/z/ of "his" becomes voiceless under the influence of the following voiceless /s/. 2  All other phonemes are realized in their canonical phonetic form. Of course, these canonical allophones might be modified by later rules involving stress, duration, and phonetic context, but the modifications are graded in nature and so do not call for separate discrete allophonic symbols.

Next, each phonetic segment is assigned an inherent duration by table lookup, and a set of duration rules is applied to predict changes to the duration of the segment as a function of sentential context. There are many such rules, so only a few will be illustrated. The final vowel of the sentence is lengthened by a clause-final lengthening rule. Stressed vowels are lengthened, as are the consonants that precede them in the same syllable. The vowels in "ate" and "soup" are shortened because the next consonants are voiceless. A special incompressibility constraint ensures that interacting rules cannot shorten a segment beyond a certain minimum.

Next, a fundamental frequency (fo) contour is determined by rules that specify the locations and amplitudes of step and impulse commands that will be applied to a lowpass filter in order to generate a smooth fo contour as a function of time. The first rule erases the verb-phrase boundary symbol ")" in the phonemic representation because the preceding noun phrase "Joe" is too short to carry its own phrasal pattern. Then, a step rise in fo is placed near the start of the first stressed vowel, in accordance with a "hat theory" of intonation ('t Hart and Cohen, 1973), and a step fall is placed near the start of the final stressed vowel. These rises and falls set off syntactic units. Stress is also manifested in this rule system by causing an additional local rise on stressed vowels, using the impulse commands. The amount of rise is greatest for the first stressed vowel of a syntactic unit, and smaller thereafter. Finally, small local influences of