MLRegTest: Stress-Testing Whether Models Learn Rules or Just Patterns

When people say “AI understands language,” they usually mean it can produce fluent text, summarize an article, or answer questions. Those abilities are impressive, but they can also hide a real problem: a model can look correct while relying on shortcuts that break in the exact cases we care about most.

That is why I have been interested in MLRegTest, a benchmark designed to stress-test sequence models using 1,800 carefully constructed regular languages. Instead of judging a model by how human its writing sounds, MLRegTest asks a simpler, sharper question: can a model learn a rule, and then apply it reliably when the test gets harder or more precise?

What is MLRegTest, in plain terms?

MLRegTest is a large collection of tiny, made-up “languages” built from simple symbols. Imagine an alphabet like A, B, C, D, and strings such as “AAB C” or “BBB A.” Each language has a hidden rule that determines whether a string belongs to it. The model learns from labeled examples and then answers a yes or no question: does this string follow the rule?

This might sound far from English or Spanish, but it is actually a powerful way to test something very relevant to computational linguistics: how models represent patterns and dependencies across sequences.

Why regular languages?

Regular languages are a class of formal languages that can be described using tools like regular expressions and finite-state machines. They are simpler than full human language, but they still capture many meaningful pattern constraints. MLRegTest uses regular languages because they let researchers control the task in a way that is difficult with natural text. The rules are fully known, the labels are unambiguous, and researchers can generate unlimited data under controlled conditions. That makes it possible to test specific kinds of generalization rather than only measuring how well a model matches the distribution of a dataset.

What makes MLRegTest different from typical benchmarks?

First, MLRegTest is not just one dataset. It is a suite of datasets drawn from 1,800 distinct regular languages, and those languages are organized by properties such as logical complexity and the kinds of constraints they express. That organization matters because “pattern learning” is not a single ability. Some rules are easy to approximate but hard to learn exactly, and some require models to track information across long spans of a sequence. MLRegTest is designed to probe those differences rather than hiding them inside one average score.

Second, the benchmark is built to examine long-distance dependencies in a controlled way. Sequence models often struggle when the relevant information is far apart in the input, and MLRegTest gives researchers a systematic way to test whether a model can handle that challenge.

Third, MLRegTest includes a kind of evaluation highlighted in Stony Brook’s write-up: border tests. These focus on edge cases where examples come in near-identical pairs. The strings might differ by only one symbol, but one is in the language and the other is not. Those are the cases where the true rule matters most, and they are also where shortcut strategies are most likely to fail. According to the Stony Brook announcement, models tended to struggle more on these boundary cases, even when they looked strong on more typical examples, which suggests that they can learn approximations instead of learning the rule itself.

What did the researchers evaluate?

The JMLR paper evaluates multiple neural architectures, including recurrent models and transformers, and reports that performance varies significantly depending on the kind of test set, the class of language, and the model architecture. That is useful because it pushes back on the idea that “a strong model is strong at everything.” MLRegTest makes it easier to ask where a system is strong and where it breaks, and to tie those results to specific properties of the pattern being learned.

Why this matters for evaluating language models

Even though MLRegTest does not test natural language directly, it targets a core issue in NLP evaluation: benchmarks can be “won” for the wrong reasons. A model can score well by picking up statistical hints that correlate with labels without learning the intended generalization. Border tests and other controlled generalization tests help researchers ask whether a model stays consistent when inputs shift in principled ways, whether it generalizes beyond the training regime, and whether it fails exactly when the rule becomes tight. Those questions matter if we want models that are dependable in real settings, especially when rare edge cases are the dangerous ones.

A quiet challenge to “just feed it more data”

MLRegTest also pushes back on a common assumption in AI right now: if a model struggles, the fix is simply more data. The benchmark is asking what happens if the deeper issue is not data quantity, but what the model is actually learning. This is not only a scientific concern but also a practical one. In high-stakes applications like robotic medical assistance or self-driving cars, the most serious situations are often rare. A particular combination of weather, road design, sensor noise, and unpredictable human behavior might occur only one in a million times. In medicine, a rare complication might be exactly the case where you cannot afford a mistake. The border tests connect directly to this idea because they emphasize edge cases where a tiny change can flip the correct decision, which is where shortcut learning becomes most dangerous.

The takeaway is simple: reliability is not the same thing as average performance. If a system only works well on patterns it has seen thousands of times, it may still be fragile in the exact scenarios we care about most. MLRegTest is valuable because it helps us measure that fragility directly instead of waiting to discover it in the real world.

A high school senior takeaway

As a high school senior interested in computational linguistics research, MLRegTest feels like a strong example of what careful evaluation looks like. It controls the task so we know what the model should learn, varies difficulty in interpretable ways so “harder” actually means something specific, and probes failure modes instead of stopping at one headline number. More broadly, it connects to a theme I keep coming back to in NLP: we do not just want systems that perform well. We want systems whose performance we can explain and trust.

References

  1. van der Poel, Sam, et al. “MLRegTest: A Benchmark for the Machine Learning of Regular Languages.” Journal of Machine Learning Research, vol. 25, no. 283, 2024, pp. 1–45. https://www.jmlr.org/papers/v25/23-0518.html
  2. Stony Brook University AI news announcement (February 13, 2026): “How Much Does AI Really Understand: Stress-testing Neural Networks with 1,800 Language Patterns.” https://ai.stonybrook.edu/about-us/News/how-much-does-ai-really-understand-stress-testing-neural-networks-1800-language

— Andrew

5,088 hits

0

How AI is Quietly Changing the Way We Talk

Introduction

In this blog post, I’d like to share recent findings suggesting AI is quietly reshaping the way we talk. You may be already aware of the trend of AI’s reshaping the way we write since the wide usage of ChatGPT and other LLM models in facilitating the text/script generation, particularly in writing research papers. See the discussion in my past blog post “Is the Increasing Trend of Leveraging LLMs like ChatGPT in Writing Research Papers Concerning?”.

Florida State University’s Study

A new study from Florida State University shows that large language models are starting to influence spoken language, not just written text. Researchers analyzed over 22 million words from unscripted science and tech podcasts, comparing episodes from before ChatGPT (2019–2021) with episodes after its release (2023–2025).

They found that words commonly used by AI models, such as “delve,” “boast,” and “meticulous,” are showing up more often in everyday conversation, while their close synonyms stayed flat.

The researchers call this phenomenon “lexical seepage,” where AI-preferred words gradually leak into the way people naturally talk.

How the Shift Happens

The study links this effect to psychology concepts like implicit learning and priming. People pick up on repeated words, even without realizing it, and then use them themselves. In other words, AI is not just helping us write. It may also be subtly shaping the way we speak. Importantly, the changes were observed in unscripted talk, in addition to formal speeches or scripted lectures.

Global Patterns and Concerns

This is not only happening in the U.S. A study in Germany found similar patterns on YouTube, suggesting the trend is global. Experts warn that if companies like OpenAI, Anthropic, and Google fine-tune their models in different ways, people might start adopting slightly different speech patterns. Over time, this could flatten dialects, erase regional slang, and reduce creativity. Some argue we need new benchmarks that push AI to use more diverse language instead of over-relying on the same set of words.

Natural Adoption vs. AI Amplification

The Florida State team also makes an important point: not everything can be pinned on AI.

“It is possible that these words have simply entered a phase of natural, rapid adoption, akin to the rise of expressions like ‘touch base,’ ‘dude,’ and ‘awesome’ in the mid-2000s.”

In this view, LLMs overuse words that were already becoming popular, but they still act as amplifiers that speed up language change. Even if AI is not the original source of these trends, the fact that machine-generated text can influence how humans speak is significant.

Final Thoughts

As a high school student, I find this both fascinating and a little worrying. On the one hand, it shows how powerful AI really is in shaping culture, not just technology. On the other hand, if AI makes everyone talk the same way, that could erase some of the creativity and uniqueness that makes language fun. Just like with social media, the full impact may take years to understand. For now, I think it’s important to keep asking questions about how AI is changing not just what we write, but also what we say.

Further Reading

  • “AI Is Quietly Reshaping the Way We Talk.” Fast Company, https://www.fastcompany.com/91398460/ai-is-quietly-reshaping-the-way-we-talk.
  • Anderson, Bryce, Riley Galpin, and Tom S. Juzek. Model Misalignment and Language Change: Traces of AI-Associated Language in Unscripted Spoken English. arXiv, 2025, doi:10.48550/arXiv.2508.00238.
  • Yakura, Hiromu, et al. Empirical Evidence of Large Language Model’s Influence on Human Spoken Communication. arXiv, 2024, doi:10.48550/arXiv.2409.01754.

— Andrew

5,088 hits

0

LATAM-GPT, Linguistic Bias, and Why Regional AI Infrastructure Matters

When a language model answers in fluent Spanish but misses local context, the problem is not grammar. The problem is representation.

That is the central issue behind linguistic bias in GPT-style systems, and it is why LATAM-GPT is such an important project for computational linguistics researchers. It pushes us to ask a better question than “Can the model generate text?” We should be asking, “Whose language realities are represented in the model?”

Linguistic bias is bigger than offensive outputs

In NLP conversations, bias is often reduced to toxic or stereotypical responses. That matters, but it is only part of the picture. Linguistic bias also includes structural imbalance: which dialects are present in training data, which cultural contexts are understood, and which institutions or histories are treated as central versus peripheral.

For many GPT-like systems, the imbalance starts at the data level. English and Global North content dominate much of the public web, so model behavior tends to be stronger when prompts align with those distributions. A model may produce polished Spanish or Portuguese and still flatten regional variation, miss sociolinguistic nuance, or rely on generic interpretations that do not fit local usage. AP’s reporting on LATAM-GPT directly frames the initiative as a response to this representational gap in mainstream AI systems.

Why regional models matter

Regional models like LATAM-GPT are not only technical artifacts. They are research infrastructure choices.

First, they can improve local relevance because the model is trained with region-specific data and priorities rather than treated as a generic multilingual extension of a primarily external corpus. AP reports that LATAM-GPT was developed specifically to better reflect Latin American language and context.

Second, regional models help build scientific and governance capacity. Reuters describes LATAM-GPT as a collaborative effort among countries and institutions in the region, which means expertise, evaluation norms, and deployment decisions are not fully outsourced.

Third, the initiative is positioned as open infrastructure for downstream applications, not just as another chatbot interface. That distinction matters for public-interest work in education, government services, and domain-specific NLP tools.

What the LATAM-GPT project is

Based on AP’s report, supported by Reuters and the official project site, LATAM-GPT is a regional open-source initiative led by The National Center of Artificial Intelligence of Chile (CENIA). AP reports early backing that included funding from CENIA and the Development Bank of Latin America (CAF), and references future training support tied to a major supercomputing investment in northern Chile. Reuters also notes cloud support in the development process.

The project is collaborative by design. AP reports participation from more than 30 institutions across eight countries, while Reuters presents a broader regional coalition narrative around deployment and adoption. The reported training pipeline includes large-scale data, combining partnership-based sources and synthetic data to improve coverage in underrepresented areas. Initial focus is on Spanish and Portuguese, with plans to expand toward Indigenous languages.

The timeline is also important. AP describes work beginning in 2023, public visibility increasing at the 2025 AI Action Summit, and launch reporting in February 2026.

Performance versus ChatGPT and Gemini

This part needs careful wording.

AP quotes project leadership saying LATAM-GPT can be more accurate and efficient for Latin American and Caribbean contexts because of regional data alignment. That is a meaningful claim and it fits the project’s objective.

At the same time, both AP and Reuters frame LATAM-GPT as not primarily intended to replace ChatGPT or Gemini as general-purpose consumer assistants. It is presented as foundational infrastructure for regional applications. Public reporting so far does not provide a single standardized benchmark table showing universal superiority over frontier global models across all task categories.

So the most responsible interpretation is this: LATAM-GPT’s strength is regional alignment and representational fit, not blanket dominance across every benchmark.

What this implies for a junior computational linguistics researcher

For early-stage researchers, LATAM-GPT signals an important shift in what counts as strong NLP work. Bigger model size is no longer the only story. Research quality increasingly depends on whether your data curation, evaluation design, and error analysis capture real linguistic diversity.

That has practical consequences. If you only run generic leaderboard-style evaluations, you may miss the most consequential failures. Region-aware testing, dialect-sensitive prompts, and sociolinguistic error taxonomies become central methods, not side tasks. Corpus documentation and annotation policy choices also become core contributions, because they shape what the model can and cannot represent.

In other words, this is an opportunity. You can build technically rigorous work while also addressing linguistic equity and real-world usefulness. LATAM-GPT makes that path visible: computational linguistics can be both advanced and locally grounded.

Final reflection

LATAM-GPT matters because it reframes AI development from pure model competition to language representation, participation, and research sovereignty. The key question is not whether it outperforms every major global model on every task. The key question is whether communities that were historically underrepresented in AI can now help shape the systems that represent them.

For junior researchers, that is a powerful direction for the next decade of NLP.

References

  1. AP News. Chile launches open-source AI model designed for Latin America (Feb 2026).
  2. Reuters. Latin American countries to launch own AI model in September (Jun 17, 2025).
  3. LATAM-GPT official site (project overview).

— Andrew

5,088 hits

0

From Hallucinated Citations to Linked Evidence: The OpenScholar Approach

In my recent blog post, I discussed Citation Hallucinations at NeurIPS and What They Teach Us. As a student researcher, I think many people are asking the same question: can we use AI tools that help us get citations right without made-up references?

I recently read a Nature article that gave a strong answer. The article introduces OpenScholar, a retrieval-augmented system that combines a language model with a database of about 45 million open-access papers. Instead of relying only on model memory, OpenScholar retrieves papers first and then generates responses with explicit citation links.

Why this matters

For research workflows, citation reliability is everything. When references are wrong, the writing process breaks down quickly. OpenScholar is designed to reduce that risk by grounding claims in retrieved literature before generating the final response.

According to the article, OpenScholar is also:

  • Open source
  • Relatively lightweight
  • Deployable locally
  • Built for scientific search and literature review

That combination is important because it supports both accuracy and reproducibility, which are essential in research settings.

Reported performance

Nature reports that in the OpenScholar evaluations, the 8B model outperformed GPT-4o on correctness in their benchmark and significantly reduced fabricated citations. The article also notes that citation behavior was described as being comparable to human experts in their testing context.

Comparison with OpenAI deep research tools

The article places OpenScholar in a broader trend. Since OpenScholar was first posted on arXiv about 14 months ago, companies such as OpenAI have integrated similar retrieval-based “deep research” methods into commercial LLM products, improving factual accuracy and citation quality compared with earlier model behavior.

OpenScholar’s main distinction in that landscape is cost-efficiency plus openness. Nature cites the OpenScholar team saying it can run at a fraction of the cost of GPT-5 with deep research, while still grounding outputs in a large scientific corpus.

Limitations to keep in mind

The article is clear that OpenScholar is not perfect. The authors acknowledge two major limitations:

  1. It does not always retrieve the most representative or most relevant papers for every query.
  2. It is limited by the scope of its indexed database.

So even though OpenScholar helps with citation hallucinations, retrieval quality remains a core bottleneck. In practice, researchers still need to verify paper relevance and coverage before relying on output.

Final thoughts

My takeaway is that this is a meaningful step forward for student researchers and independent scholars. Better grounding, lower cost, and open access can make high-quality literature review tools more available to more people.

Nature also quotes an outside researcher who argues that if OpenScholar remains free, it could become one of the most widely used tools for scientific search. I think that is very possible.

If you have tested OpenScholar, share what worked and what did not. I may feature reader feedback in a follow-up post.

— Andrew

5,088 hits

0

Citation Hallucinations at NeurIPS and What They Teach Us

I’m writing this post about a recent discovery by GPTZero, reported by Shmatko et al. 2026. The finding sparked significant discussion across the research community (Goldman 2026). While hallucinations produced by large language models have been widely acknowledged, far less attention has been paid to hallucinations in citations. Even reviewers at top conferences such as NeurIPS failed to catch citation hallucination issues, showing how easily these errors can slip through existing academic safeguards.

For students and early-career researchers, this discovery should serve as a warning. AI tools can meaningfully improve research efficiency, especially during early-stage tasks like brainstorming, summarizing papers, or organizing a literature review. At the same time, these tools introduce new risks when they are treated as sources rather than assistants. Citation accuracy remains the responsibility of the researcher, not the model.

As a junior researcher, I have used AI tools such as ChatGPT to help with literature reviews in my own work. In practice, AI can make the initial stages of research much easier by surfacing themes, suggesting keywords, or summarizing large volumes of text. However, I have also seen how easily this convenience can introduce errors. Citation hallucinations are particularly dangerous because they often look plausible. A reference may appear to have a reasonable title, realistic authors, and a convincing venue, even though it does not actually exist. Unless each citation is verified, these errors can quietly make their way into drafts.

According to GPTZero, citation hallucinations tend to fall into several recurring patterns. One common issue is the combination or paraphrasing of titles, authors, or publication details from one or more real sources. Another is the outright fabrication of authors, titles, URLs, DOIs, or publication venues such as journals or conferences. A third pattern involves modifying real citations by extrapolating first names from initials, adding or dropping authors, or subtly paraphrasing titles in misleading ways. These kinds of errors are easy to overlook during review, particularly when the paper’s technical content appears sound.

The broader lesson here is not that AI tools should be avoided, but that they must be used carefully and responsibly. AI can be valuable for identifying research directions, generating questions, or helping navigate unfamiliar literature. It should not be relied on to generate final citations or to verify the existence of sources. For students in particular, it is important to build habits that prioritize checking references against trusted databases and original papers.

Looking ahead, this finding reinforces an idea that has repeatedly shaped how I approach my own work. Strong research is not defined by speed alone, but by care, verification, and reflection. As AI becomes more deeply embedded in academic workflows, learning how to use it responsibly will matter just as much as learning the technical skills themselves.

References

Shmatko, N., Adam, A., and Esau, P. GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers, Jan. 21, 2026

Goldman, S. NeurIPS, one of the world’s top academic AI conferences, accepted research papers with 100+ AI-hallucinated citations, new report claims. Fortune, Jan 21, 2026

— Andrew

5,088 hits

1

The Productivity Paradox of AI in Scientific Research

In January 2026, Nature published a paper with a title that immediately made me pause: Artificial intelligence tools expand scientists’ impact but contract science’s focus (Hao et al. 2026). The wording alone suggests a tradeoff that feels uncomfortable, especially for anyone working in AI while still early in their academic life.

The study, conducted by researchers at the University of Chicago and China’s Beijing National Research Center for Information Science and Technology, analyzes how AI tools are reshaping scientific research. Their findings are striking. Scientists who adopt AI publish roughly three times as many papers, receive nearly five times as many citations, and reach leadership positions one to two years earlier than their peers who do not use these tools (Hao et al. 2026). On the surface, this looks like a clear success story for AI in science.

But the paper’s core argument cuts in a different direction. While individual productivity and visibility increase, the collective direction of science appears to narrow. AI is most effective in areas that already have abundant data and well established methods. As a result, research effort becomes increasingly concentrated in the same crowded domains. Instead of pushing into unknown territory, AI often automates and accelerates what is already easiest to study (Hao et al. 2026).

James Evans, one of the authors, summarized this effect bluntly in an interview with IEEE Spectrum. AI, he argued, is turning scientists into publishing machines while quietly funneling them into the same corners of research (Dolgin 2026). The paradox is clear. Individual careers benefit, but the overall diversity of scientific exploration suffers.

Reading this as a high school senior who works in NLP and computational linguistics was unsettling. AI is the reason I can meaningfully participate in research at this stage at all. It lowers barriers, speeds up experimentation, and makes ambitious projects feasible for small teams or even individuals. At the same time, my own work often depends on large, clean datasets and established benchmarks. I am benefiting from the very dynamics this paper warns about.

The authors emphasize that this is not primarily a technical problem. It is not about whether transformer architectures are flawed or whether the next generation of models will be more creative. The deeper issue is incentives. Scientists are rewarded for publishing frequently, being cited often, and working in areas where success is legible and measurable. AI amplifies those incentives by making it easier to succeed where the path is already paved (Hao et al. 2026).

This raises an uncomfortable question. If AI continues to optimize research for speed and visibility, who takes responsibility for the slow, risky, and underexplored questions that do not come with rich datasets or immediate payoff? New fields rarely emerge from efficiency alone. They require intellectual friction, uncertainty, and a willingness to fail without quick rewards.

Evans has expressed hope that this work acts as a provocation rather than a verdict. AI does not have to narrow science’s focus, but using it differently requires changing what we value as progress (Dolgin 2026). That might mean funding exploratory work that looks inefficient by conventional metrics. It might mean rewarding scientists for opening new questions rather than closing familiar ones faster. Without changes like these, better tools alone will not lead to broader discovery.

For students like me, this tension matters. We are entering research at a moment when AI makes it easier than ever to contribute, but also easier than ever to follow the crowd. The challenge is not to reject AI, but to be conscious of how it shapes our choices. If the next generation of researchers only learns to optimize for what is tractable, science may become faster, cleaner, and more impressive on paper while quietly losing its sense of direction.

AI has the power to expand who gets to do science. Whether it expands what science is willing to ask remains an open question.

References

Hao, Q., Xu, F., Li, Y., et al. “Artificial Intelligence Tools Expand Scientists’ Impact but Contract Science’s Focus.” Nature, 2026. https://doi.org/10.1038/s41586-025-09922-y

Dolgin, Elie. “AI Boosts Research Careers but Flattens Scientific Discovery.” IEEE Spectrum, January 19, 2026. https://spectrum.ieee.org/ai-science-research-flattens-discovery-2674892739

“AI Boosts Research Careers, Flattens Scientific Discovery.” ACM TechNews, January 21, 2026. https://technews.acm.org/archives.cfm?fo=2026-01-jan/jan-21-2026.html

— Andrew

5,088 hits

0

CES 2026 and the Illusion of Understanding in Agentic AI

At CES 2026, nearly every major technology company promised the same thing in different words: assistants that finally understand us. These systems were not just answering questions. They were booking reservations, managing homes, summarizing daily life, and acting on a user’s behalf. The message was unmistakable. Language models had moved beyond conversation and into agency.

Yet watching these demonstrations felt familiar in an uncomfortable way. I have seen this confidence before, often at moments when language systems appear fluent while remaining fragile underneath. CES 2026 did not convince me that machines now understand human language. Instead, it exposed how quickly our expectations have outpaced our theories of meaning.

When an assistant takes action, language stops being a surface interface. It becomes a proxy for intent, context, preference, and consequence. That shift raises the bar for computational linguistics in ways that polished demos rarely acknowledge.

From chatting to acting: why agents raise the bar

Traditional conversational systems can afford to be wrong in relatively harmless ways. A vague or incorrect answer is frustrating but contained. Agentic systems are different. When language triggers actions, misunderstandings propagate into the real world.

From a computational linguistics perspective, this changes the problem itself. Language is no longer mapped only to responses but to plans. Commands encode goals, constraints, and assumptions that are often implicit. A request like “handle this later” presupposes shared context, temporal reasoning, and an understanding of what “this” refers to. These are discourse problems, not engineering edge cases.

This distinction echoes long-standing insights in linguistics. Winograd’s classic examples showed that surface structure alone is insufficient for understanding even simple sentences once world knowledge and intention are involved (Winograd). Agentic assistants bring that challenge back, this time with real consequences attached.

Instruction decomposition is not understanding

Many systems highlighted at CES rely on instruction decomposition. A user prompt is broken into smaller steps that are executed sequentially. While effective in constrained settings, this approach is often mistaken for genuine understanding.

Decomposition works best when goals are explicit and stable. Real users are neither. Goals evolve mid-interaction. Preferences conflict with past behavior. Instructions are underspecified. Linguistics has long studied these phenomena under pragmatics, where meaning depends on speaker intention, shared knowledge, and conversational norms (Grice).

Breaking an instruction into steps does not resolve ambiguity. It merely postpones it. Without a model of why a user said something, systems struggle to recover when their assumptions are wrong. Most agentic failures are not catastrophic. They are subtle misalignments that accumulate quietly.

Long-term memory is a discourse problem, not a storage problem

CES 2026 placed heavy emphasis on memory and personalization. Assistants now claim to remember preferences, habits, and prior conversations. The implicit assumption is that more memory leads to better understanding.

In linguistics, memory is not simple accumulation. It is interpretation. Discourse coherence depends on salience, relevance, and revision. Humans forget aggressively, reinterpret past statements, and update beliefs about one another constantly. Storing embeddings of prior interactions does not replicate this process.

Research in discourse representation theory shows that meaning emerges through structured updates to a shared model of the world, not through raw recall alone (Kamp and Reyle). Long-context language models still struggle with this distinction. They can retrieve earlier information but often fail to decide what should matter now.

Multimodality does not remove ambiguity

Many CES demonstrations leaned heavily on multimodal interfaces. Visuals, screens, and gestures were presented as solutions to linguistic ambiguity. In practice, ambiguity persists even when more modalities are added.

Classic problems such as deixis remain unresolved. A command like “put that there” still requires assumptions about attention, intention, and relevance. Visual input often increases the number of possible referents rather than narrowing them. More context does not automatically produce clearer meaning.

Research on multimodal grounding consistently shows that aligning language with perception is difficult precisely because human communication relies on shared assumptions rather than exhaustive specification (Clark). Agentic systems inherit this challenge rather than escaping it.

Evaluation is the quiet failure point

Perhaps the most concerning gap revealed by CES 2026 is evaluation. Success is typically defined as task completion. Did the system book the table? Did the lights turn on? These metrics ignore whether the system actually understood the user or simply arrived at the correct outcome by chance.

Computational linguistics has repeatedly warned against narrow benchmarks that mask shallow competence. Metrics such as BLEU reward surface similarity while missing semantic failure (Papineni et al.). Agentic systems risk repeating this mistake at a higher level.

A system that completes a task while violating user intent is not truly successful. Meaningful evaluation must account for repair behavior, user satisfaction, and long-term trust. These are linguistic and social dimensions, not merely engineering ones.

CES as a mirror for the field

CES 2026 showcased ambition, not resolution. Agentic assistants highlight how far language technology has progressed, but they also expose unresolved questions at the heart of computational linguistics. Fluency is not understanding. Memory is not interpretation. Action is not comprehension.

If agentic AI is the future, then advances will depend less on making models larger and more on how deeply we understand language, context, and human intent.

References

Clark, Herbert H. Using Language. Cambridge University Press, 1996.

Grice, H. P. “Logic and Conversation.” Syntax and Semantics, vol. 3, edited by Peter Cole and Jerry L. Morgan, Academic Press, 1975, pp. 41–58.

Kamp, Hans, and Uwe Reyle. From Discourse to Logic. Springer, 1993.

Papineni, Kishore, et al. “BLEU: A Method for Automatic Evaluation of Machine Translation.” Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, 2002, pp. 311–318.

Winograd, Terry. “Understanding Natural Language.” Cognitive Psychology, vol. 3, no. 1, 1972, pp. 1–191.

— Andrew

5,088 hits

0

How Computational Linguistics Can Help Stop Phishing Emails?

I’ve always been curious about how language can reveal hidden clues. One place this really shows up is in phishing emails. These are the fake messages that try to trick people into giving away passwords or personal information. They are annoying, but also dangerous, which makes them a great case study for how computational linguistics can be applied in real life.

Why Phishing Emails Matter

Phishing is more than just spam. A single click on the wrong link can cause real damage, from stolen accounts to financial loss. What interests me is that these emails often give themselves away through language. That is where computational linguistics comes in.

How Language Analysis Helps Detect Phishing

  • Spotting unusual patterns: Models can flag odd grammar or overly formal phrases that do not fit normal business communication.
  • Checking stylistic fingerprints: Everyone has a writing style. Computational models can learn those styles and catch imposters pretending to be someone else.
  • Finding emotional manipulation: Many phishing emails use urgency or fear, like “Act now or your account will be suspended.” Sentiment analysis can identify these tactics.
  • Looking at context and meaning: Beyond surface words, models can ask whether the message makes sense in context. A bank asking for login details over email does not line up with how real banks communicate.

Why This Stood Out to Me

What excites me about this problem is that it shows how language technology can protect people. I like studying computational linguistics because it is not just about theory. It has real applications like this that touch everyday life. By teaching computers to recognize how people write, we can stop scams before they reach someone vulnerable.

My Takeaway

Phishing shows how much power is hidden in language, both for good and for harm. To me, that is the motivation for studying computational linguistics: to design tools that understand language well enough to help people. Problems like phishing remind me why the field matters.

📚 Further Reading

Here are some recent peer-reviewed papers if you want to dive deeper into how computational linguistics and machine learning are used to detect phishing:

  • Recommended for beginners
    Saias, J. (2025). Advances in NLP Techniques for Detection of Message-Based Threats in Digital Platforms: A Systematic Review. Electronics, 14(13), 2551. https://doi.org/10.3390/electronics14132551
    A recent review covering multiple types of digital messaging threats—including phishing—using modern NLP methods. It’s accessible, up to date, and provides a helpful overview. Why I recommend this: As someone still learning computational linguistics, I like starting with survey papers that show many ideas in one place. This one is fresh and covers a lot of ground.
  • Jaison J. S., Sadiya H., Himashree S., M. Jomi Maria Sijo, & Anitha T. G. (2025). A Survey on Phishing Email Detection Techniques: Using LSTM and Deep Learning. International Journal for Research in Applied Science & Engineering Technology (IJRASET), 13(8). https://doi.org/10.22214/ijraset.2025.73836
    Overviews deep learning methods like LSTM, BiLSTM, CNN, and Transformers in phishing detection, with notes on datasets and practical challenges.
  • Alhuzali, A., Alloqmani, A., Aljabri, M., & Alharbi, F. (2025). In-Depth Analysis of Phishing Email Detection: Evaluating the Performance of Machine Learning and Deep Learning Models Across Multiple Datasets. Applied Sciences, 15(6), 3396. https://doi.org/10.3390/app15063396
    Compares various machine learning and deep learning detection models across datasets, offering recent performance benchmarks.

— Andrew

5,088 hits

0

Looking Back on 2025 (and Ahead to 2026)

Happy New Year 2026! I honestly cannot believe it is already another year. Looking back, 2025 feels like it passed in a blur of late nights, deadlines, competitions, and moments that quietly changed how I think about learning. This blog became my way of slowing things down. Each post captured something I was wrestling with at the time, whether it was research, language, or figuring out what comes next after high school. As I look back on what I wrote in 2025 and look ahead to 2026, this post is both a reflection and a reset.

That sense of reflection shaped how I wrote this year. Many of my early posts grew out of moments where I wished someone had explained a process more clearly when I was starting out.

Personal Growth and Practical Guides

Some of my 2025 writing focused on making opportunities feel more accessible. I wrote about publishing STEM research as a high school student and tried to break down the parts that felt intimidating at first, like where to submit and what “reputable” actually means in practice.

I also shared recommendations for summer programs and activities in computational linguistics, pulling from what I applied to, what I learned, and what I wish I had known earlier. Writing these posts helped me realize how much “figuring it out” is part of the process.

As I got more comfortable sharing advice, my posts started to shift outward. Instead of only focusing on how to get into research, I began asking bigger questions about how language technology shows up in real life.

Research and Real-World Application

In the first few months of the year, I stepped back from posting as school, VEX Robotics World Championship, and research demanded more of my attention. When I came back, one of the posts that felt most meaningful to write was Back From Hibernation. In it, I reflected on how sustained effort turned into a tangible outcome: a co-authored paper accepted to a NAACL 2025 workshop.

Working with my co-author and mentor, Sidney Wong, taught me a lot about the research process, especially how to respond thoughtfully to committee feedback and refine a paper through a careful round of revision. More than anything, that experience showed me what academic research looks like beyond the initial idea. It is iterative, collaborative, and grounded in clarity.

Later posts explored the intersection of language technology and society. I wrote about AI resume scanners and the ethical tensions they raise, especially when automation meets human judgment. I also reflected on applications of NLP in recommender systems after following work presented at RecSys 2025, which expanded my view of where computational linguistics appears beyond the examples people usually cite.

Another recurring thread was how students, especially high school students, can connect with professors for research. Writing about that made me more intentional about how I approach academic communities, not just as someone trying to get a yes, but as someone who genuinely wants to learn.

Those topics were not abstract for me. In 2025, I also got to apply these ideas through Student Echo, my nonprofit focused on listening to student voices at scale.

Student Echo and Hearing What Students Mean

Two of the most meaningful posts I wrote this year were about Student Echo projects where we used large language models to help educators understand open-ended survey responses.

In Using LLMs to Hear What Students Are Really Saying, I shared how I led a Student Echo collaboration with the Lake Washington School District, supported by district leadership and my principal, to extract insights from comments that are often overlooked because they are difficult to analyze at scale. The goal was simple but ambitious: use language models to surface what students care about, where they are struggling, and what they wish could be different.

In AI-Driven Insights from the Class of 2025 Senior Exit Survey, I wrote about collaborating with Redmond High School to analyze responses from the senior exit survey. What stood out to me was how practical the insights became once open-ended text was treated seriously, from clearer graduation task organization to more targeted counselor support.

Writing these posts helped me connect abstract AI ideas to something grounded and real. When used responsibly, these tools can help educators listen to students more clearly.

Not all of my learning in 2025 happened through writing or research, though. Some of the most intense lessons happened in the loudest places possible.

Robotics and Real-World Teamwork

A major part of my year was VEX Robotics. In my VEX Worlds 2025 recap, I wrote about what it felt like to compete globally with my team, Ex Machina, after winning our state championship. The experience forced me to take teamwork seriously in a way that is hard to replicate anywhere else. Design matters, but communication and adaptability matter just as much.

In another post, I reflected on gearing up for VEX Worlds 2026 in St. Louis. That one felt more reflective, not just because of the competition ahead, but because it made me think about what it means to stay committed to a team while everything else in life is changing quickly.

Experiences like VEX pushed me to think beyond my own projects. That curiosity carried into academic spaces as well.

Conferences and Big Ideas

Attending SCiL 2025 was my first real academic conference, and writing about it helped me process how different it felt from school assignments. I also reflected on changes to arXiv policy and what they might mean for openness in research. These posts marked a shift from learning content to thinking about how research itself is structured and shared.

Looking across these posts now, from robotics competitions to survey analytics to research reflections, patterns start to emerge.

Themes That Defined My Year

Across everything I wrote in 2025, a few ideas kept resurfacing:

  • A consistent interest in how language and AI intersect in the real world
  • A desire to make complex paths feel more navigable for other students
  • A growing appreciation for the human side of technical work, including context, trust, and listening

2025 taught me as much outside the classroom as inside it. This blog became a record of that learning.

Looking Toward 2026

As 2026 begins, I see this blog less as a record of accomplishments and more as a space for continued exploration. I am heading into the next phase of my education with more questions than answers, and I am okay with that. I want to keep writing about what I am learning, where I struggle, and how ideas from language, AI, and engineering connect in unexpected ways. If 2025 was about discovering what I care about, then 2026 is about going deeper, staying curious, and building with intention.

Thanks for reading along so far. I am excited to see where this next year leads.

— Andrew

5,088 hits

0

From Human Chatbots to Whale and Bird Talk: The Surprising Rise of Bio-Acoustic NLP in 2025

As a high school student passionate about computational linguistics, I find it amazing how the same technologies that power our everyday chatbots and voice assistants are now being used to decode animal sounds. This emerging area blends bioacoustics (the study of animal vocalizations) with natural language processing (NLP) and machine learning. Researchers are starting to treat animal calls almost like a form of language, analyzing them for patterns, individual identities, species classification, and even possible meanings.

Animal vocalizations do not use words the way humans do, but they frequently show structure, repetition, and context-dependent variation, features that remind us of linguistic properties in human speech.

A Highlight from ACL 2025: Monkey Voices Get the AI Treatment

One of the most interesting papers presented at the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), the leading conference in our field, focuses directly on this topic.

Paper title: “Acoustic Individual Identification of White-Faced Capuchin Monkeys Using Joint Multi-Species Embeddings“

Authors: Álvaro Vega-Hidalgo, Artem Abzaliev, Thore Bergman, Rada Mihalcea (University of Michigan)

What the paper covers

White-faced capuchin monkeys each have a unique vocal signature. Being able to identify which individual is calling is valuable for studying their social structures, kinship, and conservation efforts.

The main difficulty is the lack of large labeled datasets for wild or rare species. Human speech has massive annotated corpora, but animal data is much scarcer.

The researchers address this through cross-species pre-training, a transfer learning strategy. They take acoustic embedding models (essentially sound “fingerprints”) pre-trained on: (1) Extensive human speech data and (2) Large-scale bird call datasets.

These models are then applied to white-faced capuchin vocalizations, even though the original training never included capuchin sounds.

Key findings

  • Embeddings derived from human speech and bird calls transferred surprisingly well to monkey vocalizations.
  • Combining multi-species representations (joint embeddings) improved identification accuracy further.

This demonstrates how knowledge from one domain can help another distant one, similar to how learning one human language can make it easier to pick up a related one. It offers a practical solution to the data scarcity problem that often limits animal bioacoustics research.

This paper was one of 22 contributions from the University of Michigan’s Computer Science and Engineering group at ACL 2025, showing how far computational linguistics has expanded beyond traditional human text and speech.

Another ACL 2025 Contribution: Exploring Dog Communication

ACL 2025 also included “Toward Automatic Discovery of a Canine Phonetic Alphabet” by Theron S. Wang and colleagues. The work investigates the phonetic-like building blocks in dog vocalizations and aims to discover them automatically. This is an early step toward analyzing dog sounds in a more structured, language-inspired framework.

Why This Matters

  • Conservation applications — Automated systems can monitor endangered species like whales or rare birds continuously, reducing the need for long-term human fieldwork in remote locations.
  • Insights into animal communication — Researchers are beginning to test whether calls follow rule-based patterns or convey specific information (about food, threats, or social bonds), much like how humans use syntax and intonation.
  • Transfer of AI techniques — Models originally built for human speech transfer effectively to other species. New foundation models in 2025 (e.g., like NatureLM-audio) even handle thousands of animal species and support natural language queries such as “What bird is calling here?”

While these ACL 2025 papers represent cutting-edge academic work, the broader field is gaining momentum, with related discussions appearing in events like the 2025 NeurIPS workshop on AI for Non-Human Animal Communication.

This area is growing rapidly thanks to better data availability and stronger models. In the coming years, we might see practical tools that help interpret bird alarm calls or monitor ocean ecosystems through whale vocalizations.

What do you think? Would you be excited to build a simple AI tool to analyze your pet’s sounds or contribute to dolphin communication research? Computational linguistics is moving far beyond chatbots. It is now helping us listen to the voices of the entire planet.

Thanks for reading. I’d love to hear your thoughts in the comments!

— Andrew

5,088 hits

0

Blog at WordPress.com.

Up ↑