Constraints on the onset: /r/ and /l/

The other approximants possible in second position in the onset are /r/ and /l/. When the obstruent consonant occupying initial position in such combinations is a stop, it may be either a voiced or a voiceless consonant. However, when it is a fricative, it must be [-voice] and [+anterior], i.e. it must be either /f/ or /s/. The allowed onset clusters involving initial stops are /pr/, /br/, /pl/, /bl/, /ʧr/ , /ʤr/ , /kr/, /gr/, /kl/, /l/. Those involving initial fricative consonants are /fr/, /fl/ and /sl/.

Absent from the combinations listed above, though theoretically possible based on the cluster formation constraints mentioned, are /tr/, /dr/, /tl/, /dl/ and /sr/. This absence can be explained by a constraint which blocks onset clusters of consonants specified underlyingly for the features [anterior] and [coronal]. If, however, this constraint is interpreted to apply at the phonetic level instead, the way is open for the clusters involving initial phonetically alveopalatal affricates followed by [ɹ] , i.e. [ʧr] and [ʤr] , to be regarded by speakers as the surface output of underlying /tr/ and /dr/ clusters. This would produce a merger between the phonetic outputs of underlying /tr/ and /ʧr/ , and /dr/ and /ʤr/ . Members of each pair would be realized phonetically as [ʧr] and [ʤr] respectively.

We have already seen a fusing of /tj/ and /ʧ/ realized as [ʧ] , and of /dj/ and /ʤ/ , realized as [ʤ] . Where the phonetic realizations [ʧɹ] and [ʤɹ] are interpreted as involving the phonetic realization of an underlyingly /tjr/ and /djr/, this would violate the constraint on there being no more than two consonants in the onset. This explains the fact, observed by Wells (1973: 10) that “/tr, dr/ are not altogether consistently contrastive with /ʧ/ and /ʤ/ ”. This he illustrates with some examples, e.g. the variation between [ʧruu] ~ [ʧuu] ‘true’, the latter homophonous with [ʧuu]  ‘chew’, and /ʤraa/ ~ /ʤaa/ ‘draw’, the latter homophonous with [ʤaa] ‘jaw’. In each of the preceding pairs, the first form is based on an underlying /tr/ and /dr/ whereas the second is based on an adaptation of unacceptable underlying /tjr/ and /djr/ clusters.