From the Rumelhart prize biography of Paul Smolensky.
Smolensky’s most influential work arose from what was intended to be a confrontation in 1988 with Alan Prince, a preeminent phonologist also known as a critic of connectionist research on language.
E-mail from Alan Prince, Feb. 26, 2016, in response to my question about how a debate became a collaboration:
In various substantive respects, I think the outcome is not as surprising as it might look. At the broadest level, it’s probably important that Smolensky came not out of psychology but physics, where there is a concept of theory, and where the relation between the discrete and the continuous is central. His dealings with connectionist networks have never been limited to the data-set-modeling empiricism that dominates the field. Rather, he’s been interested in the representation of structure and in optimization, and has major results to his name in both areas. On the linguistic side, the big movement in the 70’s and 80’s was away from the tediously descriptive string-manipulating regex-based story of SPE toward the articulated hierarchical structures of syllabic, metrical, and autosegmental phonology. Structure naturally comes with an inclination to think in terms of constraints on structure, in phonology as in syntax. These were often framed in optimization-talk, which was therefore poised to morph into something better defined and more interesting. So: close match waiting to be made. It was a matter of tunneling through the sociological constraints, which proved not to be that difficult.
For the record, the panel we were both on (I wouldn’t call it a “debate” — it was a matter of people getting up and making presentations, if I remember correctly) took place at the 14th meeting of the Society for Philosophy and Psychology at UNC in Chapel Hill in 1988.
E-mail from Paul Smolensky March 27th, 2016
Although all such recollections of mine must be considered marginally trustworthy at best, I seem to recall that at that SPP meeting in 1988, Alan gave what I frankly found to be a beautiful talk on a problem in tonal phonology, while I spoke about how the inadequacies of connectionist models as explanations kept me up at night. (But I might be confusing two of my SPP talks there.) I was especially primed to appreciate Alan’s talk because of the prior exposure I’d had to phonology, thanks to the 1987 LSA Summer Linguistic Institute at Stanford, which I attended because my linguist wife Géraldine Legendre needed to spend time at Stanford (where her PhD advisor David Perlmutter was visiting the Institute), and because I’d loved my undergraduate “Linguistic Inquiry” class, a general education course at Harvard taught by Jorge Hankamer. At the LSA Institute I met regularly with George Lakoff who I’d encountered through his work on “cognitive phonology” presented at a Connectionist Models Summer School. Through George I met John Goldsmith, as they were meeting regularly there, comparing notes on cognitive phonology and John’s work which was, or which was to become, “Harmonic Phonology”; I was mostly an observer. But John was kind enough to give me a draft of his text in progress, to become “Autosegmental and Metrical Phonology”. After the Institute I devoured that book, enjoying every minute of it. So when Alan and I met, I had some basic acquaintance with phonology (beyond my experience in Hankamer’s class on Turkish vowel harmony).
For his part, Alan had already invested heavily in digging deep into connectionism in his work with Steven Pinker, to become “On Language and Connectionism: Analysis of a Parallel Distributed Processing Model of Language Acquisition”, which I found to be an awe-inspiring piece of work (despite its attributing a certain amount of nonsense to me en passant). I still teach that paper every other year in my course on the Foundations of Cognitive Science.
So in addition to the more fundamental points of compatibility that Alan pointed out in his email [above], there were also points of contact in our knowledge of each other’s research area.