AI as The Best Learning Partner

I Spent Years Researching How Children Learn to Read. Then AI Arrived — and I Rebuild My Entire ESL Methodology from Scratch.

Here is what the data taught me, what the classroom taught me, and why “Collaborate with AI” is now the most important thing I teach my English language learners.

By Dr. Shu Yu Sophia Huang | Educator · ESL Methodologist · Data Scientist · Performing Arts Producer

“The most dangerous thing you can tell an English language learner is that their mother tongue is a problem to be solved.”

I want to start with a student.

She was twelve years old. She is a typical underprivileged student in a small suburban city in Taiwan. Her Mandarin was textbook-standard but her English was almost nonexistent. She was bright, curious, socially perceptive in ways that would have impressed any teacher — and she was completely silent in my classroom. Not shy. Silent. She had learned, somewhere between high and low social economic status, that the safest strategy was to produce nothing rather than produce something wrong.

I know that silence. I have seen it in hundreds of students. It is not a language problem. It is a confidence problem wearing the costume of a language problem — and almost every bad ESL methodology I have ever encountered makes it worse.

That student is why I do this work. That silence is what I am trying to break.

Who I Am — and Why Every Piece of My Background Informs How I Teach English

I am an educator with a Doctor of Education (Ed.D.) in Learning and Instruction, from the University of San Francisco. My 2008 doctoral dissertation — “The Influence of Kindergarten and First-Grade Literacy Instruction on the 3rd- and 5th-Grade Students’ Reading Achievement: Findings from the Early Childhood Longitudinal Study, Kindergarten Class 1998–1999” — used hierarchical linear modeling on data from over 20,000 American children to trace the long-term effects of early literacy instruction. The question I was asking: does what happens in kindergarten actually shape reading achievement years later?

The answer was unambiguous: yes. Profoundly. And the type of instruction mattered as much as the fact of instruction.

I am also a data scientist and machine learning engineer, holding a Master’s degree in Machine Learning and Data Science. I understand how large language models are built, how they fail, how they encode the biases of their training data, and — most importantly for my classroom — how they can serve as tireless, patient, low-stakes conversational partners for language learners who need more speaking and writing practice than any human teacher can provide alone.

I hold an MFA in Film and have produced and acted in award-winning screen projects. What this taught me about language teaching is not obvious, but it is real: story is the oldest and most powerful delivery system for meaning that human beings have ever invented. Children do not acquire language from grammar charts. They acquire it from needing to say something, to someone, about something that matters to them.

And I am the founder and CEO of the International Performing Arts Consortium (IPAC), a nonprofit I launched in 2024 in Changhua, Taiwan — an organization built on the conviction that language barriers are a design failure, not an inevitability.

I tell you all of this not to list credentials. I tell you because every one of these identities lives inside my ESL classroom. The researcher in me demands evidence for every instructional choice. The data scientist in me builds AI-integrated systems that scale personalized feedback beyond what any one teacher can deliver. The filmmaker in me knows that a student who is moved by a story will remember the language of that story far longer than any vocabulary list. And the performing arts producer in me has watched, night after night, what happens when language stops being a barrier — when a person who has been sitting outside the story finally gets to be inside it.

That is what I am trying to build in every ESL classroom I touch.

What the Research Told Me — and What It Didn’t

When I was designing my doctoral research, the “reading wars” were in full swing. Phonics advocates and whole language advocates had been fighting for decades, each convinced the other was failing children. Policy lurched back and forth. Teachers were exhausted by the pendulum. And English language learners — who faced the double cognitive challenge of acquiring a new language and learning to read simultaneously — were largely an afterthought in a debate that had been designed around native English speakers.

My research tracked reading achievement from kindergarten through fifth grade. I modeled the relationship between instruction type in the earliest grades and reading outcomes years later. What I found was clear: children who received both systematic phonics instruction and rich whole language exposure — literature-based, meaning-driven, vocabulary-embedded — showed the strongest long-term reading growth trajectories. Neither approach alone was sufficient.

But here is what the data could not tell me, and what two decades of classroom practice has filled in:

For English language learners specifically, the question of how to teach reading is inseparable from the question of what relationship to English we are trying to create. A student who learns to decode English words without any connection to what those words mean in her life will become a decoder — technically proficient, emotionally disconnected, and unlikely to become a fluent, confident English speaker. A student who falls in love with English stories but cannot crack the spelling-to-sound code will hit a wall the moment the books get harder.

The data told me what to combine. The classroom taught me why the combination matters most when the learner is arriving from another language entirely.

The ESL Problem Nobody Talks About Enough: Identity

Before I get into methodology, I need to say something that most ESL curricula skip entirely.

Every student who enters an English language classroom is carrying a fully-formed linguistic and cultural identity. They know how to make people laugh. They know how to argue. They know how to express love and frustration and boredom and delight — in their first language. They are not empty vessels. They are complete people who have temporarily lost access to their full selves because they are operating in a language that doesn’t yet fit.

The worst thing an ESL teacher can do — and many do it without realizing — is communicate, explicitly or implicitly, that the goal of language learning is to leave the first language behind. That fluent English means sounding American, or British, or however the teacher pictures “correct.” That the mother tongue is interference, not resource.

This belief produces the silence I described at the beginning of this piece. Students who are afraid to be wrong produce nothing. Students who produce nothing do not acquire language. The methodology fails before it begins.

My foundational commitment in every classroom is this: your first language is not a problem. It is the scaffold on which we are going to build your second. Every piece of English we encounter, I want my students to connect back to what they already know — in Mandarin, in Taiwanese Hokkien, in Japanese, in whatever language they dream in. Contrastive analysis is not a remediation strategy. It is a gift: when a Mandarin speaker understands why English subject-verb agreement feels strange — because Mandarin doesn’t mark it the same way — that confusion becomes a window rather than a wall.

This is not just good pedagogy. It is respect. And respect is the single most powerful variable in language acquisition that no standardized test has ever managed to measure.

How I Actually Teach — The Method in Full

Let me be specific, because vague pedagogical philosophy helps no one.

1. Comprehensible Input — But Make It Compelling

Stephen Krashen’s comprehensible input hypothesis — the idea that learners acquire language when they encounter it at a level just slightly beyond what they can currently produce — is one of the most robustly supported ideas in second language acquisition research. I build my curriculum around it. But I add a condition that Krashen’s original formulation understates: the input also has to be interesting.

This sounds obvious. It is not how most ESL classrooms are run.

I choose texts that my students actually want to read. For a class of Taiwanese students in 2025, that might mean a short documentary transcript about competitive gaming, a Reddit thread about K-pop fandom culture, a New York Times article about Taiwanese street food going global. Not because I am trying to be cool. Because language acquisition requires sustained attention, and attention follows genuine interest. A student who wants to understand what she is reading will push through unfamiliar words. A student reading a textbook passage about a topic she has never cared about will stop at the first difficulty and never restart.

The phonics and decoding work — which I do teach explicitly, because my research showed it matters and because English orthography is genuinely irregular enough to require instruction — I embed inside this compelling content. We encounter an unfamiliar word in a passage we are reading because we want to understand it. We learn the phonological pattern because it helps us read the next unfamiliar word faster. Decoding is always in service of meaning. It is never the point of the lesson.

2. Lowering the Affective Filter — Every Single Day

Krashen also identified what he called the “affective filter” — the emotional barrier that goes up when a learner feels anxious, embarrassed, or unsafe, and which effectively blocks language acquisition even when comprehensible input is present. A student who is terrified of making mistakes in front of her classmates is not acquiring language in that moment, regardless of how well-designed the lesson is.

Every structural decision I make is in service of lowering that filter.

I never cold-call. Students volunteer or pass, always. I respond to student errors by recasting — modeling the correct form naturally within my response, without marking the error as an error. If a student says, “Yesterday I go to the market,” I say, “Oh, you went to the market! What did you find there?” The correction happens. The conversation continues. The student did not feel stupid. Research consistently shows that recasting, compared to explicit correction, produces better long-term accuracy without the psychological cost.

I use pair and small group work heavily, because the affective filter is dramatically lower when a student is talking to one peer in a low-stakes side conversation than when performing for an audience of thirty. The speaking and writing that happens in those small conversations is where real acquisition often occurs — students take risks they would never take publicly.

I also talk explicitly with my students about language anxiety — about how the discomfort they feel when they reach for an English word and cannot find it is a sign that learning is happening, not a sign that they are failing. This metacognitive reframing matters more than most teachers realize. Students who understand their own acquisition process — who have a mental model of why language learning feels the way it feels — are less afraid of the discomfort and more willing to stay in it.

3. Pushed Output — Supported, Required, and Never Punished

Merrill Swain’s output hypothesis established what classroom teachers have known intuitively for a long time: learners don’t just acquire language by receiving input. They also acquire it by being required to produce — to speak or write in ways that push them slightly beyond their current comfort level. The moment of struggling to express a thought in English is itself a learning moment, as the brain searches for resources and, finding gaps, builds new connections.

I push output deliberately. Every lesson includes speaking tasks that require students to formulate and express their own ideas in English — not to fill in blanks, not to recite memorized phrases, but to actually say something they mean. A debate about whether a character made the right choice. An explanation of how something works in their own culture. A persuasive argument for a position they don’t personally hold.

The scaffold is temporary and transparent. I tell students: this sentence frame is a training wheel. We use it today. By the end of the month, I am going to ask you to throw it away and say it in your own words. That is the goal. And the goal is absolutely reachable — I have watched hundreds of students reach it.

4. Where AI Changes Everything — and Where It Doesn’t

I stopped treating AI as something my students might encounter someday. I started treating it as something they were already using, would always use, and needed to learn to work with skillfully.

AI changes ESL methodology in ways that are specific and real. I want to name them precisely, because the hype around AI in education tends to either oversell or completely miss the point.

The practice gap. The most persistent structural problem in ESL instruction is the ratio: one teacher, thirty students, forty-five minutes. Research is clear that language acquisition requires massive amounts of meaningful practice — far more than any classroom can provide. Students who have no opportunity to practice English outside school simply acquire it more slowly, full stop, regardless of how skilled their teacher is.

AI changes this equation. A student who finishes class and goes home to continue an English conversation with Claude — asking it to discuss the story we read, to role-play a scenario we practiced, to give feedback on her written draft — is accumulating practice hours that simply did not exist before. The AI does not tire. It does not judge. It is available at midnight before an exam. For language acquisition, where volume of meaningful practice correlates directly with proficiency gains, this is potentially transformative.

But only if the student has been taught to use it as a practice partner, not a ghostwriter.

Prompting as a literacy act. Writing a good English prompt for an AI is itself an act of language production that demands sophisticated skill. The student has to know what she wants to say, find the vocabulary to express it specifically, and structure the request in a way another intelligence can understand. When a student writes “help me with English” and gets back a vague, generic response, and I ask her to rewrite that prompt — to specify her proficiency level, her learning goal, the exact kind of practice she wants — she has just done real metalinguistic work. She has thought about her own language needs and expressed them precisely. That is the kind of learner autonomy that characterizes the most successful language acquirers.

I make prompting practice a formal part of my curriculum. We workshop prompts together. We compare the responses that different prompts produce. We analyze why one prompt got better output than another. This is ESL instruction and AI literacy in the same act.

Low-stakes speaking and writing practice. Students use AI for back-and-forth conversation practice — describing their week, defending an opinion, explaining a process — in ways I frame explicitly as low-stakes. The affective filter, which is so high when students must produce in front of peers, drops dramatically in private AI conversation. Students who have barely spoken in class will write paragraphs of real, effortful English when they know only the AI is reading. This is not a workaround. It is a feature. That low-stakes practice builds the confidence and fluency that eventually transfers back into the classroom.

The writing feedback loop. For writing development specifically, AI transforms the feedback cycle in ways that matter for acquisition. The traditional model — student writes, teacher marks, student revises — has a multi-day lag that is deeply suboptimal. By the time feedback arrives, the student’s attention has moved on and the cognitive opportunity for integration has partially closed. AI feedback is immediate. A student can write a paragraph, receive specific targeted feedback on vocabulary and syntax, revise in real time, and produce a third draft — all within a single class period, without consuming my attention for thirty students simultaneously.

I then spend my feedback time where it belongs: on the student’s ideas, her voice, her argument, her choice of what to say rather than just how she said it. The mechanical feedback goes to AI. The human feedback goes to meaning. This division is not a concession. It is a pedagogical clarification of what teachers are actually for.

AI ethics as ESL content. Here is where my approach becomes, I am told, distinctive: I teach AI ethics through English, not alongside it. The ethics content becomes the language lesson.

We read an AI-generated text in English and identify where its knowledge of Taiwan, East Asia, or Chinese-language culture is thin or simply wrong. We discuss, in English, why this might be. We write, in English, a correction or response. This is critical reading, cultural analysis, and authentic English writing practice simultaneously — and because students are genuinely invested in being accurately represented, the motivation to express themselves clearly is unusually high.

We read English news articles and fact-check AI summaries of them. We discuss AI hallucination in terms students can apply to their own information environment. We talk about bias — whose English counted in the training data, whose voices were centered, whose were marginal. For students in Taiwan, these are not abstract questions. They are questions about whether the most powerful AI tools in the world were built with any understanding of who they are and where they come from. That personal stake produces the most sophisticated English my students generate all year.

What AI cannot do — and why this matters for acquisition. I am direct with students about this, because the most damaging misuse of AI in language learning is using it to produce language the student didn’t create.

AI cannot give you the experience of searching for a word and finding it. It cannot give you the satisfaction of being understood in a language you built yourself. It cannot replicate the productive struggle of pushing to express your actual thought in an imperfect second language and discovering, in the pushing, that you can. These moments of difficulty — what Swain called “pushed output” — are precisely where acquisition happens at a neurological level. AI can practice with you. It cannot do the learning for you.

I make this distinction explicit and early, and then I give students plenty of legitimate, structured, purposeful ways to use AI as a genuine partner — so that the line between collaboration and shortcut is always clear, and the case for working honestly is always stronger than the temptation to let the machine do it.

From the Classroom to the World Stage: IPAC and What Theater Taught Me About Language Access

I need to tell you something that connects my ESL classroom to the performing arts organization I founded — because they are the same project.

Language is a bridge, not a barrier. And technology should serve human connection, not replace it.

In 2024, I founded the International Performing Arts Consortium (IPAC) in Changhua, Taiwan. IPAC makes live theater accessible to multilingual audiences, deaf and hard-of-hearing communities, and people with dyslexia through Storytelling Supertitles™ — a 3D AI-integrated subtitle system that embeds translations directly into the visual and spatial fabric of the performance itself, synchronized with lighting, set design, and live action.

The problem I was solving at IPAC is identical to the one I solve in my ESL classroom: people sitting outside the story because of a language barrier. The ESL student who goes silent because she is afraid of being wrong. The audience member at a Japanese classical theater performance who cannot follow the dialogue and slowly disengages. The design failure is the same. The solution — meeting people where they are and building a bridge toward the story — is the same.

IPAC has already produced cross-cultural collaborations including Kamada March Finale — Gin’s Departure and The Velveteen Rabbit with the ODC Dance Company of San Francisco and Taipei Royal Ballet, performed at the National Kaohsiung Center for the Arts (Weiwuying). In 2025, IPAC was named one of the “20 Most Innovative Companies to Watch” by Global Radiance Review.

I am more proud, though, when an audience member tells me — for the first time in their life — that they understood every word of a live performance. That feeling is exactly what I am building toward in the ESL classroom. Not translation. Not approximation. Full access to the whole story.

Why This Moment Is the Most Important One I Have Worked In

Right now is the most consequential moment for ESL methodology that I have witnessed in two decades of teaching and research.

Two forces are colliding simultaneously. The first is an unprecedented global demand for English language proficiency — English is the operating language of global science, business, and digital culture, and for young people across Asia and the developing world, English proficiency is increasingly a condition of full participation in the opportunities that matter most for their futures.

The second is the arrival of AI language tools that are genuinely, transformatively useful for language learners — and genuinely, seriously dangerous if used as a replacement for language learning rather than a support for it.

The ESL educators who meet this moment well will be those who can distinguish clearly between AI as a practice amplifier and AI as a crutch. Who can build AI collaboration skills — prompting, critical evaluation, AI ethics literacy — inside their existing methodology rather than as a separate add-on. Who understand that their students will use these tools for the rest of their lives, and that the question is not whether but how well.

That integration is what I am building. It is not finished. The tools keep changing and the students keep arriving with new questions. But the foundation is clear: every English language learner deserves a classroom that honors who they already are, gives them massive meaningful practice reaching toward who they want to become, and equips them to navigate an AI-shaped world with skill, judgment, and their own irreplaceable voice intact.

What I Am Building Next

I am expanding my AI-integrated ESL curriculum across Taiwan and the Asia-Pacific region, with particular focus on young learners for whom English proficiency is both an academic necessity and a gateway to global participation.

IPAC’s Storytelling Supertitles™ system is being developed for deployment in educational settings, museums, community events, and public spaces — anywhere that language has historically been the wall between a person and a story they deserved to be inside.

And I am writing, speaking, and building community with teachers who believe, as I do, that the most powerful thing we can do for an English language learner is refuse every false choice.

Phonics or whole language. Fluency or accuracy. AI tools or authentic language development. English or your mother tongue.

None of these are real choices. All of them are failures of imagination.

I am in the business of imagination.

And so, I hope, are you.