The NETtalk paper is Sejnowski & Rosenberg's 1987 Complex Systems 1(1):145–168 article — a distinct document from their 1986 Johns Hopkins technical report of the same name

"NETtalk" names more than one document, and the backpropagation-history literature usually means the 1987 journal article: Terrence J. Sejnowski and Charles R. Rosenberg, "Parallel Networks that Learn to Pronounce English Text," Complex Systems Volume 1, Issue 1 (1987), pp. 145–168. The publisher's own abstract page confirms venue, volume, issue, title, and authorship; two independent primary-PDF copies (Sejnowski's Salk CNL archive and a UBC teaching mirror) carry the identical running-header citation string "Complex Systems 1 (1987) 145-168."

This is distinct from the earlier 1986 Johns Hopkins EECS technical report "NETtalk: a parallel network that learns to read aloud" (JHU/EECS-86/01, 32 pp.), which was reprinted in 1988 in the edited volume Neurocomputing, Volume 1: Foundations of Research (MIT Press). The two share the "NETtalk" name and overlapping content, so citations conflate them — but the canonical "NETtalk paper" is the 1987 Complex Systems article, not the 1986 report or its 1988 reprint.

NETtalk (a network trained by backpropagation to map English text to phonemes) is one of the mid-1980s demonstrations that backprop could learn a real task, roughly contemporaneous with and preceding the Bell Labs ZIP-code work (claim-lecun-1989-first-practical-recognition).

Lead author Sejnowski was also an invited first speaker at the inaugural 1987 NIPS conference — the founding moment whose proceedings LeCun later freely released via DjVu (claim-djvu-created-to-freely-distribute-scanned-documents). NETtalk itself, published in Complex Systems rather than NIPS, is not in that freed archive, but Sejnowski's own 1987 NIPS backgammon paper is — a person-and-institution bridge between the two, not a shared document.