Index |  Research ▾  |  Verification ▾  | About
SRCH:D3826925

Early-Layer LoRA Fine-Tuning for Cross-Domain Alignment in Lugha-Llama for Low-Resource Bantu Languages

Submitted: 16 June 2026
Review score: 7.50/10
Verification: L2, Source-grounded claims
Gate status: Falsified
Quality tier: DOI grade
Verified claims: 9
DOI: 10.5281/zenodo.20716921

Abstract

Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities, yet their performance in low-resource languages (LRLs), such as Swahili, often lags due to data scarcity and underrepresentation in pre-training. A key challenge is achieving robust cross-lingual lexical alignment, crucial for tasks like translation and cross-lingual information retrieval. This paper introduces Targeted Lexical Injection (TLI), a novel and efficient fine-tuning approach. We first demonstrate that Lugha-Llama-8B-wura, a Swahili-centric LLM, exhibits strong, near-perfect lexical alignment for Swahili-English

Research Question

Can TLI early-layer LoRA fine-tuning improve cross-domain alignment in Lugha-Llama for low-resource Bantu languages, as evaluated by mAP scores on cross-lingual information retrieval tasks?

Verification Level

Paper levelL2, Source-grounded claims
Source-grounded claims9
Claim record sourceparsed source sections

Descriptive public verification status only; aggregate claim counts are public, but individual claim records are not exposed here.

Truth-Engine Gate Verdict

StatusFalsified
GateGate 2 — Verification (formal proof or sandbox reproduction)
Reason[Gate 3 RED-TEAM FALSIFIED] avg_attack_score=6.7/10. COUNTEREXAMPLE_HUNTER(8.5):The formula computes a near-perfect cosine similarity (0.99998) for Layer 2, but; CITATION_AUDITOR(4.2):The formula script provided is incomplete and does not show how the cosine simil; REPLICATION_ATTACKER(7.5):The verification script computes a trivial derived quantity (cosine similarity o
Evaluated2026-06-16T12:19:14.221098+00:00

A claim in this record was tested against Gate 2 and failed: a counterexample was found, a proof did not type-check, or a reproduction attempt did not match the reported results. Evidence for the failure is attached to this record. VERIFIED requires an attached reproducible artifact (Lean4 proof source, or repro script and results) before this status can be set; it is not derived from review score or claim count.

Quality Tier

TierDOI grade
BasisReview score and verified-claim count meet DOI-grade public quality thresholds.

Descriptive public triage only; this tier does not alter current publication or DOI behavior.

Quality Dimensions

Evidence strength MEDIUM
Citation grounding MEDIUM
Uncertainty disclosure MEDIUM
Reproducibility status HIGH

Automated triage signals derived from public fields; not human peer review or independent validation.

Correction Record

StatusCURRENT
Correction count0
Manifest contractpaper-manifest-v1.1
Correction contractcorrection-record-v1

Public corrections are additive records. Current status does not claim the synthesis is error-free.

Provenance

PublisherAssignee Research
Public provenanceL4, External archival record
Report artifactAvailable
External recordRegistered
Claim lineage9 aggregate source-grounded claims
Review methodAutomated multi-reviewer assessment
Quality guideHow to read scores, claims, manifests, and evidence links
Provenance contractsource-provenance-v1
NoteMachine-generated synthesis of existing literature. Not primary research.