What does it take to design language-learning experiences that feel as lively and practical? Silvia Place-Hildesheim, Learning Content Expert at Babbel, knows the answer better than most. As a key expert on the team behind Babbel Speak, Silvia helps shape authentic, engaging, and culturally relevant scenarios.
In this interview, Silvia shares how she approaches her work, from crafting realistic conversational tasks to balancing AI flexibility with proven teaching methods. Whether you’re a language learner, an educator, or simply curious about the magic behind Babbel Speak, her perspective offers a fascinating look at how technology and pedagogy come together to help people speak confidently in real life.
This is the second interview in our Babbel Speak series. To dive deeper into how Babbel Speak began and what inspired the feature, check out our conversation with Franz Ardito here.
Now – let’s begin.
Let’s start with a quick introduction. Can you tell us a bit about yourself and your role at Babbel?
Sure! I am a former adult education English teacher turned curriculum designer. At Babbel, I create content for learning activities and advise product teams on best practices in language teaching and learning. Before Babbel, I was a classroom teacher working with adult English learners in the US.
I’m very much still a teacher at heart, and I love working on features that can help students progress and feel more confident. Language learning is powerful and emotional; As a teacher and language learner myself, I’m always chasing those magic moments when we realize that we can connect and communicate with others in the language we’re learning.
I have worked on Babbel Speak since spring 2023 and was involved in many prototypes and experiments to determine the features included in the experience. Later, I supported with content development and content scaling.
How do you design Babbel Speak scenarios to ensure learners practice target vocabulary and/or grammar authentically?
The base of each Babbel speak scenario is a communicative situation, a real-life context in which we use language, and a Babbel CanDo, a specific language learning target adapted from the Common European Framework of Reference for Languages (CEFR). For example:
- Cando: exchange simple likes and dislikes
- Communicative situation: two friends are deciding what to eat for dinner. They suggest different options and talk about what they like and don’t like.
This base is supplemented by other guiding instructions for the Large Language Model (LLM). Together this is reviewed by experts in the Learning Content team, ensuring that target vocabulary and grammar is practiced within meaningful real-life settings. With this design, learners can practice engaging and targeted conversations that mimic how we use language in real life, making it easier to use and transfer these skills outside the app, which is the ultimate goal!
Can you share an example of how you incorporate real-life nuances into conversation practice, and explain why that’s important for language learning?
We have a great example of this for one of our Italian conversations: “Choose ingredients for a pizza”. The grammar focus is on talking about your likes and dislikes, as well as stating preferences.
My colleagues Gianluca Pedrotti (Principal Learning Content Creator) and Carolina Esposito (Senior Learning Content Quality Manager) embedded this language practice in a typical Italian situation, the pizzata (think “pizza party” or “pizza evening”), a social event centered around making and enjoying pizza together. This scenario gives our learners a glimpse of Italian life while giving space to practice target vocabulary and grammar. The scenario highlights the collaborative aspects of being together when cooking and preparing food in Italy.
We’ve also experienced some funny conversations in this scenario discussing the merits of fruit on pizza! At first glance, this can be fun for our Italian learners who are interested in traveling to Italy. On a deeper level, we are learning more about Italian life and culture. Language is not used in a vacuum. Considering the cultural context when creating scenarios is essential because it results in conversations that are more authentic and help people feel connected to the language, the people, and the culture.
How do you design authentic, real-world scenarios while meeting the pedagogical requirement to target specific learning objectives?
I don’t believe authenticity and level appropriateness are necessarily mutually exclusive, but we do need to make sure scenarios are accessible and manageable for beginners. A big way we do this is by focusing scenarios and providing guiding tasks.
The tasks are 2-3 goals we want you to do in the conversation. They break the practice down into manageable chunks and reduce cognitive load. We spent time making sure that these tasks made sense in the context of the communicative situation AND were clearly aligned to the underlying CanDo. In this way, we worked to create scenarios which are life-like, accessible to our learners, and help them progress and become more confident to speak a new language.
You mentioned supporting discovery by providing and testing new prompts. How do you evaluate whether an AI prompt will create effective learning experiences for our users?
This is a big part of my current role! We have developed a comprehensive list of criteria for what we want the AI output to look like. The criteria were developed out of best practices in language teaching and learning. To create the current Speak experience, it took many iterations of prompt development as well as working closely with the engineers in Team Asimov, the team working on Babbel Speak, to consider not only what information is in the prompts, but how the prompts link together to produce output that consistently delivers against our criteria.
This is an ongoing process for us. Through our various prompt iterations, we have also been able to learn and test the current limitations of LLM generated content. Now that we have identified some of the current limitations, we are working to improve the quality of our conversational experiences even more.
How do you balance the AI’s conversational flexibility with the structured learning outcomes that effective language pedagogy requires?
This is a good question! One thing we learned very early on was that LLMs could be conversational, but weren’t necessarily good at leveling content or creating content at specific proficiency levels. Providing the level of the student in the prompt was not sufficient and produced conversational output, but often using vocabulary or structures that would result in cognitive overload for a beginner learner. Here the tasks also play a crucial role because they focus the scenario and give guidance to the LLM, but leave room for variety within the conversation.
We’re not telling the AI exactly what to say, but we are setting guardrails for the machine to understand what type of practice we are trying to create for the learners.
What do you enjoy most about working on this feature?
What I enjoy most is hearing learners’ stories and seeing others practicing with the feature. Our main goal with this activity is not to get you to say everything perfectly, but rather to just get started exactly where you are!
The fear of making mistakes can be difficult to overcome. So when I see others speaking and working towards the communicative goals, it is very exciting. I hope if our learners can get even a slight boost of confidence with this feature, they can take this feeling and try it outside the app.
