The Language Odyssey: From Montreal to Medellín
For me, language learning has always been very personal. I was raised in an English-speaking suburb of Montreal, initially hesitant to speak French before finally embracing it as an adult. Decades later, while living in Boston, a friend’s invitation to join a beginner Spanish class sparked a learning odyssey that eventually led me to Medellín, Colombia, and to my wife. Today, the journey continues as we study Italian together.
If you’re a regular reader, you know I am a huge proponent of using AI to enhance language learning. We have entered an extraordinary new age where learning has expanded into vibrant, multimodal experiences. A robust ecosystem now exists, from Duolingo Max’s AI characters to NotebookLM’s ability to synthesize complex study systems.
What sets these modern tools apart is their ability to combine modalities—pairing a conversation with an image of a café or providing instant feedback on a spoken sentence. This creates a tight loop of action → feedback → adjustment that is ideal for skill acquisition. Unlike a recorded video or a static textbook, AI can immediately highlight a grammar error or prompt you to retry a mispronounced word. It allows for authentic communication and cultural exploration through accessible platforms that are free or cost less than a single private tutoring session.
Duolingo and the Cognitive Plateau
Duolingo has been a pillar of my practice for over a decade. Its gamified environment is fun and addictive, making daily practice engaging. For beginners (A1/A2) or low-intermediate learners (B1), it is remarkably effective for vocabulary and listening. The addition of conversational avatars like "Lily" provides much-needed speaking opportunities. But it has a critical limitation.
Duolingo is designed to keep you engaged, not to make you think hard.
Duolingo prioritizes retention over rigor. Its speech recognition is embarrassingly “permissive,” accepting garbled pronunciation or incomplete sentences just to keep your “streak” alive. If you want to move beyond a low-intermediate level, you’ll need to go beyond the “game” and embrace the “struggle.”
As language teachers will tell you: if you want to speak a language, you have to speak the language. My goal is to converse with native speakers on a wide range of topics, and it requires a level of cognitive tension that Duolingo does not demand.
Tailoring ChatGPT Voice
The evolution of ChatGPT from a text-based chatbot to an interactive voicebot—ChatGPT Voice—is the most exciting development in this space. Initially, I used it simply as a casual conversation partner, asking questions and seeking general feedback. It was fun, but it wasn’t terribly efficient.
I realized I wasn’t thinking carefully about how to prompt the AI to provide structured speaking activities. I needed a way to home in on specific weaknesses and create a progression of difficulty. This led me to experiment with practical yet challenging voice exercises designed to push me past a plateau.
In this post, I’ll delve into these specific strategies. While there is no “one” right way to learn, the goal remains the same: to move beyond conventional methods and plateaus. Here is how I use ChatGPT Voice to bridge that gap.
1) Timed “Sprints”
One of the biggest challenges in language acquisition is retrieval speed. In a real-world conversation with a stranger in Medellín, I only have a second or two to respond before the “window of opportunity” closes. Even when my confidence is high, I know that specific triggers—regional expressions or complex verb tenses—can easily trip me up.
To try and combat this, I use ChatGPT Voice for Timed Sprints. These are basically high-intensity drills designed to make a specific grammatical response “more automatic.”
The Setup: Conquering the Subjunctive Let’s take the (notoriously annoying) subjunctive. Instead of just “chatting” about it, I built a targeted drill. I provide ChatGPT with specific parameters:
The Task: Ask me 10 short questions requiring my use of the subjunctive.
The Constraint: Focus exclusively on -AR verbs in the present subjunctive.
The Pressure: Allow only four seconds for my response.
The Progression: “Laddering Up” I usually begin with a “B1 target” to find my rhythm. Once I’m in a groove with -AR verbs, I’ll rotate through -ER and -IR endings, and finally move to irregulars. To increase the cognitive load, I’ll ask ChatGPT to drop the response window to only two seconds.
If I’m feeling confident, I’ll “ladder up” to a B2 level. I’ll prompt the AI for a new series of questions requiring the Imperfect Subjunctive, and force variety by switching between different pronouns (nosotros, ellos, etc.).
Why It Works By the end of a twenty-minute session, I have a clear map of my hesitations. I can identify exactly which structures cause hesitation in my speech and refine my next sprint accordingly. It’s a low-frills, high-impact method for sharpening responsiveness and ensuring that when I step out into the streetsz my “internal processor” is ready for the speed of real life.
2) “Listen → Chunk → Retell” Mini-Stories
My Italian is currently at that frustrating “low intermediate” stage. I can follow a conversation fairly well, but my own speaking is clunky. I know the words when I hear them, but they don’t always “come out” easily when it’s my turn to talk. To bridge this gap, I use a mini-story routine to connect my listening directly to my speaking.
The Setup I start by asking ChatGPT to tell me a short story (about 60 to 90 seconds) using vocabulary from a topic I’m currently studying. For example, if I’m focusing on “the home,” I want to hear words like cucina (kitchen), soggiorno (living room), or soffitto (ceiling) used in a real context rather than just seeing them on a flashcard.
The Process Instead of trying to digest the whole story at once—which can be overwhelming—I have ChatGPT break it down into “chunks.”
Step 1: AI tells a small piece of the story (after telling me the whole story)
Step 2: I retell that specific piece in my own words. I don’t worry about being perfect; I just focus on being clear.
Step 3: Once we’ve finished all the chunks, I do a full “retell” of the entire story in about a minute.
Why It Works I like this exercise because it takes the pressure off. It breaks the story into stages, so I don’t have to “get” everything immediately. It helps the vocabulary stick because I’m hearing a word and then immediately using it in a sentence myself. It’s one of the fastest ways I’ve found to move a word from “I recognize that” to “I can say that” list.
3) Role-Play with “Constraint Cards”
Role-playing in a new language can be stressful, and when I’m stressed, I tend to revert to “safe language.” I’ll use the simple present tense, basic nouns, and predictable phrases just to get through the interaction. It’s functional, but it doesn’t lead to improvement.
To push myself out of this comfort zone, I use “Constraint Cards,” tiny sets of rules that force me to use specific vocabulary and grammar, even during a simple role-play.
The Setup I typically pick a scenario that is personally relevant. Lately, I’ve been role-playing calls to Italian real estate agents (I admit it, I spend a lot of time daydreaming about living in Italy). While the base conversation is pretty simple, I give myself a list of constraints to make it more of a workout:
Grammar Goals: Use vorrei (I would like) and potrebbe (could you) at least twice.
Interaction Rules: Ask at least one clarification question during every turn.
Target Vocabulary: Incorporate specific terms like piano (floor), lavanderia (laundry room), and dimensioni (dimensions).
The Safety Net: If I get stuck, I must ask “Come si dice...?” (How do you say...?) rather than reverting to English.
The Feedback Loop After about six or eight turns, I ask ChatGPT for a quick check-up. I tell it to identify my most “egregious” errors and suggest just one or two “upgrade” phrases that are firmly at the B1 level.
The Second Pass Then, I do the exact same role-play again. This time, I raise the stakes. I’ll look at actual real estate listings on an Italian site like casa.it and try to weave in specific details from a house I’ve seen. (Too much daydreaming!) This second round is a big challenge, but I think it;s where the real learning is ha. I’m not just “chatting”; I’m identifying useful speaking patterns and linking them to my real-world interests.
Another structured option: role-plays in Talkpal or Langua
ChatGPT Voice doesn’t always follow directions and can slip out of character. That’s annoying. So, when I want a pre-structured “simulation” feel, I turn to the role-play features in apps like Talkpal or Langua. I like the range of topics—travel, ordering food, small talk, work situations, problem-solving scenarios, even awkward moments like fighting with a friend. That variety forces me to handle different vocabulary sets and conversational actions (asking, clarifying, negotiating, etc.).
Two features make these role-plays especially valuable for speaking and listening growth:
Adjustable speaking speed of the avatar. I can slow the avatar down when I’m learning a new topic, then speed it up as I get comfortable.
Voice-only mode by hiding instant transcription. When I remove the instant transcription of a conversation, I listen intently. I’m focused on the goal to catch meaning through sound, not text.
I use Talkpal and Langua for “applied practice” sessions. In these, I’m not so worried about grammar as I am about being responsive and focused.
4) “The Transformer”
One of the hardest parts of sounding natural in Spanish is learning how to be flexible. In a real conversation, you’re constantly reshaping your thoughts: turning a statement into a question, shifting from the past to the future, or adding a “but” to a complex idea.
To practice this, I use an exercise I call The Transformer. It’s a fast-paced drill that forces me to manipulate high-level sentences under a time limit.
The Setup: I ask ChatGPT to provide a B2-level (high-intermediate) “opinionated sentence” about a specific topic, such as inteligencia artificial. I want something with a bit of meat on the bone, beyond “AI is good,” but a statement with some nuance.
For example, the AI might give me:
“La inteligencia artificial no es una moda pasajera: es una herramienta poderosa que, bien usada, puede aumentar la productividad.” > (AI is not a passing fad; it is a powerful tool that, used well, can increase productivity.)
The Challenge Once I have the sentence, I read it out loud to practice speaking and get the rhythm. Then, I give myself five seconds to transform it in three specific ways:
Change the tense: (e.g., move it to the future or the past).
Turn it into a question: (e.g., “Is AI really a powerful tool...?”)
Add a conflicting idea: (e.g., “...however, it also presents significant risks to privacy.”)
I’ll aim to do this for about five to seven different statements on a single topic.
Why It Works I like this exercise because it mimics the “mental gymnastics” required in real life. It’s one thing to understand a complex sentence when you read it and it’s another thing to pull it apart and put it back together under time pressure. It strengthens my listening and vocabulary, but more importantly, it builds the “muscle memory” needed to stay smooth when a conversation takes an unexpected turn.
Adopting a “training” mindset
Adopting a “training” mindset with ChatGPT Voice has changed how I approach my conversations in Spanish and my evenings of studying Italian. It doesn’t happen every day, but moving away from passive app engagement toward more structured, intense activities is helping me bridge the gap between knowing a language and actually speaking it.
To me, so much is about building confidence under pressure. I feel better prepared for the messy, unpredictable nature of real-world conversation. ChatGPT Voice serves essentially as my personal laboratory. It’s a place where I can make “egregious” mistakes, refine my patterns, and sharpen my responsiveness.
Finally, no matter how hard the language activity, I’m having fun. I love learning a new language, even if I’m not great at it. Perseverance and focus can take you far.
Part 2 of this series will appear next week.