The DOI matches its Crossref record; 48 of the 48 references it cites were found in a registry or an index; 5 of the 65 sources its claims cite were checked for support.
Testing the three-stage model of second language skill acquisition
Open accessAbstract
Abstract Skill acquisition theory conceptualizes second language (L2) learning in three stages (declarative, procedural, and automatic), yet competing theoretical models with fewer stages also exist, and the number of stages has never actually been tested. We tested the validity of the three-stage model by investigating the number and nature of learning stages in L2 skill acquisition. Seventy-three participants deliberately learned grammar and vocabulary of a miniature language through explicit-deductive instruction. They systematically practiced comprehending the language until their accuracy and speed of performance did not improve anymore. Participants received a battery of tests assessing individual differences in their declarative and procedural learning abilities. We first applied hidden Markov modeling to participants’ reaction time data (obtained from the language practice) to compare rival hypotheses on the number of stages in L2 skill acquisition. We then examined which cognitive variables predicted participants’ performances (accuracy and speed) in each stage. Our results indicated that participants indeed acquired L2 skills in three stages and that their performance correlated initially with declarative learning ability, but there was a tendency for procedural learning ability to take over in the later stages. Our findings provide the first formal evidence for the influential three-stage model of L2 skill acquisition.
Overview
CoChat Summary
TL;DR
This study provides the first formal test of the three-stage model of second language (L2) skill acquisition by having 65 L1 English speakers learn and practice comprehending Mini-Nihongo after explicit instruction. Hidden Markov modeling favored a three-stage model over one- and two-stage rivals, and regression analyses showed declarative learning predicted accuracy across stages while procedural learning predicted speed in later or specific stages.
Key findings
- The three-state hidden Markov model fit best: BIC = 54,921 and AICc = 54,876, with model weights ≈ 1.000, compared with one-state (BIC = 90,768; AICc = 90,741; weights = .000) and two-state models (BIC = 55,044; AICc = 55,008; weights = .000). Participants needed 18 trials on average to move to stage 2 and an additional 165 trials to reach stage 3; 16.9% did not reach the final stage.
- Declarative learning positively predicted accuracy in all three stages, with posterior probabilities of .998 (stage 1), .998 (stage 2), and .951 (stage 3). The accuracy difference between +2 SD and −2 SD declarative ability was 8.21% in stage 1, 4.81% in stage 2, and 2.57% in stage 3.
- For reaction time (RT), declarative learning predicted speed in stage 2 (Pr(b < 0) = .984) and stage 3 (Pr(b < 0) = .985). ASRT predicted RT in stage 2 (Pr(b < 0) = .925), while SL predicted RT in stage 1 (Pr(b < 0) = .884) and stage 3 (Pr(b < 0) = .883).
- RT decreased from approximately 8 s to 6 s within stage 1, by an additional 2 s from stage 1 to stage 2 (6 s to 4 s), from 4 s to 3 s within stage 2, and by an additional 1 s from stage 2 to stage 3. Within-stage and between-stage decreases were comparable, contrasting with the ACT-R assumption that most improvement occurs through stage transitions.
Methods
- Seventy-three L1 English speakers with no prior study of case-marking languages were recruited; eight were excluded, leaving 65 participants (46 female, 14 male, 5 nonbinary or not specified; mean age = 20.35 years, SD = 2.61, range = 18–30). Most were Michigan State University students not majoring in linguistics or language teaching, and they received US$60.
- Over six online sessions on Gorilla, participants received explicit-deductive instruction in Mini-Nihongo via a 19-min video, completed vocabulary and grammar tests, and practiced comprehension through a sentence-picture matching task. Practice included 128 trials per session on days 3–6, eight blocks of 16 trials each, plus 16 warm-up trials, totaling 528 practice trials in 33 blocks. Cognitive tests measured declarative learning (CVMT, LLAMA-B) and procedural learning (ASRT, SL).
- Trial-level RT histories were analyzed with hidden Markov models comparing one-, two-, and three-state models using AICc, BIC, and model weights. Bayesian generalized linear mixed models (brms; four MCMC chains of 10,000 iterations, 1,000 warm-up) then predicted accuracy (binomial) and log-transformed RT (Gaussian) from practice trials, stage dummies, declarative factor scores, ASRT z scores, SL z scores, interactions, and random intercepts and slopes.
Limitations
- The authors state that findings may not generalize to other linguistic structures (e.g., collocations, pronunciation, pragmatic routines) or modes of language use (e.g., production), and that item-based models such as the race model or CMPL theory may better account for vocabulary learning. They also state that learners might not go through three stages under incidental or inductive learning conditions.
- The authors note that the two procedural learning tasks (ASRT and SL) did not correlate (r = −.18, 95% CI [−.41, .06], p = .13), so they did not reduce them to a single procedural score; they cite construct-validity issues with implicit language aptitude as a possible reason.
- Evident from the text: the sample was restricted to mostly young university students who were L1 English speakers with no prior case-marking language study, and the study was completed online at home, which may limit generalizability.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
5 Valid · 1 Not valid · 38 Not checked · 78 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Additionally, a meta-analysis by Hamrick, Lum and Ullman (2018) confirmed this dynamic role of declarative and procedural learning abilities in L2 learning, particularly among adult learners (but see also Oliveira, Henderson & Hayiou-Thomas, 2023, for counterevidence).”
§ Testing the three-stage model of second language skill acquisition > Note:
What the sources say
- Supportedcites Hamrick, Lum & Ullman, 2018 source
The Results report meta-analyses of adult L2 learners: grammar was significantly associated with declarative learning at lower L2 experience and procedural learning at higher L2 experience. The Discussion summarizes this changing pattern, supporting the claim about a dynamic role for the two learning systems. (LLM, conf=0.93)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Our results indicated that participants indeed acquired L2 skills in three stages”
§ Testing the three-stage model of second language skill acquisition > Abstract
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“their performance correlated initially with declarative learning ability”
§ Testing the three-stage model of second language skill acquisition > Abstract
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“there was a tendency for procedural learning ability to take over in the later stages.”
§ Testing the three-stage model of second language skill acquisition > Abstract
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“At the trial level, regressing the logarithm of mean RT on the logarithm of practice trials yielded R 2 = .90.”
§ Testing the three-stage model of second language skill acquisition > Descriptive results
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
53 more claims
Create a free account to see all 122 claims and every source verdict.
- Probably not a substantial claim. 28
- It may state two separate claims as one. 39
- Not yet classified. 2
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ Testing the three-stage model of second language skill acquisition > Reaction time
“the effect of SL seemed larger than that of declarative learning in stage 3 (0.51 for SL versus 0.27 s for declarative).”
§ Testing the three-stage model of second language skill acquisition > Reaction time
“The difference in RT between individuals with the highest and lowest declarative learning scores (at +2 SD and -2 SD away from the mean) was 0.55 s (95% CrI [0.52, 0.60]) and 0.51 s (95% CrI [0.44, 0.56]) in stage 2 and stage 3, respectively.”
Citations
Cites
Gelman, A., Carlin, J. B., Stern, H. S., & Rubin, D. B. (2003) · doi:10.1201/9780429258480 · Cited by 27,276 · cited 1×
“In Bayesian analysis, prior knowledge in the form of probability distributions is combined with data to generate the posterior distribution of model parameters (Gelman et al., 2013).”
Inference from iterative simulation using multiple sequences
Gelman, A., & Rubin, D. B. (1992) · Statistical Science · doi:10.1214/ss/1177011136 · Cited by 16,627 · cited 1×
“We monitored the value of b R associated with each parameter to assess whether the MCMC simulation converged on a stable solution (Gelman & Rubin, 1992).”
brms : An R Package for Bayesian Multilevel Models Using Stan
Bürkner, P.-C. (2017) · Journal of Statistical Software · doi:10.18637/jss.v080.i01 · Cited by 9,476 · cited 1×
“We used the R package Testing the three-stage model of second language skill acquisition brms (version 2.19.0; Büerkner, 2017) to estimate the posterior distributions through the Markov chain Monte Carlo (MCMC) simulation, consisting of four MCMC chains of 10,000 iterations each, with the first…”
InfluentialAnderson, J. R. (2013) · doi:10.4324/9781315799438 · Cited by 6,761
(1983) · Cited by 5,041
Speaking: from intention to articulation
(1989) · Choice Reviews Online · doi:10.5860/choice.27-1947 · Cited by 5,003
Speaking: From Intention to Articulation
Levelt, W. J. M. (1989) · doi:10.7551/mitpress/6393.001.0001 · Cited by 4,804
Colman (2022) · doi:10.1073/pnas · Cited by 4,370
Acquisition of cognitive skill
Anderson, J. R. (1982) · Psychological Review · doi:10.1037/0033-295x.89.4.369 · Cited by 3,663 · cited 6×
“In our study, we focused on declarative and procedural learning as cognitive abilities underlying the acquisition of cognitive skills (Anderson, 1982, 1983).”
InfluentialToward an instance theory of automatization.
Logan, G. D. (1988) · Psychological Review · doi:10.1037/0033-295x.95.4.492 · Cited by 3,466 · cited 1×
“The race model, proposed as part of the instance theory of automatization (Logan, 1988, 2002), claims that learners store and represent each experience of skill performance as an instance.”
Research works printed but not linked: 3
- Trahan, 1988, Continuous visual memory test
- Meara, 2019, The LLAMA tests v3. LLABA-B v3.2 beta
- Maie, R. (2022). Testing the three-stage model of second language skill acquisition (Publication No. 29998996) [Doctoral dissertation], Michigan State University. ProQuest Dissertations and Theses Global.
Cited by
Maze tasks for L2 morphosyntactic skill acquisition
Serfaty, J. (2026) · Language Teaching Research · doi:10.1177/13621688261422857 · Cited by 1
Communicative Drills: The Impact of Communicative Intent on Oral Proficiency
Serfaty, J. (2026) · TESOL Quarterly · doi:10.1002/tesq.70155 · Cited by 2
Differences in characteristics of two aspects of procedural learning in action video game players
Guo, W., Tao, T., Jiang, J., & Wang, B. (2026) · PeerJ · doi:10.7717/peerj.21013
Decoupling abstractness and granularity in L2 hierarchical syntactic building: An ERP study
朱素琼, Li, S., & Yang, Y. (2026) · Journal of Neurolinguistics · doi:10.1016/j.jneuroling.2026.101362
Gérat, P., & Campo, M. (2026) · Journal of Applied Sport Psychology · doi:10.1080/10413200.2026.2716240
Metadata
- Authors
- Maie, R., & Godfroid, A.
- Affiliations
- Tohoku University (JP)
- Michigan State University (US)
- Corresponding author
- Maie, Ryo
- Venue
- Studies in Second Language Acquisition
- Publisher
- Cambridge University Press (CUP)
- Published
- 2025-04-23
- Volume
- 47
- Issue
- 2
- Pages
- 617-649
- ISSN
- 0272-2631 (ISSN-L)
- 1470-1545
- OpenAlex
- W4409694183
- Cited by
- 11
- References
- 43
- Citations by year
- 2026: 11
- FWCI
- 18.02
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Locations
- Studies in Second Language Acquisition (publishedVersion, cc-by)
- OSF Preprints (OSF Preprints) (submittedVersion, cc-by)
- OSF Preprints (OSF Preprints) (submittedVersion)
- Funders
- Division of Behavioral and Cognitive Sciences (2140704)
- Topics
- Social Sciences › Arts and Humanities › Language and Linguistics › EFL/ESL Teaching and Learning
- Social Sciences › Psychology › Developmental and Educational Psychology › Second Language Acquisition and Learning
- Life Sciences › Neuroscience › Cognitive Neuroscience › Neurobiology of Language and Bilingualism
- Keywords
- three-phase model, declarative knowledge, procedural knowledge, automaticity, individual differences, reaction time, hidden Markov models, grammar learning, vocabulary acquisition
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-09-28, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-09-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Testing the three-stage model of second language skill acquisition. https://app.cochat.ai/papers/10.1017/s027226312500021x
- The paper
- Maie, R., & Godfroid, A. (2025). Testing the three-stage model of second language skill acquisition. Studies in Second Language Acquisition. https://doi.org/10.1017/s027226312500021x
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.