Identifying Political Campaign Winners Through Linguistic Features

by Paul Sterr and Ken Sultant
E. Lektorell College, Wyoming

Introduction

Analysis of political discourse has primarily focused on such areas as power differentials (Flywater 1997), construction and maintenance of tribal identity (Flock and Clock 2009), the “average guy” persona projection (Walls, Lipinski and Mothhound 2002), fundraising rhetoric (Pekin and Bu 2010), and, more recently, indirect incivility (Jones, Smith and Roachstrand 2018). Taken together, these themes achieve a level of insight that rises almost to the threshold of “so what?”

Linguistics has much more to offer. In this paper, we show that examination of seemingly innocuous linguistic features of political speeches enables us to identify, with almost absolute certainty, the winners of political campaigns.

Methodology

From the Penn Political Corpus we selected speeches made by over 500 individual political candidates in English-speaking countries. These candidates represented a wide range of political parties and levels of political sophistication (i.e., experience), ranging from multi-term heads of state to serially unsuccessful campaigners at the local level. Our corpus included one speech per candidate per election cycle, so that an individual who contested many elections is represented multiple times.

We represented major and minor parties. Thus, since there is only one winner of any given election, non-winners outnumber winners in our sample.¹

To ensure comparability across various election systems, we selected each candidate’s speech for inclusion in our study based on a standard temporal position with respect to the election event itself.

Within each of the speeches we examined, we identified candidate linguistic features, and then compared the linguistic behavior of campaign winners and campaign non-winners. To prevent researcher biases, our research assistants coded speeches without knowledge of whether the speaker in question ultimately achieved electoral success or not.

Results

Our analysis of the linguistic data reveals significant differences between the linguistic devices employed by campaign winners and campaign non-winners.

Morphological differences were noted. The speeches of both winners and non-winners included many imperatives, but the non-winners’ speeches a significantly higher incidence of negative imperatives, such as “don’t give up”, “do not lose heart”, etc.

Non-winners’ speeches had a much higher use of the past tense, while winners make much greater use of the future tense. Winners employed the metaphor of journeys that have begun beginning, while non-winners favored journeys that have ended.

Both winners and non-winners frequently used first person singular pronouns. First person plural pronouns are somewhat more oft-used by winners, while second person pronouns are more common in the speeches of non-winners.

By far the most revealing linguistic feature, though, is semantic. All candidates thanked their supporters and rehashed the main rallying cries of their campaigns. Winning candidates’ speeches contrasted with non-winners, though, through subtle thematic/semantic emphases. These included messages of “unity” (winners), negative feelings such as “pain”, “loss”, and “disappointment” (non-winners), feelings of “humility” (winners), and “choice” (non-winners).

Surprisingly, patterns of usage related to “work” and infrastructure, ratios of thanks to family/campaign staff/voters, and references to both Triple Crown–winning horses and “carnage” were not reliably predictive.

Conclusion

The data support a very strong and somewhat surprising conclusion: based on linguistic features alone, it is possible to identify, almost unfailingly, the winners of political campaign contests.

Although sociolinguists have long claimed that people make judgements of others based purely on features of their speech, we do not believe that any previous research has successfully demonstrated a nearly perfect correlation between specific linguistic features and nonlinguistic outcomes such as successful political campaigning. Therefore, we believe that our study represents a major step forward in applied linguistics, offering promise well beyond the political arena of our own subjects, and into such important areas of everyday life as hiring decisions.

Further Research

We consider these results highly promising, and intend to continue studying the speech of candidates, with a view toward strengthening our ability to identify winners based on a wider range of data. In particular, we hope that future examination of candidates’ speeches delivered prior to an election will yield similarly predictive results.

	On Being [-AGR]—How to Disagree in Linguistic Academese—Bella Coes & P. Vish
	Good Enough for Folk Etymology—Part XI—A. Pocryphal & Verity du Bius
	SpecGram Vol CLXXXVIII, No 4 Contents