCrunch. https://social.techcrunch.com/2018/08/01/duolingo-hires-itsfirst-chief-marketing-officer-as-active-user-numbers-stagnate/

latinofluency. (2013, July 10). What Language Acquisition Theory(s) Is Used in Duolingo? – Duolingo [Question and Answer]. Duolingo Forum. https://forum.duolingo.com/comment/579004/What-Language-Acquisition-Theory-s-Is-Used-in-Duolingo

Schroth, S. T., PhD. (2019). Grounded theory. In Salem Press Encyclopedia. Salem Press.

Schukei, A. (2020, August 28). Finding Success in a Blended Learning Environment [Educational blog]. The Art of Education University. https://theartofeducation.edu/2020/08/28/finding-success-in-ablended-learning-environment/

Appendix

Duolingo users in the thousands. Courses that are in development (“hatching”) are represented by eggs.

A spreadsheet of courses in Duolingo as of October 2020. Courses in development (or "hatching" in Duolingo parlance) are indicated as eggs.
Duolingo courses