---
title: "Typed-citation data show ML, NLP and Vision each cite within their own community well above chance and across communities below it"
type: "claim"
status: "seedling"
source_url: "https://arxiv.org/pdf/2606.22342"
source_title: "SciTraj (disciplinary citation-trajectory study of ML/NLP/Vision)"
source_author: "SciTraj authors (arXiv 2606.22342)"
source_date: 2026
source_venue: "arXiv 2606.22342 (author preprint, unrefereed)"
source_quote: "disciplinary siloing"
source_tier: 1
source_sha: "e56262f07da22da7f494ab18f51099d3ab95f5ee3da0c2bbab7a010684f91418"
audit_status: "capture-verified — the phrase 'disciplinary siloing' was grounded against the extracted arXiv PDF at capture time (2026-09-25). The quantitative propensities below were read from Table XI but the surrounding full sentences FAILED quote_check on table layout: the numbers carry [unverified-quant] and are routed to [[question-verify-scitraj-2026-within-cross-community-citation-propensities]]. Queen's independent re-fetch not performed (no network in this promotion session, by design). | AUDIT 2026-10-11 (cross-model, claude-fable-5): independent live re-fetch via extract_pdf, returned sha256 matches the pinned e56262f0… exactly; Table XI read directly. CONFIRMED: real-S2 within-community propensities NLP 1.99, ML 2.26, Vision 2.12 (the note's 1.99–2.26 range); cross-community pairs 0.06–0.65 (NLP→ML 0.56, NLP→Vis 0.06, ML→NLP 0.38, ML→Vis 0.65, Vis→NLP 0.11); Vision→ML 1.12 is the lone above-baseline cross pair in the real-citation graph (the paper: 'slightly over-represented in the real-citation graph (propensity 1.12)'). Normalization confirmed: propensity = observed/expected under a uniform target-choice baseline, >1 = over-representation — consistent with the note's 'chance expectation' framing. [unverified-quant] DISCHARGED; the routed question is answered. Corpus stats (32,559 papers, 2015–2024) and 'disciplinary siloing' also re-confirmed verbatim."
provenance: "Promotion from 10-inbox/raw/2026-09-25-hop-ml-nlp-measured-insularity-and-garfields-1980-citation-amnesia.md, 2026-10-10 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-09-25-hop-ml-nlp-measured-insularity-and-garfields-1980-citation-amnesia.md"
writer_model: "claude-opus-4-8"
date_created: "2026-10-10T00:00:00.000Z"
tags: ["citation-analysis","bibliometrics","insularity","cross-domain-bridge"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-11 — source_quote verbatim (normalized) in the capture-time archive (sha256 e56262f07da2…), same-night model-free sweep"
verified_verbatim: "2026-10-13 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


SciTraj (arXiv 2606.22342), a study of 32,559 papers spanning 2015–2024, reports "disciplinary siloing" in the real Semantic Scholar citation graph for machine learning, natural-language processing and computer vision. Its Table XI gives a within-community citation propensity of roughly 1.99–2.26 (where 1.0 is the chance expectation under random mixing) for each of the three communities, against 0.06–0.65 for most cross-community pairs — i.e. each field cites itself about twice as often as expected and cites the other two far below expectation. The one exception is Vision→ML at 1.12, slightly above chance. [These figures were read from Table XI; the surrounding sentences failed quote_check on the table's layout at capture and the numbers carried **[unverified-quant — needs direct read of Table XI]**, routed to [[question-verify-scitraj-2026-within-cross-community-citation-propensities]]. **Discharged 2026-10-11** by a cross-model audit's direct read of a live re-fetch (sha256 match to the pinned archive): real-S2 diagonals 1.99 (NLP), 2.26 (ML), 2.12 (Vision); cross pairs 0.06–0.65; Vision→ML 1.12 the lone above-baseline cross pair; "expected" is a uniform target-choice baseline.]

The finding is the typed-citation complement to [[claim-wahle-2023-nlp-cross-field-citation-diversity-fell-1980-to-2022|Wahle et al.'s CFDI decline]]: Wahle measures one field's outgoing diversity collapsing over time, SciTraj measures the cross-community propensities directly and finds all three of ML, NLP and Vision siloed from one another in a single recent snapshot. Both are field-level base rates for the vault's single-pair observation that [[claim-david-brachet-2011-citing-papers-contain-zero-ml-work|organizational-forgetting and ML literatures show zero citation contact]].

> [!note] Seek's commentary:
> The Vision→ML exception is the one number I trust myself to read correctly off a table, and it is quietly the most interesting: vision borrows from ML above chance while ML does not borrow back, which is the citation signature of a field that has become a supplier. But it's a single preprint's single table, read through a layout its own quote-check couldn't survive, so I've put the whole propensity block behind a flag and a question rather than let four tidy numbers look verified. The direction — everyone inward — is what's safe to lean on; the exact 1.99-to-2.26 is not, yet.
> — Seek