The Humanitarian Turn at the UNSC: Explaining the development of international norms through machine learning algorithms

Changing values in international society

Few questions in international relations are as fundamental as those that ask about the extent to which ideas about appropriate behavior influence conduct and, in contrast, how much relations between states are driven by narrowly defined interests. Scholars in recent years have produced a series of powerful works showing how the values of international society have changed over time (Pinker, 2011; Hathaway & Shapiro, 2017; Goldstein, 2012; Gat, 2006; Mueller, 2009). Yet, while the question of where norms come from has proved more difficult to answer than the question of whether they are influential, certain ideas have gained currency. Historical study reveals the prominent role played by norm entrepreneurs, who are unsurprisingly most likely to succeed when they have the backing of a powerful state (Ikenberry, 2009; Kaufmann & Pape, 1999). In addition to processes based on power disparities, scholars also emphasize the importance of persuasion and social influence in international politics (Johnstone, 2011; Johnston, 2001).

There are two main ways that international organizations in particular can spread norms to the rest of the international community. First, in areas where they exercise power, they can take direct action, or make policy that incorporates certain norms. International organizations often have their own budgets and mandates from states, and control over resources and their distribution can directly shape behavior (Hurd, 2008). Once a state accepts a new practice in order to comply with a regulation from abroad, that norm can become embedded in domestic politics through the activities of bureaucracies and institutions created in order to ensure compliance (Park, 2005). International organizations can also directly sanction behavior that violates widely accepted norms through refusing to provide resources, placing certain conditions on their distribution, or even by simply deciding who gets to take advantage of mechanisms designed to facilitate mutually beneficial cooperation (Hathaway & Shapiro, 2017).

In addition to taking direct action, international institutions can spread norms through statements and declarations that are aspirational in nature, or normative declaration. This can contribute to the creation of new logics of appropriateness that can change the behavior of domestic and international actors (Finnemore, 1993). For this reason, nongovernmental organizations (NGOs) and governments fight to control the agenda and influence the pronouncements of international institutions, expending tangible resources in order to do so, even when there is no direct payoff or lawmaking power at stake such as in the case of the UN General Assembly (Hurd, 2008:112–118; Hanania, 2019).

The Security Council as a diffusor of norms

The UNSC is a singular institution, as its ‘ability to mobilize massive coercive resources is unprecedented among international organizations, and almost all states in the system have consented to it in a highly public way’ (Hurd, 2008: 30). The UNSC spreads norms through both methods described above. With regards to direct action, recent decades have seen a rise in UN-created peacekeeping initiatives, and they appear to reduce the likelihood of developing nations reverting to civil war (Fortna, 2008). In authorizing missions in countries such as Afghanistan and Sierra Leone, the Security Council promotes values such as good government, including women in decisionmaking processes, and a free and open press. Similarly, it sanctions the outcasts of international society such as terrorists and those who produce prohibited weapons, and such punishments tend to be enforced by states and other international institutions (Reinisch, 2001).

The Security Council also spreads norms through normative declaration, a practice that has increased over the last few decades. While the effect of direct action can be traced quite easily, how can we likewise be confident that normative declarations have had their intended effect? There are two lines of evidence that suggest that they do. First, as mentioned above, states and other actors such as NGOs invest tangible resources in trying to influence the Security Council, including lobbying for seats (Hurd, 2008: 111–136). This even extends to controlling the agenda, or whether certain issues will or will not be discussed, regardless of whether those issues have a chance of finding their way into resolutions. NGOs likewise push for resolutions that on their surface appear to have no direct effects in terms of the expenditure of resources (Shepherd, 2008: 386–392; True-Frost, 2007: 156–157).

The UNSC has the power not only to allow the use of force on the part of states, but to require it, making it the global institution with perhaps the greatest ability to diffuse norms. While dealing with international conflicts was its main goal at creation, the UNSC has become more concerned with issues that in previous decades were seen as purely domestic in nature (Joachim, 2003). We can call this change in the focus the ‘Humanitarian Turn’, which can be defined as the increasing tendency to address issues that pertain to the rights of individuals and issues within countries. One may conceptualize this as a shift away from a focus on protecting states to a focus on protecting individuals (Teitel, 2001).

This means that not only has the nature of peacekeeping changed, but also we have seen the passing of what have been called thematic issues of peace and security resolutions (‘TIPS resolutions’), which ‘address issues as diverse as women’s equality, HIV/AIDS, and children’s rights, and urge all – not just select – states to take action’ (True-Frost, 2007: 116). These documents are not tied to any particular conflict, nor even the original mission of the UN, and lack any enforcement mechanism. Nonetheless, a series of case studies show how UN resolutions passed through the Security Council and General Assembly have influenced domestic actors seeking political change (Risse & Sikkink, 1999; Cohn, 2008).

Overcoming inertia

This section argues that major changes in international politics can be created through highly visible exogenous shocks to the system, or focal point events. O’Neill introduces the concept of focal point outcomes in international politics, which are game outcomes that individuals expect to arrive at due to their beliefs about what others believe about their own ‘consideration of extra-game-theoretic factors’ (O’Neill, 2005: 45–50). In other words, second- and higher-order beliefs about other actors make certain outcomes more likely than others. Similarly, in international politics there are events that are important because everyone believes that everyone else believes that they will be seen as important. Actors are therefore more likely to change their behavior under such circumstances, as the recent shock causes them to rethink ‘the way things are done’ and come to new understandings about the role of institutions (Widmaier, Blyth & Seabrooke, 2007).

Socialization between states is likely to take hold when the socialized country faces a crisis that calls into question the legitimacy of those in power (Ikenberry & Kupchan, 1990; Kim & Sharman, 2014). International institutions in particular can be quite resistant to change, particularly if their practices have become routinized and their attention to rules and procedures prevents them from effectively facing new challenges (Barnett & Finnemore, 1999; Shanks, Jacobson & Kaplan, 1996; Hopf, 2010). Thus, just as we see exogenous shocks being the catalysts for change in domestic politics, we may expect a similar dynamic to develop within international institutions.

Focal point events depend on what economists call common knowledge, or knowing that everyone else knows some relevant information (Simler & Hanson, 2017: 61–62). Common knowledge is of fundamental importance to international law, as are higher-order beliefs. For example, the justification for punishing an aggressor lies not only in the first-order effect of not punishing, or letting everyone know that the actor got away with the act. The second-order effect is also important, as each actor knowing that every other actor is more likely to believe that aggression pays can make states more afraid of one another and exacerbate the security dilemma (Hathaway & Shapiro, 2017: 345–349).

Actors and institutions are therefore unlikely to develop a new logic of appropriateness based on information that only they themselves know. The United Nations depends on states for funding and personnel, and states in turn look to it for legitimacy (Hurd, 2008). An international organization is less likely to try to spread norms if it has no way of enforcing them, while a state is less likely to carry out enforcement if doing so will be seen as illegitimate. A focal point event can modify each side’s expectations about what others will do, thereby changing behavior. The same applies to other actors such as NGOs, and how they interact with other agents in the international system. NGOs greatly increased their activity after the collapse of the Berlin Wall, likely in part because donors and others saw more potential for their actions to be effective (Joachim, 2013).

This article therefore argues that we should expect to see major changes in the topics that the United Nations addresses in the aftermath of public events of high symbolic importance. Inertia being the default state of institutions, focal point events create new expectations among participants in such organizations and those that may seek to influence them. In a model in which political actors are rational, in the sense that they invest time and resources where they will be most effective (Tsebelis, 1990: 18–22), there should be relatively short windows of opportunity in which individuals and organizations can expect to have a particularly large impact. Leaders that believe in advocating for human rights as an important aspect of foreign policy will consider the period after a focal point event as a good time to direct political capital towards influencing the Security Council. Human rights NGOs and other activists, always facing trade-offs in what kinds of activities they emphasize, should likewise see such moments as times when they would be best served by directing more of their resources towards powerful and influential organizations, as opposed to, for example, grass roots activism (Roth, 2004). Events identified by historians as politically or culturally important, even if they are not associated with shifts in power relationships, should be associated with major changes in the practices of states. The study of history can guide and be supplemented by text analytic methods.

Given that new logics of appropriateness can result from focal point shocks, two candidates emerge as possible catalysts for the Humanitarian Turn. The Helsinki Accords (1975) are considered a breakthrough in human rights law and motivated reform movements across the Eastern bloc (Thomas, 2001: 206). Before that time, communist power had seemed to be permanently entrenched, as highlighted by a recent crackdown on political activity. The Helsinki Accords were a catalyst for the mobilization of transnational networks in countries such as Poland and Czechoslovakia, and motivated nations to begin putting pressure on the USSR by emphasizing human rights in the course of diplomacy centered around other issues. These agreements signed were therefore ‘an essential feature of the crystallization of international human rights consciousness in 1976–77’ (Moyn, 2010: 150).

The signing and ratification of the Helsinki Accords was not an isolated event, but rather part of a movement that throughout the 1970s for the first time made human rights a priority in the normal course of international politics (Moyn, 2010: 148–155; Denhert, 2013). This plausibly helped lead to the end of the Cold War, the second major shock of the second half of the 20th century. The events of late 1991 meant that, practically overnight, the great ideological conflict of the era was over, the largest empire in the world had collapsed, and the international system went from a bipolar to a unipolar structure. For the reasons discussed above, we should see major changes at the Security Council around the time of these two focal point events.

Even if we do, however, there may be other explanations as to why any shifts we find happened when they did. First, some ground the Humanitarian Turn in specific events within the Security Council itself. For example, in 1991, Secretary General Javier Pérez de Cuéllar issued a report calling on the United Nations to take a stand against human rights violations, and scholars have marked this as an important milestone (Mertus, 2009: 100–101). However, it is the large powers that vote on and can potentially veto UNSC resolutions, which means that if the secretariat has power, it is only because the permanent members allow it. Secretaries-general have shown a passion for human rights for at least half a century, which means that if the UNSC only began dealing with such issues in recent decades, it is unlikely to be because the leadership of the institution has changed (Fröhlich, 2007).

Sometimes explanations also account for the role of outside actors. For example, in recent decades the UNSC has seemed to be particularly sensitive to how war affects women and girls, a change that is often traced to Resolution 1325 (2000) (Tryggestad, 2009). Yet, while norm entrepreneurs in the form of NGOs have helped bring attention to this issue, such groups have been seeking to influence the international community for over a century and a half. Why, then, have they only had success in the last few decades (Park, 2005: 116)? If this is to be attributed to socialization undertaken by NGOs, one has to ask whether they have suddenly become more persuasive in recent years (O’Brien, 2000). A much more likely scenario is that there has been a psychological change among those who are being socialized, or the members of the Security Council. Furthermore, the rise of TIPS resolutions and an increase in concern with human rights in peacekeeping missions seem to call for a more holistic explanation. The Security Council became more interested in the rights of women and girls around the same time it was deciding to start taking on environmental issues and protecting other vulnerable groups. As it is highly unlikely that NGOs working on these various issues all became more effective in their work at around the same time, we need a more general theory to account for the Humanitarian Turn.

Indeed, scholars have not failed to notice that the rise in humanitarian pronouncements at the Security Council came after the end of the Cold War. Thus, according to one theory, the great superpower conflict of the era prevented the Council from cooperating as much as it could have (Mertus, 2009: 8–9; Schweigman, 2001: 287–288). The problem with this view is that much of the rise in Security Council activity seems to be related to topics that are among the least relevant to Cold War rivalries – for example women, children, the protection of civilians, and human trafficking. As will be shown below, the portion of Council work devoted to conflict itself, in contrast, has remained relatively constant over time. During the Cold War, the UNSC did have some peacekeeping operations, but largely ignored human rights issues. For example, the UN Peacekeeping Force in Cyprus was formed in 1964, and although the Security Council passed dozens of resolutions addressing that mission and the conflict itself, in those texts we see very little discussion of the humanitarian concerns that would later become the hallmarks of resolutions related to ongoing conflicts. Given that the permanent members of the Council could trust one another enough to cooperate in creating specific missions for the UN, why could they not agree to include in their resolutions language pertaining to less controversial matters of human rights, or to draft TIPS resolutions? It seems highly likely that the end of the Cold War is in some ways related to changes at the Security Council, but the exact mechanisms involved remain to be explored.

Methodology

In recent years, unsupervised topic modeling has emerged as a powerful tool to find evidence for theories that had previously been difficult to validate (Grimmer & Stewart, 2013). The last two decades have seen a series of works that have enriched our understanding of statements of public officials and legal and political texts, particularly parliamentary speeches, treaties, constitutions, and documents produced by and submitted to courts (Spirling, 2012; Evans et al., 2007; Law, 2016; Diermeier et al., 2012; Slapin & Proksch, 2008; Stewart & Zhukov, 2009). Among the most advanced and fruitful ways that unsupervised text analytic models have been used is through latent Dirichlet allocation (LDA) (Blei, Ng & Jordan, 2003). Analyzing UNSC resolutions is well suited for this method, as the institution provides constant output in a standardized format that is centered

OPEN IN VIEWER

Figure 1.

LDA model

For this article, the corpus is all 2,397 UNSC Resolutions between the founding of the institution in 1946 and 2017. This article uses the ‘bag of words’ method, rather than n-grams, to determine the unit of analysis. The UNSC produces legalistic documents of the kind that do not include much in the way of concepts such as sarcasm, humor, or irony. The use of n-grams is appropriate when the order of words is important for determining context, as it often is in conversational language (Amir et al., 2016). In contrast, in legalistic documents the bag of words method is more appropriate, as drafters aim for precision in their writing (Law, 2016; Slapin & Proksch, 2008). As a first step, the names of countries, certain international institutions, and common phrases are each combined into single terms. Among these are terms such as ‘Democratic People’s Republic of Korea’, ‘United Nations’, ‘Western Sahara’, ‘Secretary General’, ‘al-Qaida’, and ‘Sierra Leone’. Certain phrases that are common parts of country names are removed, such as ‘Kingdom of’ and ‘Republic of’. Then terms are stemmed and de-capitalized, and stop words, punctuation, and numbers are all removed, before being put into a document-term matrix (see Online appendix).

Much discussion revolves around how to choose the number of topics k, with some adopting what has been called a hierarchical Dirichlet process (HDP) (Teh et al., 2005). However, this process has been criticized for finding topics that group together terms that appear

OPEN IN VIEWER

Figure 2.

Finding ideal number of topics according to four different methods

Fortunately, all methods have statistical profiles that show the process of model selection to be informative, as the models perform worse after a certain point, indicating that they avoid the problem of overfitting (Griffiths & Steyvers, 2004: 5231–5232). They also converge in agreement that the ideal number of topics for the UNSC resolution dataset is between around 100 and 130 topics, with at least one method suggesting that the ideal number of topics is closer to 150. Despite the finding that 100 or more topics would be most appropriate, it is not reasonable to conclude that the analysis cannot benefit from fitting models with a smaller number. Sometimes, models that are ‘better fits’ statistically perform relatively poorly from the perspective of the theoretical concerns of the researcher, such that it is standard practice that ‘several different numbers of topics are fitted and the optimal number is determined in a data-driven way’ (Grün & Hornik, 2011). The purpose of this article is to track changes in abstract concerns over time, and while a model that gives us individual topics of ‘North Korea’, ‘the Cyprus conflict’, and ‘the former Yugoslavia’ may be the ‘best fit’ from a statistical perspective, using such a model would provide limited insight.

In order to judge models, researchers have tended to use two methods: the cohesiveness of topics, or word intrusion, and how closely the topic composition of individual documents matches what would be expected from reading them, or topic intrusion (Chang & Blei, 2009: 81; Ponweiser, 2012: 26). This article therefore uses these methods in order to start with the smallest possible number of topics and see whether underlying concepts of interest can be found. Because of the focus of this article, a desirable model will create topics on the basis of abstract qualities and not region or any other more concrete metric. Once that is done, however, I also create a model with k = 110. This is near the lower range of the estimates given by the four methods used to select the correct number of topics in Figure 2. The k = 110 model serves as a robustness check to investigate whether similar topics of interest emerge in models that are better performing statistically and also gain more refined insight into the change in data over time.

The content of each topic is determined by finding the terms with the highest β values for each topic within

OPEN IN VIEWER

Table I.

Topics by k value

Metatopic	Punitive			War				Humanitarian
2	Punitive							Humanitarian
3	Punitive			War				Humanitarian
4	WMD/Terror	Other		War				Humanitarian
5	States	Individual		Central/West Africa		Other		Humanitarian
6	WMD/Diamond	Police		Central/West Africa	Europe	Other		Humanitarian
7	WMD/Diamond	Police		Interstate	Intrastate	NATO		West Africa Other
8	WMD	Police	Piracy/Terror	South Africa/Israel	West/CentralAfrica	Middle East	Other	Humanitarian

where β i t is how strongly word i is associated with topic t, all other topics in the model are represented by j, and k is equal to the total number of topics in the model. Only words that appear with a probability greater than .001 in any of the topics in the model are considered. All models put the entire corpus through the Gibbs sampling process 6,000 times after the initial random allocation of terms to topic. This entire process is repeated four more times for each model with a different k value, and only the model with the highest overall posterior likelihood is maintained (Grün & Hornik, 2011).

The models are also used to find the θ distribution for each document. If we allow θ d t to represent the percentage of document d devoted to topic t, we end up with the following equation:

The values β and θ can be used to judge word intrusion and topic intrusion, as the terms with the highest β values within a topic can indicate whether a topic has coherence, at which point that topic can be given a name. Once that is done, θ distributions for specific documents can be used in order to determine whether the topics most closely associated with any particular document actually appear to reflect its content. Finally, θ distributions for individual documents over time can be used to measure how the focus of the UNSC has changed across its history.

Results

Table I shows the topics that appear depending on the value of k, representing number of topics, as judged by the terms with the highest β values for each topic for models k = 2 through k = 8 (see Online appendix). I call metatopics those topics that are broken down into more specific components as k increases. We see that when k = 3, the first topic is Punitive, which involves the international community sanctioning a bad actor, including terrorist organizations and states undertaking WMD programs. War refers to an armed contest between two or more sides, whether states or factions within a country. Finally, the topic Humanitarian is associated with terms focused on making life better for individuals within a country, particularly in a peacekeeping environment. Running the same models but with more topics leads to the emergence of subtopics that are based on region or more specific versions of the main metatopics. For example, when k = 4, Punitive is broken down into one topic addressing WMDs and terrorism and another

OPEN IN VIEWER

Figure 3.

β values for 40 terms most closely associated with each topic when k = 3

The patterns are clear. The first topic, War, has as its most distinctive terms those that relate to armed conflict. This includes words referring to actions that parties are to take (e.g. ceasefir, withdraw, refrain), locations that may be in dispute (island, zone) or the combatants themselves (parti, side), and processes undertaken to deal with the aftermath of conflicts (agreement, judg). The countries in that category are ones that have been involved in conflict, whether civil wars (bosniahrzg, somalia) or international disputes (israel, southafrica, somalia). Topic 3 also contains terms that refer to war zones, but the abstract words that are most characteristic of that topic generally refer to humanitarian issues that have little direct bearing on the battlefield or international security (e.g. protect, children, women, sexual). We can call this topic Humanitarian, and its rise over time can be taken to represent the Humanitarian Turn at the Security Council. It also includes many terms relating to peacekeeping and rebuilding and reforming societies in the aftermath of conflict (polic, disarma, demobil, reintegr). One may be tempted to see Topic 3 as relating to peacekeeping rather than humanitarian issues, yet the most frequent terms in Topic 1, War, include references to several countries that have had UN peacekeeping missions, such as Somalia and Cyprus, and the missions themselves such as UNFICYP and UNISFA. Thus, Topics 1 and 3 are not distinguished by the abstract categories of peacekeeping and non-peacekeeping. Rather, Topic 1 deals directly with conflict and negotiations between parties, while Topic 3 is mostly characterized by issues relating to improving the lives of people within a state.

Finally, Topic 2, or Punitive, is composed partly of a rogues’ gallery of outcasts of international society in the postwar era, whether they have been state (dprk, iran, iraq, libya) or non-state (isil) actors. It includes the policy areas that the international community takes issue with (chemic, nuclear), along with the actions that the UNSC takes to punish or deter these bad actors (embargo, prohibit). The two most distinctive terms within the topic are shall, referring to concrete actions that the Security Council or others are required or resolved to take, and individu, reflecting the growing tendency of the institution to penalize individual actors. If we expand our focus beyond Figure 2, notable terms among those with the 100 highest β values for each topic are area, border, delay, extend, and sovereignty for War; alqaida, asset, counterterror, materiel, and transfer for Punitive; and civilian, crime, develop, transpar, and humanitarian for Humanitarian.

The categories created thus pass the test of topic coherence, meaning that they group words together in ways that make sense to an observer (Foltz, Kintsch & Landauer, 1998). Another way to investigate whether the model returns sensible results is to plot individual resolutions along the three dimensions and see whether they fall about where we would expect. This is done in the top triangle in Figure 4, where each point represents a resolution, each of which is placed into one of eight categories indicated by color and shape. Together, the points on the figure represent 460 resolutions, or about an eighth of the total corpus.

In the top graphic, the bottom left corner includes the various conflicts, the bottom right features resolutions related to terrorism and North Korea, and those focusing on women and children are near the top. UNSC resolutions focused on Liberia include components of all three topics, as the UN has engaged in peacekeeping in that country in the midst of a civil war and also sanctioned human right violators. Finally, the Indonesian struggle for independence lies between War and Punitive, as the UNSC occasionally pressured the Netherlands in language that is usually reserved for more punitive documents, such as Resolution 67, wherein the Council demanded, among other requests, that the Netherlands release political prisoners and transfer authority over to the Indonesians. The bottom triangle in Figure 4 shows all resolutions from 1992–2017 and reveals changes in topic frequency over time.

The three-topic model thus works to provide coherent topics and a reasonable way to classify individual

OPEN IN VIEWER

Figure 4.

Placement on 3D scale for selected categories of resolutions (top graphic); placement on 3D scale for all resolutions, 1992–2017 (bottom)

In the three-topic model, Punitive stays about constant, dropping off a bit in the first few decades but then resurfacing with the emergence of terrorism and the North Korean nuclear problem as issues. From the founding of the UN until the first decade of the 21st century, War is the most prominent topic at the Security Council. Finally, Humanitarian starts low and remains constant, but then shoots up immediately after the fall of the Berlin Wall. The bottom graph of Figure 5 is presented as a robustness check, showing how the topic Humanitarian varies over time when k is set to different values. Regardless of the model specification, we see nearly the exact same pattern, with Humanitarian taking off around the beginning of the 1990s. Using time-series analysis as developed by Bai & Perron (1998), the trend of the category Humanitarian in Figure 5 reveals two structural breakpoints. For each of the models, regardless of the value of k, an analysis breaking up the data by

OPEN IN VIEWER

Figure 5.

Change in various topics over time

Arguably, the results observed might be explained by changes in power dynamics that made cooperation between superpowers easier. Yet, the nature of the changes in topic distributions over time appears to contradict this simple theory. The topics War and Punitive address issues that are likely to be of greatest concern in a superpower rivalry: which state and non-state actors to sanction and how to settle disputes between countries. Great powers often disagree about whether a bad actor should be punished if one of the powers is allied with the norm violator, as can be seen in efforts to hold the Syrian government accountable for atrocities committed in recent years. They also may take different sides in a conflict, seeing any potential action in zero-sum terms. Yet, after the Cold War ended, the most contentious issues began to receive less attention relative to what seem to be less controversial topics. While it is true that the collapse of the Berlin Wall coincided with an increase in peacekeeping, the fundamental nature of peacekeeping missions changed. In the few cases that ended before 1989, ‘the primary purpose was less to prevent the resumption of war than to contain the conflict to prevent direct superpower interventions’ (Fortna, 2008: 4). Although UN peacekeepers were sent to Cyprus and the Republic of Congo in the 1960s, the resolutions authorizing and supporting those missions are closer in content to the resolutions surrounding interstate disputes than they are to modern peacekeeping resolutions.

Furthermore, if we are tempted to explain the increase in the topic Humanitarian by pointing to better relations between the USA and Russia, then we may have reason to expect the trends that began at the end of the Cold War to have reversed themselves by now due to deteriorating relations between those powers. The two countries have moved further apart since the mid-1990s, as can be seen in differences regarding NATO expansion and conflicts in the former Yugoslavia and the Ukraine, and as is reflected in voting patterns at the General Assembly (Bailey, Strezhnev & Voeten, 2017). Yet this increase in tension does not seem to have mitigated or reversed trends related to increasing Security Council output, particularly regarding the Humanitarian topic, that were set in motion with the collapse of the Soviet Union. This supports the view that international institutions are often

OPEN IN VIEWER

Figure 6.

Words per document and number of resolutions by year, 1946–2017

It is important to note that after the Cold War, it cannot be said that the Security Council began to pay less attention to War and Punitive in absolute terms. In fact, if measured by total number of words or resolutions devoted to each topic, activity has universally increased over the last few decades. Figure 6 shows the total number of resolutions passed and words per document for the period between 1946 and 2017.

Once again, the end of the Cold War stands out. After that point, the Security Council started passing a larger number of resolutions with a higher average word count. Total activity was stable from the founding of the UN up to the end of the Cold War. Interestingly, the number of resolutions has stayed relatively constant since its initial

OPEN IN VIEWER

Figure 7.

Change in topic focus over time for selected topics, k = 110

OPEN IN VIEWER

Figure 8.

Change in topic focus over time for selected topics, k = 110

Returning to Figure 2, we see that the ideal number of topics in the corpus at the lower range is between 100 and 120. This part selects the mid-point of this range, choosing a model where k = 110. As expected, doing so produces a large number of topics pertaining to specific regions and conflicts. At the same time, we see the appearance of a number of distinct abstract categories, which are as expected more specific than those in the models with smaller numbers of topics. Figure 7 shows the change in topic frequency over time for selected topics in the k = 110 model. At the top of each graph are the top ten terms that are most associated with the given topic.

We see the clear emergence of categories for Women, Children & Abuse, and Reform, the last of which contains words typically used in order to set and meet goals within an institutional framework. Again, the end of the Cold War appears to have motivated change. We also see sharp increases in Human Rights Law and Elections that correspond to the signing of the Helsinki Accords. Thus, while the Helsinki Accords do not appear to have had an impact from the perspective of models with lower k values, a more fine-grained analysis reveals that they may have been important for increasing attention paid to these two topics. This is consistent with assessments about the central role that the Helsinki Accords played in motivating human rights reforms across the Eastern bloc.

Elections in general refers to UN actions undertaken and aspirations put forth in peacekeeping missions; resolutions before the 1970s focused relatively little on the form of governments that states hosting missions should have. Because the changes that we see come near the end of the 1970s, rather than the start of the decade, we have evidence that the Helsinki Accords themselves were the cause of greater focus on human rights law and elections, rather than both being driven by the same underlying historical process. The change that resulted from the Helsinki Accords may have set the stage for the more comprehensive Humanitarian Turn that would arrive a decade and a half later.

As in the models with smaller numbers of topics, it is important to emphasize that Figure 7 addresses topic per document (tpd) frequency. Thus, considering the already noted increase in Security Council activity, the rise of the humanitarian topics is even more striking. Taking into account increased UNSC activity also reveals new topics that took off after the signing of the Helsinki Accords. Figure 8 shows trends regarding the topics Human Trafficking and Media & Medicine, the latter of which refers to groups that are expected to be protected within conflict zones. The top graphs show tpd, or relative frequency, while the bottom graphs take into account the increasing number of resolutions over time. No account is taken of changes in words per document, as doing so is not necessary to see the trends. The bottom graphs in Figure 8 thus show the total number of resolutions devoted to each topic per year.

Because tpd is a zero-sum measure, the increase in the prominence of the topics shown in Figure 7 must come at the expense of other topics. This means that using tpd to investigate change in focus over time with regards to Media & Medicine and Human Trafficking leads to a misleading picture of the interest of the UNSC in these matters. When we account for the increasing number of resolutions per year, we see more focus on these topics that began approximately around the time of the Helsinki Accords. Because these changes appear to have begun before 1975, they can be seen as part of the same process that led to the signing of those agreement. Thus, while focal point events are important, we see some evidence for a trend that was based on a more gradually forming foundation.

Conclusion

Through the use of automated content analysis, this article shows that we can trace changes in the behavior of the Security Council back to the focal point events of the signing of the Helsinki Accords and the end of the Cold War. In the future, text analysis has the potential to aid in more exploratory analysis that can help scholars develop precise theories. By showing the importance of focal point events, the findings presented here can hopefully make the project of studying norm diffusion in international politics more manageable by suggesting a temporal focus. Researchers who are interested in tracing the Humanitarian Turn at the Security Council but unsure about where to focus their attention would likely be best served by beginning with an investigation into the time periods surrounding the two focal point events highlighted in this article. The second breakpoint found for the topic Humanitarian, taking place in the mid-2000s, was not predicted by the theory developed in this article. This demonstrates how text analysis can, in addition to providing support for existing theories, open up new areas of exploration for social scientists and suggest novel theories. Although it is speculative at this point, it is possible that that the war on terror ushered in an era of nation building that is reflected in UNSC resolutions.

The Humanitarian Turn can also be understood in a broader context. In recent years, scholars have made strong arguments that humanity has seen moral progress (Hathaway & Shapiro, 2017; Pinker, 2011). Support for this view is provided in the form of quantitative evidence on the decline of violence and qualitative work on changing practices over time. Automated text analysis provides yet another method to understand these kinds of historical processes. One might always suspect that violence and other undesirable practices have declined due to economic, technological, or other material reasons (Waltz, 1990). Text, however, is perhaps the most direct indicator of what people are thinking in any given time and place. Using it to make empirical arguments is difficult, however, as there is often too much information to meaningfully review, in addition to inherent challenges in interpretation. Here, automated text analysis can supplement historical inquiry by processing large amounts of data and finding underlying patterns. When the results match what we suspect from the historical record, such findings can give us more confidence regarding theories of how international society develops.

In arguing for the importance of focal point events, this article has implications not only for scholars but also for activists and others who would like to effect change in the practices of states and international organizations. Inertia may be the natural state of politics and bureaucracies, yet once in a while we are faced with events that have the potential to create new social realities. The story behind the Humanitarian Turn at the Security Council suggests that the future is often made by those who are able to seize such opportunities.