1.3 Data and methods
2.2.3 Effects and non-effects
Iscontraction is conditioned by Subject Type (pronoun vs. NP), phonological conditioning, and following constituent.1 Each of these conditioning factors have also been reported
for AAVE copula deletion (Labov, 1969), providing unifying constraints with which to analyze and connect these case studies. With regard toSubject Type,iscontraction is near- ceiling after pronouns. Non-pronominal noun phrase (NP) subjects, however, are followed by auxiliaries that are 40% contracted, and are affected significantly by factors such as following constituent, subject length, and preceding vowel vs. consonant (MacKenzie, 2012). Other possible effects, like preceding syllable stressedness or the grammatical class of the preceding word are not significant.
Isallomorphy selection is also affected byphonological conditioning, such that short allomorph selection is higher when the segment preceding the auxiliary is a vowel rather than a consonant (Labov, 1969, Baugh, 1979, Rickford et al., 1991, MacKenzie, 2013). Labov attributes this to a disfavoring effect of consonant clusters (1969). MacKenzie points out that this would predict that allomorphy selection forhasshould follow the same pattern, but vowel vs. consonant is not a significant factor forhas. To account for this, she proposes that instead of consonant cluster avoidance, the effect onisallomorphy selection is instead the result of vowel hiatus avoidance. Rather than disfavoring short allomorph selection after consonants to avoid consonant clusters, short allomorphy is favored after vowels to avoid vowel hiatus.
Preceding-ingand preceding non-sibilant fricatives also select more for the long allo- morph than the short allomorph, which can also be attributed to vowel hiatus avoidance. However, preceding /r/ goes in the opposite direction, significantly favoring short allo- 1There may also be social effects on contraction that have not been fully examined. For instance, males
morph selection even more than preceding vowels do. MacKenzie notes that many words ending in /r/ are kinship terms (likemother, daughter, brother, sister), and that when these are removed, the phonological effect of /r/ disappears. I will return to this anomaly in the results section, and propose that the explanation for these terms acting differently than oth- ers may not be their phonological shape at all. Instead, they all represent human subjects and thus are more likely to select the short allomorph.
In the following constituent effect, the grammatical category of the constituent im- mediately following the copula strongly predicts which allomorph is selected. Following constituents are separated into a hierarchy of grammatical categories, where verbal con- stituents promote highest contraction rates, followed by adjectival constituent, followed by noun phrases (Labov, 1969, Rickford et al., 1991, and Frank and Jaeger, 2008, among many others). Adjectival constituents are sometimes split into adjectival vs. locative, and are a key data point in the origins controversy of AAVE. However, these finer-grained dis- tinctions in following constituent categories do not improve statistical modeling in MAE contraction, therefore following constituents are collapsed into a three way hierarchy, from most likely to contract to least likely: verb>adj>NP (MacKenzie, 2012).
Conditioning onhasandwill
Likeis, Subject Typehas a strong effect on has, such that short allomorphy selection for
hasafter pronouns is near ceiling. Pronoun subjects ofwilldo not categorically select short allomorphs, anditis significantly less likely to occur with the short allomorph ofwillthan any of the other pronouns. One possible reason is that the phonology of it, which is the only personal pronoun to end in a consonant, is driving this effect. Has, have, hadand is
show no difference between the different pronouns in short allomorph selection probability, whilewould also shows this special status of it, leaving it difficult to determine if this is driven by the phonology ofit, and if so, if the effect is generalizable (MacKenzie, 2012). In
the results section of this chapter, I propose a possible reason, animacy, for the seemingly special resistance thatit has toward contraction ofwill. The rest of this literature review only refers to previously demonstrated conditioning for NP subjects.
Preceding segment conditions contraction of has and will, but not along the vowel- consonant split that affectsis. Unlike is, the model forhasandwill is significantly better when the place, rather than manner, of articulation is a factor. However, none of the levels of place of articulation (vocalic, velar, labial, coronal) reaches significance. There is no effect for auxiliary identity, indicating thathasandwillare not acting significantly differ- ently from one another. Neither is there any interaction between auxiliary identity and any of the fixed effects.
Extra-grammatical effects
Contraction is also affected by extra-grammatical effects, such as subject length and subject frequency. The effect ofsubject length, counted as the number of orthographic words in a subject, is that the longer the subject, the less likely that the short allomorph will be selected (MacKenzie, 2013). Because there is no grammatical cut-off for how long a subject can be and still occur with the short allomorph, the gradient decline of short allomorphy selection by NP length is analyzed as the result of processing effects (MacKenzie, 2012). One expla- nation is that the longer a subject is, the more likely it is to be planned in a separate chunk from the auxiliary, which decreases the likelihood of a clitic (the short allomorph) being able to attach to the NP. In other words, if the subject is not planned in the same chunk as the auxiliary, it is unlikely to surface as one prosodic word.
Word frequency is known to affect cognitive processes and behavioral responses in speech production (e.g. Jescheniak and Levelt, 1994) and processing times (e.g. Rayner and Duffy, 1986), and affects contraction. The more frequent a subject head is, the more likely it will select the short allomorph (Lignos and MacKenzie, 2013, Lignos and MacKen-
zie, 2014).