"Leaky phonology and the design of language" [slides]
Recent work on sign language - notably the work of Sandler and colleagues on Al-Sayyid Bedouin Sign Language - has shown that phonology emerges in a new language. There does not appear to be a clear transition point from not having phonology to having it. Rather, the phonological organisation of signs becomes gradually more systematic, and even mature sign languages continue to show a variety of features that are difficult to analyse in terms of a smallish finite inventory of phonological elements. In this paper I suggest that the same is true of spoken language phonology as well, and that spoken languages exhibit phonological phenomena analogous to the hard-to-analyse features in sign phonologies. Sapir famously said that "all grammars leak". We are used to thinking that phonology does not leak, and that all phonetic aspects of an utterance can be analysed in terms of a language's finite inventory of phonological elements, independently of grammatical structure. "Leaky" aspects of spoken language phonology are usually idealised out of consideration in various ways (e.g. as paralinguistic or expressive, as dialect mixture or unassimilated borrowing, as historical change in progress), or simply ignored. I discuss several such phenomena, and suggest that they are comparable to what we find in sign languages. The existence of such phenomena argues against a conception of language design in which phonology is autonomous and unconnected to grammar and meaning (the design feature called "duality of patterning" by Hockett and "double articulation" by Martinet). Rather, phonology is in the first instance a property of the internal structure of signs (or words or morphemes), and it is not always possible or desirable to distinguish phonological from grammatical aspects of this internal structure. If phonology is viewed in this light, it is not surprising that it can "leak" like grammar.
"DADDY, EDDY, NINNY, NANNY and BALDEY: Big Data for speech perception" [slides]
In its concluding decade, the Comprehension Group at the Max Planck Institute for Psycholinguistics created several very large sets of speech perception data, all of which are publicly available for the use of any interested parties (such as, for instance, the laboratory phonology community). The Database of Dutch Diphones (DADDY; Smits, Warner, McQueen & Cutler, JASA, 2003) comprises more than a half a million identification responses, by 18 listeners presented with gated fragments of diphones representing every phonetic segment of Dutch in every possible left and right phonetic context. Even larger is DADDY's English counterpart, the English Diphones Database (EDDY; Warner, McQueen & Cutler, JASA, 2014) which used the same technique with American English input and 22 listeners. The Noise-masked Identifications by Native and Non-Native listeners data set (NINNY; Cutler, Weber, Smits & Cooper, JASA, 2004) contains identification responses by American English (native) and Dutch (non-native) listeners to American English vowels and consonants presented under three levels of noise masking. The other two data sets do not concern phonemic identifications. One is a large corpus of input to 11-month-old infants learning Dutch, with speech from multiple caregivers to the infant and to other adults (NANNY; Johnson, Lahey, Ernestus & Cutler, JASA, 2013). Finally, the BALDEY data set (Ernestus & Cutler, submitted 2014) contains lexical decisions to 2780 spoken words of Dutch and a like number of spoken nonwords. Indicative analyses will illustrate the possibilities of these multifaceted data sets.