Psycholinguistics · Second Language Acquisition · Counterfactual Processing · L1 Transfer · Python · Statistical Modelling
🧠 Knowing a grammatical form is not the same as having it available when you need it.
An English learner may recognise the Past Perfect in an exercise and still struggle to use it when explaining an unrealised past:
If I had considered this type of client, our turnover could have increased.
This project asks whether the counterfactual context surrounding a difficult L2 form can influence what learners subsequently do with it.
More specifically, it investigates whether Goal structure, Agency, and Focus in English third conditionals are associated with how L1-Spanish adult learners subsequently select the Past Perfect with past-subjunctive meaning in a desiderative sentence.
The project reconstructs and extends a 2022 psycholinguistic experiment with adult L1-Spanish learners of English.
Participants encountered third-conditional counterfactuals in which three dimensions were manipulated:
- Goal Type — frequent goal, non-frequent goal, or no explicit goal
- Agency — explicit or implicit antecedent agency
- Focus — self-focused (
I) or other-focused (They)
They subsequently completed a desiderative sentence by selecting among four verb-form alternatives: one correct/target Past Perfect response, one L1-transfer response, and two L2-other alternatives.
The original prediction was straightforward:
Frequent goal + Explicit agency + Self-focus
should provide the most favourable environment for subsequent target Past Perfect selection.
The analyses revealed a less additive — and more interesting — behavioural organisation.
The response process became clearer when examined in two stages:
1. Reaching a resolution opportunity
Did the response move beyond the competing L2_other alternatives?
In goal-directed contexts, explicit agency was strongly associated with greater movement into resolution.
2. Resolution destination
Once a resolution opportunity existed, did the response resolve toward L1 transfer or toward the correct/target form?
Here, the relationship depended on the experimental configuration. Focus and Agency jointly organised resolution, while Goal Type shifted the relative destination of that resolution.
One particularly interesting pattern appeared within self-focused goal
contexts: configurations associated with lower predicted probabilities of
moving beyond the L2_other response space were also associated with higher
predicted probabilities of correct resolution once resolution occurred.
The context most strongly associated with reaching resolution was therefore not necessarily the context most strongly associated with resolving correctly.
➡️ The full behavioural and theoretical synthesis is developed in
15_findings_and_theoretical_synthesis.ipynb.
The findings do not identify a universally optimal learning context. Instead, they suggest that the conceptual organisation of language practice may itself be worth treating as an experimental variable.
A learning activity can vary more than vocabulary. It can vary:
- who is represented as acting;
- whose unrealised outcome is represented;
- what kind of goal structures the alternative past;
- and how those dimensions change as the learner becomes able to use the target form across different contexts.
This opens several directions for further research.
The same design could be replicated across different L1 backgrounds to investigate which response patterns reflect particular L1–L2 mappings and which generalise more broadly.
It could also become longitudinal: rather than repeatedly practising one supposedly optimal configuration, learners could encounter systematic variation in Goal, Focus, and Agency while researchers examine retention, error type, response latency, and generalisation to novel or productive contexts.
Large-scale digital language-learning environments would make these questions particularly testable through controlled experimental variation and A/B testing across learner populations.
The broader question is therefore not only:
Which environment makes a target response easiest?
but:
How can systematic variation across environments help a grammatical resource become available beyond the exercise in which it was learned?
The project brings together three perspectives:
- linguistic processing and computational resources, drawing on Gibson;
- functional counterfactual thinking and goal-directed behaviour, drawing on Roese & Epstude;
- perspective and counterfactual representation, informed by the actor–reader findings of Girotto et al.
The theoretical framework, experimental manipulations, actual stimuli, and original predictions are documented in:
➡️ 00_theoretical_framework_and_predictions.ipynb
The notebooks are organised as an analytical journey rather than as isolated analyses. Readers can follow the complete sequence or move directly to the stage relevant to their interests.
00_theoretical_framework_and_predictions.ipynb
Original theoretical framework, experimental manipulations, stimuli, and predictions.
01_initial_data_loading.ipynb ·
02_data_quality_checks.ipynb
Dataset construction, initial inspection, and quality checks.
03_visual_exploration.ipynb ·
04_visual_exploration_self_focus.ipynb ·
05_visual_exploration_other_focus.ipynb
Overall response patterns followed by self- and other-focused exploration.
06_metadata_quality_checks.ipynb ·
07_proficiency_effects.ipynb ·
08_exposure_and_use_effects.ipynb
Metadata quality, proficiency, and English exposure/use as variables to target a subgroup.
09_targeted_subgroup_analysis.ipynb ·
10_targeted_subgroup_analysis_self_focus.ipynb ·
11_targeted_subgroup_analysis_other_focus.ipynb
Targeted exploration of the overall, self-focused, and other-focused behavioural patterns.
12_inferential_tests_overall_subgroup.ipynb ·
13_inferential_tests_self_focused_subgroup.ipynb ·
14_inferential_tests_other_focus_subgroup.ipynb
Inferential follow-up of the subgroup patterns identified during exploration.
15_theoretical_subgroup_exploration.ipynb ·
15b_theoretical_subgroup_exploration_by_goal.ipynb ·
15c_theoretical_subgroup_exploration_by_agent.ipynb ·
15d_theoretical_subgroup_exploration_by_goal_and_agent.ipynb ·
16_summary_tables.ipynb
An exploratory theoretical branch examining consequence reversibility. It was not retained as part of the final explanatory account, but is preserved to document the analytical process and negative/exploratory findings.
01_focus_main_effect.ipynb ·
02_focus_goal_type_effect.ipynb ·
03_focus_agent_effect.ipynb ·
04_focus_agent_goal_type_effect.ipynb ·
05_focus_effects_summary_plots.ipynb
Progressive examination of Focus in relation to Goal Type and Agency, followed by a visual synthesis.
06_response_opportunity_analysis.ipynb
Introduces the distinction between moving beyond competing L2_other
responses and the subsequent destination of resolution.
07_from_effects_to_models.ipynb
Guide to the modelling strategy and the transition from exploratory effects to the model families used in the subsequent analyses.
08_escape_L2_models.ipynb ·
09_transfer_resolution_models.ipynb ·
10_correct_resolution_models.ipynb ·
11_no_goal_behaviour_models.ipynb
Models the two-stage response process: movement beyond competing L2 responses, followed by transfer or correct resolution, with no-goal behaviour examined separately.
12_escape_L2_summary.ipynb ·
13_resolution_summary.ipynb ·
14_no_goal_behaviour_summary.ipynb
Compact summaries of the principal modelling patterns.
➡️ 15_findings_and_theoretical_synthesis.ipynb
Returns the modelled behaviour to the actual counterfactual stimuli and the original theoretical framework, distinguishing statistical findings, interpretive possibilities, and questions for future research.
16_appendix_combined_data_escape_L2.ipynb ·
17_appendix_goal_behaviour_data_escape_L2.ipynb ·
18_appendix_combined_data_transfer_resolution.ipynb ·
19_appendix_goal_behaviour_transfer_resolution.ipynb ·
20_appendix_combined_data_correct_resolution.ipynb ·
21_appendix_goal_behaviour_data_correct_resolution.ipynb
Full model specifications and supporting analyses for readers who want the complete statistical record.
For the research question and theoretical rationale:
start with 00_theoretical_framework_and_predictions.ipynb.
For the main behavioural findings without following every analytical step:
go to 12_escape_L2_summary.ipynb,
13_resolution_summary.ipynb, and
14_no_goal_behaviour_summary.ipynb.
For the overall interpretation and future research implications:
go directly to 15_findings_and_theoretical_synthesis.ipynb.
For the full analytical journey:
follow the EDA and Modelling sections above in numerical order.
This project was developed through an extended analytical dialogue with ChatGPT (OpenAI), used throughout the reconstruction as a coding, statistical, methodological, and editorial assistant.
Its contribution went considerably beyond code generation. Across several months of iterative work, ChatGPT was used to help debug and explain Python workflows, examine alternative statistical approaches, challenge interpretations, identify inconsistencies, reorganise the analytical architecture, translate model outputs back into the experimental stimuli and theoretical questions, and improve the documentation and communication of the project.
The collaboration was deliberately interactive rather than automatic: analytical decisions were discussed, questioned, revised, and frequently rejected or reformulated before being incorporated into the repository. ChatGPT also provided an unusually patient sounding board during the less quantifiable stages of research — including false starts, theoretical reconsiderations, stubborn notebooks, and considerably more analytical rabbit holes than originally anticipated.
All research questions, experimental materials and original data derive from the author's Master's thesis. The reconstruction, analytical decisions, interpretation of results, and final content of this repository remain the author's responsibility.
AI assistance was therefore used here as a tool for reasoning, coding, critique, and communication — not as a substitute for researcher judgement.