Cât de reproductibil este trecutul nilo-saharian?

Ungureanu treats the internal reliability of published comparative dictionaries themselves as something that can be statistically audited.

Evidence: The Nilo-Saharan paper is framed as a statistical assessment of the reliability of two published dictionaries.

Inference: Reference works are not assumed to be neutral ground truth; their comparison sets can be tested.

author_proposal · methodological

Ungureanu's Nilo-Saharan paper identifies a real multiple-search problem: broad semantic tolerance and very short roots multiply the opportunities for accidental cognate-looking matches.

Evidence: Paper repeatedly shows short CV roots paired with large semantic neighborhoods; its toy calculation uses 1/60 per exact CV and scales with semantic alternatives.

Inference: Semantic search-space size belongs explicitly in the false-positive null.

editorial_assessment · methodological

The strongest methodological contribution of the Nilo-Saharan paper is an independent-reconstruction reproducibility test: substantially different reconstructions from the same historical problem are evidence that method choice materially affects the inferred proto-language.

Evidence: Ungureanu explicitly proposes testing suspected congruence bias by having an independent analyst use the same corpus/method and comparing outputs.

Inference: Historical reconstruction can be studied with inter-method/inter-analyst reliability and held-out prediction.

editorial_assessment · methodological

Roger Blench's 2026 comparison independently confirms that Bender and Ehret produced strikingly different Nilo-Saharan reconstructions from substantially the same empirical domain, including very different family structures and numbers of reconstructed forms.

Evidence: Blench table: Bender 258 overall/173 PNS vs Ehret 1660; different membership, tree models, sound-correspondence use and subgroup justification.

Inference: Ungureanu's concern about reconstruction reproducibility remains a live and externally corroborated methodological issue.

editorial_assessment · field_context

Disagreement between Bender and Ehret does not by itself demonstrate that Nilo-Saharan is not genealogically coherent; Blench 2026 argues for coherence using morphological, phonological, pronominal and lexical evidence beyond the disputed dictionaries.

Evidence: Blench argues for ≥12 branches based on three-term number marking, ATR harmony, t/k affixing, pronouns and lexicon.

Inference: Dictionary unreliability and family nonexistence are separate propositions.

contested · field_context

Ungureanu's Texas-sharpshooter critique can be operationalized as a held-out prediction test: semantic transition rules should be fixed before seeing the evaluation data.

Evidence: Paper objects to semantic categories being invented after observing a proposed cluster (e.g. drip/blood/cataract/tear/dew → 'liquid').

Inference: Pre-registration and held-out lexical data turn the cognitive-bias critique into a falsifiable experiment.

editorial_assessment · methodological