A voice based companion for practicing real world
conversations in new languages.

Winner of student design competition | UXPA Intl. 2018

Sound wave gif

Problem space

Engaging in real world conversations is recognised as an important and difficult part of learning a new language.


Chattie is a voice-based companion for language learners to practice real-world conversations and improve their speaking proficiency.

April - June 2018

Pallavi Benawri, Puhe Liang, Sonali Tandon, Ruchi Ookalkar

View poster submission for UXPA 2018
View article in UX Matters


My role

This was a self-initiated side project wherein I identified language learning as a possible domain for voice-based intervention. I was deeply involved in every step of the process, from research and ideation to user testing. I took charge of drafting the initial overall architecture for Chattie, considering high-level structure, main features and omnipresent background commands, and conducted two small user tests with the initial conversation dialog. Prior to usability testing, along with my teammate, I used Google text-to-speech software to record snippets of the conversation dialog. In the Wizard-of-Oz usability tests, I acted as a notetaker in the first test and the moderator in the second test. 


What are the problems faced by language learners?

Comparative Analysis

To understand the gaps, we compared features of current physical and digital language learning solutions, which include Duolingo, Memrise, Speechace, Spotify, Coursera, Google Translate, and an in-person language class. We found that there was an overwhelming gap surrounding speaking in a new language. Current tools provide reading and writing practice but speaking practice, translation from other languages, and detecting incorrect pronunciation remain unaddressed by most existing solutions. 



We conducted a survey with 148 respondents to understand people's motivation behind starting and continuing learning a new language, the most difficult parts of learning a new language, and tools/ features people found helpful in their learning.

How do owners interact with their voice assistants?


Through our initial research, we identified voice intervention as a potential solution to address the lack of speaking practice. To decide whether this solution would be viable in the language learning context, we conducted four interviews with voice assistant owners who have previously attempted to learn a language. This exercise helped get deeper insights into language learning habits, and analyse how people currently interact with their voice assistants. We created an affinity map to evaluate the interviews.

Affinity mapping


Who are the target users?


Our research helped us identify the set of users we wanted to target, which is intermediate language learners, i.e. learners who know basics of the language but wish to improve their proficiency.


Designing the conversation flow

Hey, let’s talk food.

We conducted a fun exercise where we talked to people about a selected topic (say, food) with the same initial question but no further guidance. The natural conversation flow helped us understand how people talk about food and guided our first draft of the script. We iterated over this conversation dialog by doing multiple small user tests which helped us identify the major gaps and account for them. The insights from our research helped us define the features of the product and the overall information architecture of the VUI.


Wizard of Oz usability testing

Once we had designed the features and the script, we tested them with two users through wizard of oz tests. We used Google text-to-speech software to record and play Chattie’s part of the conversation in response to the user. The tests revealed that users thought of Chattie as a teacher but wanted it to be more of a “friend”. We then made changes such as adding ad-libs like “That was a great answer!”, and “That dish sounds yummy!” to make Chattie more encouraging. In addition to this, we also added more edge cases and created fallbacks for unscripted responses.

Screen Shot 2018-09-22 at 3.29.51 AM.png
Screen Shot 2018-09-22 at 3.28.57 AM.png


This was a self-initiated project and my first step into the domain of Voice User Experience. With VUI penetration in households increasing significantly every year, the scope and opportunities are tremendous. With the field being relatively new, the limitation on available resources was a challenge. However, Cathy Pearl’s ‘Designing for Voice User Interfaces’, along with Google and Amazon’s guidelines for voice based design served as wonderful guides for our process.