Affricates

The two English affricates, /tʃ, ʤ/, follow the patterns of stops and fricatives with regard to fortis/lenis (voiceless/voiced) distinction. Thus, we can state that /ʤ/ is fully voiced only in intervocalic position (e.g. agent [eʤənt], ledger [lεʤɚ]); in initial and final position it is only partially voiced (e.g. Jane [d̥ʒen], fudge [fΛd̥ʒ]).

Also, like stops and fricatives, sonorants (vowels, diphthongs, and sonorant consonants) are longer preceding the voiced (lenis) affricate than when preceding the voiceless (fortis) affricate. The vowels in ridge [ɹ̣ɪʤ] and badge [bæʤ] are longer than the ones in rich [ɹ̣ɪtʃ] and batch [bætʃ]; we obtain a similar difference for the nasal consonant in binge [bɪnʤ] and pinch [pɪntʃ], and lunge [lΛnʤ] and lunch [lΛntʃ]. Since the above two generalizations are valid for stops, fricatives, and affricates, we can reformulate the rule and make it general for all obstruents. Thus, we can state that (a) lenis obstruents are voiced only when they occur intervocalically; they are partially voiced in initial and final position, unless immediately followed by a voiced sound; and (b) sonorants are longer preceding a voiced (lenis) obstruent than when preceding a voiceless (fortis) one.

There is, however, a notable difference between the affricates on the one hand and the stops and the fricatives on the other with respect to lengthening in geminates. Unlike stops and fricatives, in which one long articulation with no separate release is observed in cases of two adjacent identical segments (e.g. stop Peter [stɑp:itɚ], rough features [ɹ̣Λf:it ʃɚz]), affricates have separate releases. Thus, sequences such as much cheaper [mΛtʃ tʃipɚ] and orange juice [ɔɹ̣əndʒ dʒus] cannot be pronounced as *[mΛtʃ:ipɚ] and *[ɔɹ̣əndʒ:us]. The same principle holds when the two affricates are different in voicing (e.g. much jollier [mΛtʃ dʒɔliɚ], large chair [lɑɹ̣dʒ tʃεɹ̣]).

Although affricates are phonetically made up of two sounds /t + ʃ / and /d + ʒ/, phonologically they behave like one segment and not like consonant clusters. There are several supporting arguments for this assertion. Firstly, English does not allow any onsets with a stop + fricative combination. Secondly, as indicated earlier in section 1.3.4, data from speech errors (spoonerisms) show that affricates, when transported from one position to another, fill the space that is vacated by a single segment, as illustrated by key chain [ki tʃen] becoming [tʃi ken], Ray Jackendoff [ɹ̣e dʒækəndɔf] becoming [dʒe ɹ̣æk əndɔf], and last cigarette Tim had in June [...tɪm hæd ɪn dʒun] becoming [...dʒɪm hæd ɪn tun] (Fromkin 1973). Thus, if roughly cheaper [ɹ̣Λfli tʃipɚ] were to suffer a spoonerism, the likely form would be [tʃΛfli ɹ̣ipɚ], and never *[tɹ̣Λfli ʃipɚ]. Lastly, sounds representing an affricate are noticeably shorter than the sequence of the sounds that make up the affricate. Thus, /tʃ/ in watch ear is shorter than a sequence of [t] + [ʃ] in what sheer.