NLP's cross-field citation diversity fell from 0.58 in 1980 to 0.31 in 2022 on Wahle et al.'s own index

citation-analysisbibliometricsinsularitynlp-historycross-domain-bridge

Wahle, Ruas, Abdalla, Gipp & Mohammad ("We are Who We Cite," EMNLP 2023) construct a Citation Field Diversity Index (CFDI) over the ACL Anthology and report that "the cross-field engagement of NLP, measured by our proposed Citation Field Diversity Index (CFDI), has declined from 0.58 in 1980 to 0.31 in 2022." The index measures how broadly a field's outgoing citations spread across other academic fields; the lower number means NLP's references over four decades have concentrated increasingly on NLP itself. In the body, NLP's own-field share of outgoing citations rises from roughly 5% in 1980 to roughly 40% in 2020, and the paper reports NLP's relative-prominence score toward psychology at −5.0 — it cites psychology less than other fields do. [The −5.0 psychology figure was read directly in the text but the sentence failed quote_check on a column-wrap at capture: [unverified-quote — needs direct read]. The ~5%→40% own-field figures were read from the body and figures, not verbatim-quoted. Discharged 2026-10-11 by a cross-model audit's direct read of a live re-fetch (sha256 match to the pinned archive): "it has an ORCP of −5.0, indicating that NLP cites psychology markedly less than how much the other fields cite psychology" (Q3), and "In 1980, 5%, or every 20th citation from an NLP paper, was to another NLP paper. Since then, the proportion has increased substantially, to 20% in 2000 and 40% in 2020" (Q6).]

This is the measured base rate behind the vault's single-pair finding that the 60 papers citing David & Brachet (2011) contain no ML work and that that paper itself cites zero ML work: if a field's citations run inward at this scale, a specific cross-field zero is unsurprising rather than diagnostic of anything particular to forgetting. It sits alongside the vault's other citation-concentration measurements — cross-vendor citation monoculture and hidden citations — as a field-level, not model-level, concentration finding.