A Corpus Assisted Comparative Analysis of Donald Trump’s and Hillary Clinton’s Speeches during the 2016 Presidential Election

First of all, it must be specified that the corpus that will be taken into account was compiled by D. W. Brown (D.W. Brown, 2016)2, who assembled various speeches of the two politicians in question. Since this paper proposes to analyze texts pronounced or produced by Trump and Clinton, the choice could have fallen on the aforementioned speeches, on interviews or on tweets. All these kinds of discourse have advantages and disadvantages. In the case of speeches, it is useful to remember that they are usually produced by a speechwriter and that the role of the candidate is sometimes limited to its enunciation. However, the fact that the speaker is likely to make changes during the production of the text is not to be excluded; indeed, it is probably wrong to assume that the text is reported exactly as it was written. So, although the president is not often the author of the speech, he can characterize it in a personal way, through his/her way of speaking, becoming almost co-author.
On the other hand, interviews were not chosen because of the difficulty that arises from ensuring that the final product has not been cut or reworked, as in the case of television interviews and transcripts for newspaper articles. Furthermore, interviews would have been the most difficult genre to find, since it is not easy to encounter a large, representative number of interview transcriptions of both candidates. The obvious advantage given by an interview is however the certainty of the authorship of the text; in fact, even in the event that the utterer had received help from advisors in the preparation of some answers, his/her role in the enunciation and in the linguistic act is much more central.

The speaker’s authorship is even more prominent in tweets, which have however been discarded because the idea of analyzing texts that were more complete, articulate and easily identifiable as being in a specific context prevailed. Moreover, it would have required more time for the selection, given the large number of tweets, combined with the variety of topics. Another problem represented by tweets is the fact that some of them have suspension points and therefore clearly have a following tweet that completes the precedent one, however until now Donald Trump has never united them according to the subject dealt with in a so-called thread, that is a series of connected Tweets from one person with which you can provide additional context, an extended point or an update. This certainly makes it more difficult, months later, to link them in a coherent way in the corpus. The person who assembled the corpora that will be the object of this study, D. W. Brown, chose texts that were pronounced during official election campaign events.
Furthermore, texts containing sections of Q&A and interviews or press conferences with other Heads of State or other important political figures deemed overly long were discarded. In fact, an interactive structure would have deviated too much from discourse genre. To standardize these types of documents to other texts, the interventions of speakers other than Trump should have been eliminated. Of course, this would have deprived the President's answers and comments of their context, so Brown chose to discard these texts.