Linking phonology and phonetics: An implementation model of tones

The best way to understand the phonology/phonetics interface is to build a phonetic implementation model that generates surface variations from the phonological representation. This paper describes such a model using Mandarin tones as an example. In Chinese conversational speech, sometimes the pitch contours move in the opposite direction of the expected lexical tone, yet native speakers cannot hear the errors and consider the production to be natural. It is argued that this phenomenon is a general reduction process that occurs in all languages and that the level of reduction can be simulated using weights. The weights represent the speaker’s balancing act to meet two conflicting demands in speech communication: for ease of production, the speaker would like to minimize articulatory effort; for ease of perception, the speaker would need to maintain production accuracy. Local reduction/inversion is then explained as global optimization. Following the model assumption, it is concluded that the Mandarin Third Tone Sandhi is a phonological process while the Mandarin Second Tone Sandhi is a phonetic one.