Fascinating research about applying ML techniques to cuneiform and other early languages.

I recently received a request to comment on the general tone of international rhetoric. The question posed to me was this: "the language of world leaders today has changed a lot. No one is polite anymore. Even Roosevelt and Stalin treated each other with more respect. It's very dangerous." Do you recognize this perception? And do you agree on the dangerous aspect? Why?

My response was this:
Your question seems to be referring to the perception of a decline in civility in politics, or political language. It may be true in some ways, but here is my perspective: First, in the 20th century, polite and civil leaders in public forums committed some of the worst atrocities, like genocide, in interstate wars and civil wars. Does it matter if words are demure, when civilians and particular identity groups are targeted for violence and victimization? Second, there are plenty of counter-examples of lack of politeness, such as Khrushchev banging his shoe on the table in 1960? Third, the number of democracies in the international system has steadily been increasing since the Second (Cold War) and Third (1980s) Waves of democratization, which corresponds to overall less formality in political language.

What may be the most dangerous is not the decline in civil or polite language, but the weakening of democratic institutions or preferences for liberal democratic values in mature democracies. When institutions like an independent judiciary are threatened, or when citizens show preferences for authoritarian leadership and are complacent to (or acceptant of) weakening respect for human rights (at home or abroad) – these things are problematic. There are many examples of populist candidates and political parties in the Philippines, Hungary, Denmark, France, and Germany.

The map below shows which countries use the most angry language. By far, Iran uses the most angry language, followed by the United States, Venezuela, and France. This roughly corresponds to the use of formal language, which also allows us to infer about politeness. People are more formal and more polite when they are unfamiliar with each other, or when they are speaking to their superiors. It is possible that the decline in civility or politeness also is a reflection that many countries are broadly using less formal language with each other.

Formality in the world
In a previous blog post, I showed how formality in the world varies. The more democratic and/or institutionalized a state is, the less formal their language at the United Nations General Assembly. Conversely, the less democratic and/or institutionalized, the more formal the state representative's language. This is a curious, perhaps counter-intuitive issue. Initially, I assumed that for democracies, public speeches (especially such high-profile ones as the General Debate) would be highly vetted and wordsmithed - and more formal. This is not the case! They are, in fact, crafted to be less formal. I interpret this as a signal that they are members of an in-group, and as such need fewer formalities to signify rank or proof of status. Even non-democracies like China have informal language on par with Western Europe and the United States; I attribute this to their degree of institutionalization and similar scrutiny at the hands of the Communist party's speechwriters and policy powerholders.

Brevity is the soul of wit
Why, then, do non-democracies use more formal language before their global peers on the world stage at the United Nations? One interpretation is that they are political outsiders who seek to bolster their reputation by appearing and sounding more credible. Ironically, this type of language sometimes backfires and serves to discredit where it is intended to impress. There is well established literature that shows that, in the hierarchy of relationships, subordinates use verbose and often clunky phrases as compared to their more terse superiors. (Think: the last time you emailed your boss, carefully checking and re-checking for spelling and other errors, ensuring that your language was clear and conveyed *exactly* what you intended...only to get a three-word response in reply.)

A recent article reinvigorated my thinking about this phenomenon of formality, and what it means for international rhetoric and relationships. Slate discusses why teachers should ban the thesaurus in descriptive writing. Specifically, students have been instructed to replace 'said' with 'exclaimed, chortled, retorted' and the like to provide variation - and ostensibly depth - to their writing. Instead, author Gabriel Roth argues that, "Replacing the word said with “colorful” or “lively” synonyms is a ubiquitous symptom of bad writing." In other words, it discredits the writer. Dense vocabulary makes for unreadable writing, specifically because the writer is trying too hard.

This is exactly the type of writing found in the speeches of many authoritarian regimes: dense writing peppered with honorifics. Their highfalutin language serves to undermine the legitimacy they wish to convey and instead solidifies their position as outsiders in the global community. The implications for this feature of language include demonstrating political transitions (both democratization as well as backsliding) with linguistic measures, as well as emerging and disintegrating coalitions.

Grayson Cupit presented his work entitled "Conflict Forecasting Using Deep Neural Architecture" at the Works In Progress Symposium yesterday. His goal is to improve the accuracy of event data coding and forecasting by implementing an NLP approach to parsing text.

This poster was presented at the 2016 Peace Science Society (International) (PSSI) conference in South Bend, IN. It describes a project on discourse analysis in authoritarian regimes, specifically North Korea. We are investigating ways to better understand the semantic relationships between and within topics, ways to visualize the relationships in the data, and future directions for political forecasting. We are deeply grateful to the many scholars who gave us valuable feedback and encouragement on this project!

Several scholars have weighed in recently on language in the political debates, including CNN's science of deception, and Jeff Hancock on Trump's language. For the third debate, we found many of the same trends as in the first two debates, indicating that the candidates are consistent in their rhetorical approaches, as well as some new features. For instance, Trump used more references to "other" in the third debate than did Clinton. This is compatible with his rhetoric, as his campaign has identified many sources of "other" like Syrian and Mexican immigrants.

The next two graphs complement each other: Clinton uses more positive emotion, and Trump uses more negative emotion. Some research has shown that positive language is contagious: leaders who use more positive and optimistic language "infect" the population and cause citizens to improve their evaluation of the leader. Positivity breeds positivity, which increases leaders' approval ratings.

The next two graphs show the syntactic and semantic patterns of the candidates' language. Both indicate that Clinton again is taking the central route to persuasion, which requires more cognitive processing and is conceptually complex. Trump also continues to pursue the peripheral route to persuasion, which appeals more to emotions. In short: Clinton is appealing to voters heads, and Trump is appealing to their hearts.

Finally, as in previous debates, Clinton used more presidential language than did Trump.

The following graphs represent statistically significant differences in the way that Clinton and Trump used language in the second presidential debate last night. We see some familiar patterns from the first debate, but also some departures as well. Last night, Clinton used much more complex syntax than did Trump.  Complex syntax is conceptually more difficult to follow and suggests that the speaker is using the central route to persuasion (appealing to heads, rather than hearts). Trump used less complex syntax in the debate, suggesting that he pursued the peripheral route to persuasion, which often resonates better with the voting audience.

This next graph shows that Clinton used less referential cohesion in her language than did Trump. Referential cohesion refers to the overlap between a noun, pronoun, or noun phrase and another constituent in the text. Lower levels of referential cohesion indicate that the speaker does not connect concepts between sentences, and/or covers a lot of ground conceptually in the allotted speaking time. During the debate, Clinton tried to cover more policy ground than in the first debate, which may account for why her referential cohesion is low. The following excerpt illustrates that point:

But everybody else, the 170 million of us who get health insurance through our employees got big benefits. Number one, insurance companies can't deny you coverage because of a pre-existing condition. Number two, no lifetime limits, which is a big deal if you have serious health problems. Number three, women can't be charged more than men for our health insurance, which is the way it used to be before the Affordable Care Act. Number four, if you're under 26, and your parents have a policy, you can be on that policy until the age of 26, something that didn't happen before.

Because she provides a list of distinct reasons, there is little overlap between content words and concepts. On the other hand, Trump had higher levels of referential cohesion. The following is an excerpt of his debate language:

Obamacare will never work. It's very bad, very bad health insurance. Far too expensive. And not only expensive for the person that has it, unbelievably expensive for our country. It's going to be one of the biggest line items very shortly. We have to repeal it and replace it with something absolutely much less expensive and something that works, where your plan can actually be tailored. We have to get rid of the lines around the state, artificial lines, where we stop insurance companies from coming in and competing, because they want -- and President Obama and whoever was working on it -- they want to leave those lines, because that gives the insurance companies essentially monopolies. We want competition. You will have the finest health care plan there is. She wants to go to a single-payer plan, which would be a disaster, somewhat similar to Canada. And if you haven't noticed the Canadians, when they need a big operation, when something happens, they come into the United States in many cases because their system is so slow. It's catastrophic in certain ways.

In this text, Obamacare is the key term used in the first sentence. Trump uses both "it('s)" and "plan" subsequently, rather than restating the key term, Obamacare.

On the other hand, Clinton has a higher level of deep cohesion in her narrative than Trump. According to Zwaan and Radvansky (1998) deep cohesion has five components: causation, intentionality (goals), time, space, and protagonists. Deep cohesion refers to the 'mental model' of the text, like whether the speaker weaves together a cohesive narrative with themes that the audience can follow.

The next three graphs help to illustrate the differences between the content of Clinton's and Trump's language. Trump references language related to work less than does Clinton; this involves terms like "corporation, tax, executive, finance, and stock." Clinton also used more words related to achievement than did Trump, including "advance, plan, solve, and succeed." Clinton also used more language related to family than did Trump.

The final graph shows the candidates' consistency between the first and second debates. Again Clinton used more tentative language than did Trump.

The candidates in the vice presidential debate showed some linguistic similarity to their running mates. The following graphs show the statistically significant differences between the two vice presidential candidates. The tentative language results mirror those from the first presidential debate. Like Clinton, Kaine used tentative language more than Paine or Trump did.

Kaine used more simple syntax than did Pence during the debate. A previous analysis of candidates' debates during the 2012 election showed just the opposite effect - that the Republican candidates used more simple syntax and the Democratic candidates used more complex syntax. Complex or simple syntax provides a clue to the speaker's route to persuasion: more complex language indicates a central route, and less complex indicates a peripheral route. The central route to persuasion involves more cognitively demanding processing of the message, whereas the peripheral route is easier to follow.

Finally, we evaluated the candidates' use of honest language, generated by Slatcher et al. (2007). They note that more honest language is associated with increased usage of first person pronouns, references to other people, and exclusion words (except, but). In the VP debate, Kaine used more honest language than did Pence.

The two presidential candidates have very different campaigns, policy prescriptions, and rhetorical styles. In our analysis of the first presidential debate, we treat each speaking turn as a separate document and report the statistically significant differences between Trump and Clinton. The first graph shows that Clinton used more tentative language than did Trump. In her diplomatic role as Secretary of State as well as during her time as a Senator, Clinton was responsible for negotiating across the aisle and across the world. Negotiation requires nuance and compromise, and this is reflected in her more cautious and tentative language. On the other hand, as a career businessperson, Trump has sought to project power, confidence, and certainty, which is also reflected in his language.

This next graph shows that Trump used more negatively valenced language than did Clinton during the debate.

The following graph shows the difference in nonfluencies used by Trump and Clinton. Nonfluencies are filler words like um, err, ah, and mmm. We see here that Clinton uses many more nonfluences than Trump does. This is curious given that men usually use more nonfluencies than women do. One reason for this is that they try to "hold the floor" while formulating their next thought, so as to not leave "dead air" where someone else might start speaking. Given that Trump interrupted Clinton around 51 times, Clinton may have been using this typically masculine strategy in order to "hold the floor" to ensure she could finish her statements.

The next graph (femininity) confirms that Clinton's language was indeed more masculine than Trump's.

Finally, we evaluated Clinton's and Trump's language on the presidentiality dimension generated by Slatcher et al. (2007). There is a statistically significant difference in the level of presidential language; this graph shows that Clinton uses more presidential language than does Trump.