---
id: "20260709-0242-what-are-the-exact"
title: "Exact publication year and page range of the NETtalk paper (Sejnowski & Rosenberg)"
type: "capture"
status: "promoted"
promoted_to: ["30-notes/claim-nettalk-1987-complex-systems-distinct-from-1986-report.md"]
not_promoted: []
promotion_note: "Promoted 2026-07-09 (Fable clean-lane marathon). NETtalk = 1987 Complex Systems 1(1):145-168, distinct from the 1986 JHU report — promoted at Tier 1-2. The pdfinfo failure the capture flagged is now repaired; a first-person render of the Wolfram PDF is a clean upgrade target, not needed for promotion."
origin: "batch"
model: "claude-sonnet-5"
date_created: "2026-07-09T00:00:00.000Z"
provenance: "batch run 2026-07-09"
derived_from: []
tags: ["nettalk","sejnowski","rosenberg","backpropagation","connectionism","neural-networks","history-of-ai","bibliographic"]
source_url: "https://www.complex-systems.com/abstracts/v01_i01_a10/"
source_author: "Terrence J. Sejnowski, Charles R. Rosenberg"
source_date: "1987"
source_tier: "1-2 (see per-claim notes; primary PDF text could not be directly rendered in this environment — see caveat)"
---


Harvested from `2026-06-30-who-first-applied-backpropagation-to-a-practical-recognition-task-and-when`. Core question: what is the exact publication year and page range of the NETtalk paper, per the primary PDF?

> [!note] Tooling caveat (read before trusting the tier ratings below)
> Per operating spec, PDF sources should be fetched with `extract_pdf` and read directly. In this session, `extract_pdf` failed on all three candidate PDF URLs with `[Errno 2] No such file or directory: 'pdfinfo'` — a missing environment dependency (poppler-utils), not a source-availability problem. `WebFetch` on the same PDFs returned only raw/garbled binary (the Salk and Wolfram-hosted copies appear to be scanned image PDFs, e.g. CCITT-encoded fax-style scans, with no accessible text layer via this tool). Direct network access via Bash (curl) was blocked by the sandbox. So the claims below rest on: (a) the journal's own abstract page (true Tier 1, publisher-hosted, HTML — resolves and was read directly), and (b) search-engine-indexed text snippets that appear to be extracted verbatim from the primary PDF's own running header on two independent host copies (Salk CNL's own archive of Sejnowski's lab, and a UBC teaching-materials mirror). These snippets are exact quoted strings, not paraphrase, and two independent copies agree — but neither this agent nor the vault's tools directly rendered the PDF page image to confirm with its own eyes. Flagging this so Seek can re-attempt with working PDF tooling before final promotion if a higher bar is wanted.

## Claim: The NETtalk paper (the one usually meant by that name) was published in 1987 in *Complex Systems*, Volume 1, Issue 1

The paper "Parallel Networks that Learn to Pronounce English Text" by Terrence J. Sejnowski and Charles R. Rosenberg — the document almost universally cited as "the NETtalk paper" — appeared in *Complex Systems*, Volume 1, Issue 1 (1987).

- source_url: https://www.complex-systems.com/abstracts/v01_i01_a10/
- source_author: Complex Systems journal (publisher's own site, hosted by Wolfram)
- source_date: page undated but is the permanent abstract page for the article
- exact_quote: page confirms "Volume 1, Issue 1" and lists title "Parallel Networks that Learn to Pronounce English Text" with authors "Terrence J. Sejnowski and Charles R. Rosenberg"; PDF download link resolves to `https://content.wolfram.com/sites/13/2018/02/01-1-10.pdf`
- source_tier: 1 (publisher's own journal site; this is the primary venue of record, though the year "1987" itself is corroborated below rather than printed as bare text on this abstract page)
- claim_type: quantitative (date) — Tier 1-2 required, met via publisher site + corroboration below

Corroborating exact-quote (from search-engine indexing of the primary PDF text itself, two independent host copies of the same PDF):
- exact_quote: "Complex Systems 1 (1987) 145-168 Parallel Networks that Learn to Pronounce" (appears identically as the indexed title/snippet for both of the URLs below, consistent with this being the paper's own printed running header)
- source_url (copy 1): https://papers.cnl.salk.edu/PDFs/Parallel%20Networks%20That%20Learn%20to%20Pronounce%20English%20Text%201987-3351.pdf (Sejnowski's own lab archive, Computational Neurobiology Laboratory, Salk Institute)
- source_url (copy 2): https://www.cs.ubc.ca/~murphyk/Teaching/CS340-Fall07/reading/nettalk.pdf (UBC CS teaching archive, scanned copy of the same article)
- source_tier: 2 (primary-document text, but surfaced via search-index snippet rather than a direct read by this agent — see tooling caveat above)

## Claim: The page range is 145–168

Both independent PDF-host copies' indexed text give the same page range, matching the standard bibliographic citation for this paper.

- exact_quote: "Complex Systems 1 (1987) 145-168 Parallel Networks that Learn to Pronounce" (same snippet as above; page numbers 145 and 168 appear as the volume's start/end pages for this article)
- source_url: https://papers.cnl.salk.edu/PDFs/Parallel%20Networks%20That%20Learn%20to%20Pronounce%20English%20Text%201987-3351.pdf
- source_author: Terrence J. Sejnowski, Charles R. Rosenberg (paper itself, via Salk CNL archive)
- source_date: 1987
- source_tier: 2 (see tooling caveat)
- claim_type: quantitative (page range) — Tier 1-2 required, met per above with the caveat noted

Wikipedia's article on [[NETtalk]] independently states the same figures in its reference list, giving a Tier 3-4 secondary corroboration (definitional/historical-adjacent, not load-bearing on its own but consistent):
- exact_quote: "Sejnowski, Terrence J., and Charles R. Rosenberg. \"Parallel networks that learn to pronounce English text.\" Complex systems 1.1 (1987): 145-168."
- source_url: https://en.wikipedia.org/wiki/NETtalk_(artificial_neural_network)
- source_tier: 3-4 (encyclopedia secondary source; used only as corroboration, not as the load-bearing citation for the number)

## Claim: "NETtalk" names more than one document — the 1987 Complex Systems article is distinct from an earlier 1986 technical report and its 1988 reprint

Sejnowski and Rosenberg first circulated NETtalk as a Johns Hopkins University EECS technical report in 1986, titled "NETtalk: a parallel network that learns to read aloud" (JHU/EECS-86/01), which was later reprinted in 1988 in the edited volume *Neurocomputing, Volume 1: Foundations of Research* (MIT Press). This is a different document from the 1987 *Complex Systems* journal article ("Parallel Networks that Learn to Pronounce English Text," 145-168) that is the primary target of this capture and the one most commonly cited as "the NETtalk paper" in the backpropagation-history literature. The two are easy to conflate because they share the "NETtalk" name and overlapping authorship/content.

- exact_quote: "(1986) Terrence J. Sejnowski and Charles R. Rosenberg, NETtalk: a parallel network that learns to read aloud, The Johns Hopkins University Electrical Engineering and Computer Science Technical Report JHU/EECS-86/01, 32 pp."
- source_url: https://direct.mit.edu/books/edited-volume/5431/chapter/3958546/1986-Terrence-J-Sejnowski-and-Charles-R-Rosenberg
- source_author: MIT Press (Neurocomputing, Vol. 1: Foundations of Research, edited volume gateway page)
- source_date: chapter reprint dated 1988 in the edited volume; original tech report dated 1986
- source_tier: 2 (publisher's own book-gateway page, listing the primary bibliographic record of the reprinted chapter)
- claim_type: historical/bibliographic (uncontested) — Tier 3-4 floor applies and is exceeded here

> [!note] Seek's commentary
> This capture answers the "exact year and page range" question about as tightly as the available tooling allowed: 1987, *Complex Systems* 1(1), pp. 145-168. Confidence is high because two independent host copies of the same primary PDF agree character-for-character on the citation string, and the publisher's own site confirms venue/volume/issue/title/authors. The one open loop is that this agent never got a clean render of the PDF itself (missing poppler-utils in this environment, and the hosted copies appear to be scanned images rather than text-layer PDFs) — worth a re-run with working PDF tooling if the promoted note needs a first-person primary read rather than triangulated corroboration.

## Further leads

- The 1987 Complex Systems abstract page lists the direct PDF at https://content.wolfram.com/sites/13/2018/02/01-1-10.pdf — retry with working PDF-extraction tooling to get a first-person primary read of the header/footer.
- ScienceDirect hosts "NETspeak — A re-implementation of NETtalk" which could be useful for tracing NETtalk's reception/replication history: https://www.sciencedirect.com/science/article/abs/pii/0885230887900131
- The 1986 JHU technical report JHU/EECS-86/01 (32 pp.) is the earliest NETtalk document and may be relevant to the broader "who first applied backprop to a practical recognition task" question this was harvested from — not yet located as a standalone PDF.
- Semantic Scholar has separate paper records for the 1987 Complex Systems article and the 1988/1986 "reads aloud" piece — worth cross-checking citation counts and abstracts if disambiguating impact: https://www.semanticscholar.org/paper/Parallel-Networks-that-Learn-to-Pronounce-English-Sejnowski-Rosenberg/de996c32045df6f7b404dda2a753b6a9becf3c08