One Feedback System Does Not Fit All: Localising Data-to-Text Driver Coaching for the United Kingdom and Nigeria
Organizations: Department of Computing Science, University of Aberdeen, King’s College, Aberdeen AB24 3UE, United Kingdom
Abstract
Data-to-text driver coaching is often presented as a generic pipeline from telematics events to advice. This paper argues that its content requires localisation because usefulness and credibility depend on drivers' knowledge, prevalent risks, regulation, infrastructure, and available data. Two independently developed systems in the United Kingdom and Nigeria are compared by tracing requirements through content selection, generation, and field evaluation. The UK system prioritises post-trip reflection, explanations tied to road and place context, and tone-sensitive wording. The Nigerian system combines legally grounded, once-daily Tips based on detected events with weekly persuasive Reports; it foregrounds safety education and alcohol-related risk in response to reported gaps in formal training and traffic-rule knowledge, as well as local road-safety priorities. Reliable speed-limit metadata supported speeding feedback in the UK, whereas its scarcity led the Nigerian evaluation to exclude speeding from its outcome metric. Both interventions were associated with reduced distance-normalised unsafe-event rates in their own field studies, although their designs and metrics preclude an effect-size comparison. The analysis yields a requirements-to-content design process for localising safety-critical NLG without treating a high-income deployment as the default.
Figures & tables
| Design question | UK evidence | Nigerian evidence | Consequence for generated content |
|---|---|---|---|
| What help is needed? | Drivers sought fair, actionable explanations of detected events and preferred post-trip reflection to distracting live alerts. | Stakeholders, specifically drivers and Federal Road Safety Corps (FRSC) officials, identified lack of training and knowledge of traffic rules as causes of unsafe driving, alongside demand for education and awareness. | UK reports explain behaviour in context; Nigerian Tips explicitly teach rules and consequences. |
| Which risks deserve emphasis? | Speeding, phone use, harsh manoeuvres, and risk near schools and other busy or sensitive places. | Non-compliant/aggressive driving, distraction, fatigue, and alcohol/substance use; the app could reliably instrument harsh braking, acceleration, swerving, and alcohol risk. | Each system selects locally relevant, credibly detectable behaviours rather than a universal inventory. |
| What makes feedback acceptable? | Supportive, non-judgemental language; clarity, fairness, and sensitivity to preferred feedback style. | Credible Nigerian legal references plus respectful, ability-building coaching that acknowledges local road realities. | Personalised tone in the UK; two complementary Nigerian surfaces for authority and persuasion. |
| What can data support? | Telematics supplied speed and speed-limit fields plus mapped/geofenced context. | Digital posted speed limits were too sparse and unreliable for consistent legal-speed classification. | UK content can discuss speeding and place risk; the Nigerian event rate and feedback avoid unsupported speeding claims. |
| Content layer | Locally motivated configuration |
|---|---|
| Core purpose | UK: explain post-trip events in context. Nigeria: teach rules and support behaviour change. |
| Evidence | UK: event, time, and road/place. Nigeria: event statistics plus retrieved law, fine, and penalty. |
| Risk focus | UK: speed, phone use, manoeuvres, and sensitive places. Nigeria: manoeuvres and alcohol influence. |
| Adaptation | UK: profile-sensitive tone and road advice. Nigeria: legal retrieval, local realities, and TPB coaching. |
| Surface | UK: integrated daily/weekly Report. Nigeria: event Tips plus weekly Report. |
| System | Local evidence | Content selection | Generation control | Driver-facing output |
|---|---|---|---|---|
| UK | Telematics events, speed-limit fields, maps, and geofences | Rules select incidents and road/place advice | Constrained paraphrase of the rule-selected draft | Contextual daily or weekly Report |
| Nigeria | Smartphone events, alcohol classification, and legal knowledge base | Keyword legal retrieval for Tips; weekly statistics for Reports | Structured Tip output; two-step Report generation and revision | Once-daily legal Tips; weekly persuasive Reports |
| Dimension | DRIVINGBEACON: UK | Safe Drive Africa: Nigeria |
|---|---|---|
| Design | 12-week within-subject deployment: 4-week pre-intervention, 5-week intervention, and 3-week post-intervention with feedback withdrawn. | Within-subject pre-intervention/intervention deployment over approximately 25 weeks, excluding transition trips. |
| Data | 30 drivers; daily or weekly Reports; phase-level driver event rates. | 54 analysed drivers (27 commercial, 14 private, 13 company); 802 trips; once-daily Tips and weekly Reports. |
| Outcome | Unsafe events/km from speeding, phone use, and harsh braking/acceleration; geofenced events remain in the total and place context supports interpretation. | Unsafe events/km from harsh braking, rapid acceleration, and swerving; speeding excluded. |
| Result | 0.4275 pre-intervention, 0.2270 intervention, and 0.1493 post-intervention; significant phase effect and all pairwise differences. | 1.44 pre-intervention and 1.16 intervention, a 19.4% reduction ( ); 43/54 drivers (79.6%) improved. |
| Limitations | No parallel control group; all drivers received feedback during intervention; the personalisation effect was not isolated; post-intervention observation lasted three weeks. | No parallel control group; intervention trip count did not correlate significantly with improvement; alcohol-classifier outputs were excluded from the outcome. |
Text Generation
Text generation studies methods for producing coherent, useful language from prompts, data, and dialogue context.
See all 424 papers