Articles for the paper

Twelve recommendations

Twelve articles that make good papers, covering every area of the course, in course order. Eight were proposed in THA3; four were added to cover neurolinguistics and animal communication, found by Claude Opus 5.5 at the instructor’s request in reputable journals beyond the syllabus list (Nature, Brain, Science). You are encouraged to take one of these twelve. You can choose any of the Top-12 without checking with me first. You can also take any article approved for you in THA3. The three checks above apply to every article.

1. First languageBilingualismpitched and approved

   Messenger et al. (2026). Priming beyond syntax: Crosslinguistic evidence for shared syntactic and thematic representations in Polish–English bilingual children. Journal of Memory and Language 151, 104799. doi.org/10.1016/j.jml.2026.104799

   Why Three careful priming experiments with bilingual children and adults: do bilinguals share sentence structure across their languages, and even who-did-what-to-whom? A model of experimental design.

   Watch out for Dense and on the long side, so budget extra time, and learn the jargon first: structural priming, thematic roles, Polish word order and case (OVS). Within- and between-language priming, and children and adults, are compared across different groups, and some child effects are small, so check which comparisons are actually tested.

2. First languagePsycholinguisticspitched and approved

   Rabagliati & Robertson (2017). How do children learn to avoid referential ambiguity? Insights from eye-tracking. Journal of Memory and Language 94, 15–27. doi.org/10.1016/j.jml.2016.09.007

   Why Eye-tracking shows what children check for while they describe a scene, and why they still produce ambiguous descriptions. Older than the usual rule, but a really nice paper.

   Watch out for From 2017, older than the usual rule; the year has been waived for this one. The key child result sorts trials by what the children went on to say, which is a correlation inside an experiment, and some measures were changed after null results (the paper says so openly): a good case for discussing analytic flexibility. The adult groups are small (12).

3. Second languagePsycholinguisticspitched and approved

   Kuperman (2025). How does language distance affect reading fluency and comprehension in English as second language? Studies in Second Language Acquisition 47(3), 757–773. doi.org/10.1017/S027226312510065X

   Why An eye-movement corpus of readers from 18 first languages: does the distance between your first language and English slow your reading? A big question with big data.

   Watch out for Big claims (a “mechanism”) from observational data at a single point in time: a mediation analysis on such data can fit a causal story but cannot show one. Language distance comes bundled with country, university and testing site; the effects are small; and “no effect on comprehension” rests on a non-significant result. Big claims require big evidence: check how far the design carries them. The conclusion is intuitive (the paper itself opens with “Intuitively”), which makes it easy to accept on plausibility alone: a case for belief bias.

4. Second languagePsycholinguisticspitched and approved

   Şafak & Hopp (2025). Learning L2 grammar from prediction errors? Verb biases in structural priming in comprehension and production. Bilingualism: Language and Cognition 28(5), 1249–1265. doi.org/10.1017/S1366728925000033

   Why Two eye-tracking experiments test whether second-language learners learn grammar from being surprised (prediction error). Dense, but very solid.

   Watch out for Very solid but dense: learn the dative alternation, the visual-world paradigm and cluster-based permutation tests first. The claim of “longer-term learning” rests on a before/after comparison without a control group, so it is the weakest link; the verb-bias effect itself is robust.

5. Second languageNeurolinguisticspitched and approved

   Cieśla et al. (2025). The impact of orthography versus images on foreign language learning: Evidence from behavioral and neural markers. Bilingualism: Language and Cognition 29(3), 563–575. doi.org/10.1017/S1366728925100242

   Why Learning new words from pictures or from written translations, with brain responses (ERPs) to see what changes. A compact, well-planned lab experiment that links word learning to the brain.

   Watch out for A focused lab experiment (40 new words, tested 24 hours later, with brain responses), so much of the work is placing it in context: should new words be taught with pictures or with written translations? Read up on the picture superiority effect, dual coding and the N400 first. A strength: it reports a power analysis, which few papers here do.

6. Bilingualismpitched and approved

   García González, Lahdenranta & Lehtonen (2025). Are executive functions engaged in language switching? The role of language proficiency. Bilingualism: Language and Cognition 29(2), 516–526. doi.org/10.1017/S1366728925100199

   Why Does switching between your languages draw on general executive control, and only until you are proficient? A neat, open-data study at the centre of the bilingualism debate.

   Watch out for Correlational: executive functions and proficiency were measured, not manipulated, so “engaged” is an inference from how two tasks go together. The key result rests on one interaction (p = .027) and smaller subgroups, so treat it as promising rather than settled. Its review section is a good map of the bilingual-advantage debate.

7. Psycholinguisticspitched and approved

   Sidhu, Parajuli & Muraki (2026). That sounds exciting! Emotional iconicity facilitates processing and recall. Journal of Memory and Language 150, 104769. doi.org/10.1016/j.jml.2026.104769

   Why Do words that sound like what they mean (an exciting-sounding word like shock) get processed faster? Several experiments and a megastudy on emotional iconicity.

   Watch out for Congruent words were selected, not created, so other word properties could differ: the authors match many (length, frequency, valence, concreteness) but not all (for example, age of acquisition). The effects are small (about 20–25 ms) but replicated in a second experiment with a power analysis. For memory, read past the title: the recall results are mixed across the three tests.

8. NeurolinguisticsBilingualismfree to choose

   Bhaya-Grossman et al. (2025). Shared and language-specific phonological processing in the human temporal lobe. Nature 649(8095), 140–151. doi.org/10.1038/s41586-025-09748-8

   Why Spanish, English and Mandarin speakers listen to their own language and to an unfamiliar one: which parts of speech processing in the brain are shared across languages, and which are learnt? Linguistic questions at the heart of a major neuroscience result.

   Watch out for Demanding methods (recordings from electrodes on the brain surface, encoding models): focus on the question, the figures and the conclusions. The participants are epilepsy patients, some groups are tiny (3 Mandarin speakers), and listening to a known language also means understanding it, which could add to the word-level effects.

9. Neurolinguisticsfree to choose

   Turkeltaub et al. (2025). Right hemisphere language network plasticity in aphasia. Brain 149(4), 1396–1409. doi.org/10.1093/brain/awaf308

   Why A large fMRI study (76 stroke survivors with aphasia, 69 controls) tests three classic claims about how the right hemisphere helps recovery, and finds them wanting. It says openly that earlier studies were too small: a field correcting itself.

   Watch out for Cross-sectional: “more right-hemisphere activity with more time since the stroke” compares different people rather than following anyone over time, and the links to recovery are correlations. Get the published version (Brain) through the FSU library, not the earlier preprint.

10. Animal communicationfree to choose

    Arnon et al. (2025). Whale song shows language-like statistical structure. Science 387(6734), 649–653. doi.org/10.1126/science.adq7055

    Why Methods used to study how infants find words in speech, applied to eight years of humpback whale song, find the same statistical structure as in human language. A high-profile result linking animal communication with language learning.

    Watch out for Short and dense (5 pages; the methods are in the supplementary materials, which you will need). The pattern is robust (it beats shuffled baselines and holds across several thresholds), but the explanation, cultural transmission, is inferred rather than tested, and the data come from one population. “Whales have language” is a conclusion many people would like, so watch for belief bias; the authors themselves make no claim about meaning. An erratum has been published: check what it corrects.

11. Animal communicationfree to choose

    Elie et al. (2025). Categorical and semantic perception of the meaning of call types in zebra finches. Science 389(6766), 1210–1215. doi.org/10.1126/science.ads8482

    Why A true experiment: do zebra finches sort their calls by what they mean rather than by how they sound? A clean design for the question of whether animals have word-like categories.

    Watch out for Short (6 pages plus supplementary materials). Few birds (12 in the main task), though each does many trials. The step from “errors follow context more than sound” to semantic depends on how well sound similarity was measured: is “meaning” an explanation, or a label for what is left over? “Birds know what their calls mean” is an appealing conclusion, so watch for belief bias.

12. Computationalpitched and approved

    Lim et al. (2025). Format inertia: A failure mechanism of LLMs in medical pre-consultation. Proceedings of EMNLP 2025: Industry Track, 1437–1450. doi.org/10.18653/v1/2025.emnlp-industry.101

    Why Why do medical chatbots get stuck asking repetitive questions? A short, clear computational study that tests its explanation with an intervention.

    Watch out for A strength: the causal claim is backed by an intervention (retraining on rebalanced data), across several models. Things to check: each model seems to have been trained only once, with no statistical tests; the patients were simulated; and clinical usefulness was judged by a language model. Short main text, with an appendix.

All proposed articles

These are the articles proposed in THA3 for the Article Summary & Critique, grouped by the course’s areas. Each comes with what to watch out for, from the instructor’s notes and Claude Opus 5.5’s reading of each article at his request: pointers for the critique part of your paper, not verdicts on the articles. Pitched and approved means a student proposed it in THA3 and it is approved for them (that student may be you). The twelve recommended articles are open to everyone, whoever pitched them. If you would like another article that is not yours, ask the instructor, as it may free up once its student has chosen. To switch to an article not listed here, pitch it first.

First language acquisition

• First languageBilingualismpitched and approved

  Messenger et al. (2026). Priming beyond syntax: Crosslinguistic evidence for shared syntactic and thematic representations in Polish–English bilingual children. Journal of Memory and Language 151, 104799. doi.org/10.1016/j.jml.2026.104799

  Watch out for Dense and on the long side, so budget extra time, and learn the jargon first: structural priming, thematic roles, Polish word order and case (OVS). Within- and between-language priming, and children and adults, are compared across different groups, and some child effects are small, so check which comparisons are actually tested.

• First languagePsycholinguisticspitched and approved

  Rabagliati & Robertson (2017). How do children learn to avoid referential ambiguity? Insights from eye-tracking. Journal of Memory and Language 94, 15–27. doi.org/10.1016/j.jml.2016.09.007

  Watch out for From 2017, older than the usual rule; the year has been waived for this one. The key child result sorts trials by what the children went on to say, which is a correlation inside an experiment, and some measures were changed after null results (the paper says so openly): a good case for discussing analytic flexibility. The adult groups are small (12).

Second language acquisition

• Second languagepitched and approved

  Alqarawi & Alshammari (2026). How do Arabic college students perceive AI in L2 poetry interpretation? Asian-Pacific Journal of Second and Foreign Language Education 11, article 34. doi.org/10.1186/s40862-026-00410-y

  Watch out for A one-off survey of students’ opinions (202 students), so its paths show how opinions go together, not what causes what. Read closely: the Abstract calls AI “minimally effective”, while the Conclusion says students saw it working “beyond surface-level pattern matching”; check whether the learner types in the Discussion are backed by a measure in the Method; and the data are not shared. Expect extra critique work.

• Second languagepitched and approved

  Barbour & Debreli (2026). Exploring the psychological dimensions of poetry- and drama-based instruction in EFL learning. Frontiers in Psychology 17, 1802084. doi.org/10.3389/fpsyg.2026.1802084

  Watch out for An interview study (focus groups with 24 students, interviews with 6 teachers): it shows how participants describe poetry and drama lessons, not whether anxiety actually fell. Participants were chosen because they already use these methods, so look at the interview questions in the Appendix: do any assume a benefit? A pleasant conclusion, so watch for belief bias. Expect extra critique work.

• Second languageNeurolinguisticspitched and approved

  Cieśla et al. (2025). The impact of orthography versus images on foreign language learning: Evidence from behavioral and neural markers. Bilingualism: Language and Cognition 29(3), 563–575. doi.org/10.1017/S1366728925100242

  Watch out for A focused lab experiment (40 new words, tested 24 hours later, with brain responses), so much of the work is placing it in context: should new words be taught with pictures or with written translations? Read up on the picture superiority effect, dual coding and the N400 first. A strength: it reports a power analysis, which few papers here do.

• Second languageBilingualismpitched and approved

  Cruz & Suethanapornkul (2025). Second language learners’ sensitivity to grammatical gender in Spanish–English codeswitching environments. Second Language Research 41(4), 743–767. doi.org/10.1177/02676583251314000

  Watch out for A small sample (26 learners). The authors conclude that learners are “not sensitive” to gender in these switches, but their own figures (86% expected answers for masculine nouns, 45% for feminine ones) could also be read as a strong masculine bias plus some sensitivity: worth checking which reading the data support.

• Second languagepitched and approved

  Kim et al. (2026). Comparing the effectiveness of Duolingo, classroom instruction, and classroom + Duolingo instruction conditions on beginner-level French language development. Studies in Second Language Acquisition 48(3), 680–707. doi.org/10.1017/S0272263126101521

  Watch out for Groups were not randomised: classroom students chose whether to add Duolingo, and the app-only learners were recruited separately and were much older (average age 42, against 19 and 22). So ask: do keen learners choose the combination, rather than the combination making them better? The headline is a null result, and “no significant difference” is not the same as “equally effective” without an equivalence test. The study was funded by Duolingo, as the paper states. Many readers would like apps to work as well as class, so this is also a case for belief bias.

• Second languagePsycholinguisticspitched and approved

  Kuperman (2025). How does language distance affect reading fluency and comprehension in English as second language? Studies in Second Language Acquisition 47(3), 757–773. doi.org/10.1017/S027226312510065X

  Watch out for Big claims (a “mechanism”) from observational data at a single point in time: a mediation analysis on such data can fit a causal story but cannot show one. Language distance comes bundled with country, university and testing site; the effects are small; and “no effect on comprehension” rests on a non-significant result. Big claims require big evidence: check how far the design carries them. The conclusion is intuitive (the paper itself opens with “Intuitively”), which makes it easy to accept on plausibility alone: a case for belief bias.

• Second languagepitched and approved

  Maie & Godfroid (2025). Testing the three-stage model of second language skill acquisition. Studies in Second Language Acquisition 47(2), 617–649. doi.org/10.1017/S027226312500021X

  Watch out for Mostly careful about causation: the learning abilities were measured, not manipulated, and the Abstract says correlated. Watch for places in the Discussion that slip into causal wording (contributed to, crucial role). Long (33 pages) but readable: one group of 65 learners and an artificial language.

• Second languagepitched and approved

  Mehmood et al. (2025). Psychological impact of languages on the human mind: Research on the contribution of psycholinguistics approach to teaching and learning English. Journal of Applied Linguistics and TESOL 8(3), 1160–1174. doi.org/10.63878/jalt1089

  Watch out for Raises questions at first glance, so budget for extra critique work. Look into the venue (not on the syllabus list: how does the journal review its papers?) and the authors (what else have they published, and where?), and check the references: do the DOIs lead to the papers cited? A survey with interviews, so the correlations it reports (r = 0.58 and r = −0.47) show what goes together, not what causes what. Compare what the Method says will be done (it is written in the future tense) with what the Results report, and check which variables were actually measured.

• Second languagepitched and approved

  Renner et al. (2017). Dutch courage? Effects of acute alcohol consumption on self-ratings and observer ratings of foreign language skills. Journal of Psychopharmacology 32(1), 116–122. doi.org/10.1177/0269881117735687

  Watch out for From 2017; the year has been waived for this one. A real randomised experiment, but small (25 per group), and the control drink was water, so participants knew whether they had had alcohol and expectations could play a part (the authors say so). The headline, better pronunciation, was not the authors’ prediction (they expected worse ratings), comes from one of several exploratory tests, and has p = .05 (.051 unrounded): a textbook case for false-positive psychology. A fun conclusion, so watch for belief bias. Short (7 pages).

• Second languagePsycholinguisticspitched and approved

  Şafak & Hopp (2025). Learning L2 grammar from prediction errors? Verb biases in structural priming in comprehension and production. Bilingualism: Language and Cognition 28(5), 1249–1265. doi.org/10.1017/S1366728925000033

  Watch out for Very solid but dense: learn the dative alternation, the visual-world paradigm and cluster-based permutation tests first. The claim of “longer-term learning” rests on a before/after comparison without a control group, so it is the weakest link; the verb-bias effect itself is robust.

Bilingualism

• Bilingualismpitched and approved

  Chamorro, de la Viña & Janke (2025). The effects of L2 exposure at school on the cognitive development of children from monolingual backgrounds: A longitudinal study. Bilingualism: Language and Cognition 29(4), 829–843. doi.org/10.1017/S1366728925100448

  Watch out for Families chose the schools, and the immersion school is also the only private one, so school choice and family background could explain the differences; the authors themselves note that higher English exposure already went with better scores at the first test. Read it next to the large meta-analyses that find little or no reliable bilingual advantage once publication bias is corrected (for children: Lowe et al., 2021), and the course side note on the benefits of bilingualism. Many people want this conclusion to be true: a classic case for belief bias.

• Bilingualismpitched and approved

  García González, Lahdenranta & Lehtonen (2025). Are executive functions engaged in language switching? The role of language proficiency. Bilingualism: Language and Cognition 29(2), 516–526. doi.org/10.1017/S1366728925100199

  Watch out for Correlational: executive functions and proficiency were measured, not manipulated, so “engaged” is an inference from how two tasks go together. The key result rests on one interaction (p = .027) and smaller subgroups, so treat it as promising rather than settled. Its review section is a good map of the bilingual-advantage debate.

• Bilingualismpitched and approved

  Koronkiewicz (2025). Adverb position in heritage and L2 Spanish–English code-switching. Borealis: An International Journal of Hispanic Linguistics 14(2), 357–384. doi.org/10.7557/1.14.2.8165

  Watch out for The groups (heritage speakers, L2 learners) are compared as they come, not assigned. Ratings were analysed as z-scores, so check how the conclusion that the groups do not differ was actually tested. Only four adverbs were used. Long in pages, but in large type.

• BilingualismPsycholinguisticspitched and approved

  Mavrou et al. (2026). Cognition and emotion in moral decision-making: The role of working memory, emotional intelligence, and language factors in bilinguals’ moral judgments and emotions. Studies in Second Language Acquisition 48(2), 424–449. doi.org/10.1017/S0272263125101484

  Watch out for A tricky topic that needs careful reading: does deciding a moral dilemma in a second language change the choice? They mostly find no effect. A null result in one study means “no clear evidence either way”, not proof of no effect, and the earlier positive findings deserve the same scrutiny, since small samples and flexible analyses produce false positives. A strength: it is preregistered. The ability results are correlational, so read “influenced” as “was associated with”, and check that the Abstract’s inhibition result (“only in their L1”) matches the results tables.

• BilingualismSecond languagepitched and approved

  Rosales & Torres (2024). The role of metalinguistic knowledge in third-language development among heritage and late bilinguals. International Journal of Bilingualism 29(3), 566–582. doi.org/10.1177/13670069241229161

  Watch out for Metalinguistic knowledge (MK) was measured, not manipulated, so “MK significantly contributed to” (a phrasing the paper uses often) means “went together with”, not “caused”, and the paper does not say so. Check the statistics carefully, too: some reported values look inconsistent (for example, t = −23.5 next to a small effect size, d = 0.29). Online in 2024; it appeared in a 2025 issue, so it counts as 2025.

Psycholinguistics

• Psycholinguisticspitched and approved

  Sidhu, Parajuli & Muraki (2026). That sounds exciting! Emotional iconicity facilitates processing and recall. Journal of Memory and Language 150, 104769. doi.org/10.1016/j.jml.2026.104769

  Watch out for Congruent words were selected, not created, so other word properties could differ: the authors match many (length, frequency, valence, concreteness) but not all (for example, age of acquisition). The effects are small (about 20–25 ms) but replicated in a second experiment with a power analysis. For memory, read past the title: the recall results are mixed across the three tests.

Computational linguistics

• Computationalpitched and approved

  Lim et al. (2025). Format inertia: A failure mechanism of LLMs in medical pre-consultation. Proceedings of EMNLP 2025: Industry Track, 1437–1450. doi.org/10.18653/v1/2025.emnlp-industry.101

  Watch out for A strength: the causal claim is backed by an intervention (retraining on rebalanced data), across several models. Things to check: each model seems to have been trained only once, with no statistical tests; the patients were simulated; and clinical usefulness was judged by a language model. Short main text, with an appendix.

• Computationalpitched and approved

  Salim et al. (2025). MedCOD: Enhancing English-to-Spanish medical translation of large language models using enriched chain-of-dictionary framework. Findings of the Association for Computational Linguistics: EMNLP 2025, 6579–6597. doi.org/10.18653/v1/2025.findings-emnlp.350

  Watch out for A specialised engineering paper (machine translation of medical texts), so expect background work: BLEU, chrF++ and COMET scores, LoRA fine-tuning and the UMLS database. The test set is small (100 sentences), some margins over GPT-4o are tiny, and “clinical accuracy” is claimed without a human evaluation. Potentially challenging: read with extra care.

• ComputationalPsycholinguisticspitched and approved

  Salminen et al. (2025). Decoding deception in the online marketplace: Enhancing fake review detection with psycholinguistics and transformer models. Journal of Marketing Analytics 14(3), 731–748. doi.org/10.1057/s41270-025-00393-8

  Watch out for Validity is in the method details. The fake reviews were written on request by online participants (“Pretend to be an owner of this phone”), each followed by a real one, so they may differ from fakes in the wild. Many word-category tests were run, and reviews by the same writer may sit in both training and test data. Check the claim of “overly positive language” in fakes against the paper’s own sentiment table.

• ComputationalFirst languageask the instructor first

  Wilcox et al. (2025). Bigger is not always better: The importance of human-scale language modeling for psycholinguistics. Journal of Memory and Language 144, 104650. doi.org/10.1016/j.jml.2025.104650

  Watch out for The instructor advises against this one for the paper: it is partly a position paper and partly a report on a shared task (the BabyLM Challenge), so it does not fit the standard criteria and would be a heavy extra load. Come and talk to the instructor first if you really want it.