The emergence of phonetic enhancement of phonological features

It has been shown that phonetic enhancement occurs only for contrastive phonological features. This constitutes a challenge for linguistic modelling: if in speech production the phonology comes first, and phonetic implementation has its input the output of the phonology, the enhancement seems to be unpredictable. After all, how should the phonetic implementation phase know which of its input features are contrastive and which aren’t?

One answer that has been given is that the phonology knows about the phonetics: Stanton (to appear in Phonology) follows Flemming (2008) in claiming that in speech production a part of the phonology (namely, phonotactics) follows the phonetic implementation. My computer simulations show, however, that the same results emerge automatically from learning a bidirectional multi-level grammar in Optimality Theory: if evaluation and learning proceed in parallel in both modules of grammar (Boersma 2007, 2008; Apoussidou 2007; Boersma & Van Leussen 2017), and simultaneously for comprehension and production, the rankings of the relevant faithfulness constraints (phonology) happen to become correlated with the rankings of the relevant cue constraints (phonetic implementation). Neither knowledge of the phonetics by the phonology, nor knowledge of the phonology by the phonetics, nor interleaving of phonological and phonetic submodules, turns out to be needed to account for these facts. I conclude that the specialised devices proposed by Flemming and Stanton are superfluous when it comes to modelling phonetic enhancement.