Annotations under a paradigm of dependency relationships between words in an utterance differs drastically from the constituent annotations for this particular dataset. Dependency annotations presupposes no essential null elements in the utterance structure, thus all the null elements that were introduced in the previous section are once again stripped from this dataset, so that no deletion of any elements is assumed theoretically.
A benefit of this is that moving outside of the frame of reference of what is 'deleted' between General American English and African American English allows for the analysis of African American English as a language unto itself rather than through comparison with the standard, while a drawback is that it becomes much harder for those interested in morphosyntactic variation to study phenomena that are variable between Standard English and African American English, and, indeed, within African American English, because the context of variation is not preserved.
The null elements that were so difficult to handle under the constituent annotation paradigm become fairly straightforward with dependency annotations. The null elements are simply left out of the relations.
An utterance containing have to in the predicate with a modal function would receive the root node in the original dependency annotation schema, but this does not represent the intuition that the main verb of the sentence, and thus the root, is in fact the one following the have to. Have to should be treated as a multiword auxiliary, similar to must or other modals expressing a similar effect. The dependency should branch from the root verb to have, with to receiving an 'mwe' dependency stemming from have.
Similar to have to being synonymous with must in this spoken data, get also carries a very similar meaning to become in informal and/or spoken registers of English. Therefore, get, like become, should be used with an 'xcomp' dependency annotation.
A substantial difference between the PTB tagging and the dependency annotations is that the discursive function of discourse markers can be highlighted using the 'discourse' dependency in a way in which the PTB is unequipped to handle. An example follows, in which the utterance where the discourse function of the imperative wait went unnoted in the constituent annotation.