Search the research
Searches titles, authors, topics, charts, data stories and the full text of the essays.
25 results for “AI & evaluation”
- Paper · Extrafemi Research · 2026
Naija Eval: An Open Evaluation Pipeline for Nigerian-Language Translation and Speech Recognition
…umentary audit of a pinned revision, not a new experimental run; 14 references. AI & evaluation Method & verification Extrafemi Research…
- Paper · Extrafemi Research · 2026
Assay: Checking AI-Written Text Against Evidence, One Question at a Time
Obasa Olorunfemi, MBA Precious Olusegun, PhD Evidence-specific checks for AI-written citations, quotations, figures and links, with fail-safe handling of unavailable sources and an explicit distinction between source exi…
- Paper · Extrafemi Research · 2026
Lodestone: Keeping AI Calls Working on Unreliable Free Models Inside a Serverless Time Limit
Obasa Olorunfemi, MBA Precious Olusegun, PhD Time-budgeted failover, circuit breakers, validated outputs, spending controls and privacy-aware routing for unreliable hosted language models. Technical report v1.0, November…
- Paper · SSRN · 2026
AI, Knowledge Accessibility, and the Democratisation of High-Quality Research
…a well-prompted LLM can produce a defensible literature review in an afternoon. AI & evaluation Policy & society SSRN…
- Table · Water & health
Table A3-5: Rainfall statistical Evaluations
Table 16 in “Impacts of Climate Change on Water Quality of Self-Supply Wells”: 20 rows, 6 columns. Impacts of Climate Change on Water Quality of Self-Supply Wells 1 -.009 .199 .002 1 .966 2 -.056 .195 .086 2 .958 3 -.19…
- Table · AI & evaluation
Lodestone’s seven requirements and the failure each answers
Table 1 in “Lodestone: Keeping AI Calls Working on Unreliable Free Models Inside a Serverless Time Limit”: 7 rows, 2 columns. Lodestone: Keeping AI Calls Working on Unreliable Free Models Inside a Serverless Time Limit…
- Table · AI & evaluation
Adversaries and faults, the asset at risk and the main mitigation
Table 2 in “Lodestone: Keeping AI Calls Working on Unreliable Free Models Inside a Serverless Time Limit”: 7 rows, 3 columns. Lodestone: Keeping AI Calls Working on Unreliable Free Models Inside a Serverless Time Limit…
- In an essay · Naija Eval
5. Evaluation protocol and inspected implementation
…nts. The source article is credited to Olorunfemi Obasa, MBA, and this paper retains that attribution. Translation sample. The scripts obtain FLEURS test transcriptions for English, Hausa, Igbo, and Yoruba, deduplicate s…
- In an essay · Naija Eval
6. Translation evaluation
The pinned archive records seven translation model labels. The main article emphasizes six, but its broader summary also contains Mistral Small 3.2. To avoid selection-dependent claims about rankings and price, Figure 1…
- In an essay · Naija Eval
7. Speech evaluation and orthographic sensitivity
…g. Both conditions should be reported, with the stricter orthographic result retained rather than replaced. Human error analysis should distinguish omitted tone marks, changed vowel identity, lexical substitutions, delet…
- In an essay · Who Shares in Corporate Growth?
7. Reassessing the five-decade claim
Nigeria’s post-1973 corporate history contains plausible reasons for labour’s bargaining position to have weakened: oil-boom volatility, the 1986 Structural Adjustment Programme, repeated devaluations, industrial declin…
- Chart · AI & evaluation
Hosted models score highest on Nigerian-language translation; N-ATLaS is well ahead of the open baseline
…and no confidence intervals. chrF++ is an overlap score, not percent accuracy. Naija Eval, archived summary at the pinned revision (publisher-reported, not re-run) Extrafemi Research, technical report v1.0 (Naija Eval) C…
- Chart · AI & evaluation
English to Yoruba is the weakest direction for every one of the seven models
…h direction uses 100 aligned sentences; no confidence intervals were supplied. Naija Eval, archived summary at the pinned revision (publisher-reported, not re-run) Extrafemi Research, technical report v1.0 (Naija Eval) C…
- Chart · AI & evaluation
Removing combining marks cuts Yoruba word error from 64.7% to 33.5%, but changes what is being measured
…ly tone marks, so the two conditions measure different notions of equivalence. Naija Eval, archived summary at the pinned revision (publisher-reported, not re-run) Extrafemi Research, technical report v1.0 (Naija Eval) H…
- Table · AI & evaluation
The checks in Assay and what each compares text with
Table 1 in “Assay: Checking AI-Written Text Against Evidence, One Question at a Time”: 10 rows, 3 columns. Assay: Checking AI-Written Text Against Evidence, One Question at a Time citations public bibliographic records…
- Table · AI & evaluation
What counts as evidence for each check, and what does not
Table 2 in “Assay: Checking AI-Written Text Against Evidence, One Question at a Time”: 10 rows, 4 columns. Assay: Checking AI-Written Text Against Evidence, One Question at a Time citations a bibliographic record with m…
- In an essay · Naija Eval
Abstract
Background. Nigerian-language AI needs evaluation that distinguishes languages, tasks, orthographic conventions, and the conditions under which a system will be used. A fluent demonstration or an aggregate leaderboard c…
- In an essay · Naija Eval
1. Introduction
The availability of a model that can process Nigerian languages is not, by itself, evidence that it can reliably serve Nigerian users. A system may translate a formal sentence while failing on conversational code-switch…
- In an essay · Naija Eval
2. Background and related work
…tions across 17 African languages. These projects are not direct validation of Naija Eval's translation or speech results; they demonstrate that task diversity and language-community participation can inform benchmark de…
- In an essay · Naija Eval
4. System architecture and data flow
…underlying content comparable between directions, while imposing the shared-domain and sampling limits described in Section 9. Task expansion and generation. Each aligned group supplies English-to-Hausa, Hausa-to-Englis…
- In an essay · Naija Eval
9. Limitations and threats to validity
Incomplete prediction retention. Files labelled as raw translation results retain aggregate scores and three examples per direction, not all 600 predictions. Speech artifacts likewise retain only a small example set. Co…
- In an essay · Naija Eval
10. Verification and reproducibility protocol
…run software tests, perform mutation testing, or reproduce the scores. Public availability of scripts is not evidence that those checks have passed. In contrast to a library verification study, no new pass rate or test c…
- In an essay · Naija Eval
11. Responsible use and security considerations
…mended controls; this report has not conducted a penetration test or a supply-chain audit and does not assert that the inspected system implements them. Reporting responsibility. Publish failed runs and missing compariso…
- In an essay · Naija Eval
12. Conclusion
Naija Eval offers useful preliminary evidence and an inspectable starting point for Nigerian-language evaluation. Its archived scores support improvements against selected translation and speech baselines under a docume…
- In an essay · Naija Eval
Declarations
…echnical report, version 1.0, revised 4 November 2026. The evidence boundary remains the documentary review conducted on 1 November 2026. This report version does not identify a Naija Eval software release. Competing int…