Fighting in the First Year

The Korean War erupted after years of escalating tensions following World War II. Under Japanese occupation since 1910, the Korean peninsula was mutually occupied and divided by the Soviet Union and the United States in 1945. The two states split the peninsula along the thirty-eighth parallel and took control of the territory north and south of this line, respectively. Figure 6.1 depicts the peninsula and its geographic division. By 1948, both sides established their own governments. The Democratic People’s Republic of Korea (DPRK, informally North Korea) took root in the North, while the Republic of Korea (ROK, informally South Korea) was established in the South. Each regime had the support of a different superpower and claimed to govern the entire peninsula.³ On June 25, 1950, North Korean troops launched a sudden offensive across the entire parallel. The UN Security Council declared this an invasion and called for a ceasefire on the same day. North Korean troops did not observe this request. Beyond harboring a fundamental suspicion of the United Nations, Kim Il-sung, chairman of the DPRK, had anticipated accomplishing victory within a month,⁴ and he therefore had no interest in heeding the Security Council’s appeals.

Historical treatments of the conflict divide it into four distinct phases, of which the first three took place in the opening year. The first phase occurred between June 25 and September 14, 1950, as the North Korean People’s Army (KPA) engaged in a relatively swift southward sweep down the peninsula. Troops of the Republic of Korea were poorly equipped and had no means to deal with the tanks or heavy artillery at the invaders’ disposal. US troops, few in number, were also unprepared for an attack and did not have the appropriate weaponry to stage a proper defense. The Truman administration had placed far more of its focus on Europe and also considered Japan to be the critical factor to containing China and the Soviet Union in East Asia. Unable to quickly recover from these handicaps, South Korean and US forces were in a state of constant retreat and driven all the way down to Busan, a port city on the southern tip of the Korean peninsula.⁵ On July 7, the Security Council adopted Resolution 84, which recommended that member states provide military forces for a “unified command” that would be led by the United States.⁶ The United Nations Command (UNC) was formally established on July 24, 1950. One week later, ROK president Rhee Syng-man transferred operational command of the ROK Army to the United States and hence the UNC. Given that South Korea was not a member of the United Nations at that time, this was a significant decision.

The second phase began on September 15, 1950. The UNC, under the command of General Douglas MacArthur, was tasked with restoring the prewar status quo. This effort started with a surprise amphibious landing in Incheon, a town near Seoul and the thirty-eighth parallel. As a result, the UNC cut off supplies and communication between the North Korean government and KPA troops that had moved into the South. The Incheon landing greatly exceeded the UNC’s expectations. The KPA fled back to the thirty-eighth parallel in only two weeks and offered very little resistance as the battle front swung back up the peninsula. The UNC revised its original goal of restoring the prewar status quo, instead opting to eliminate the DPRK altogether. UNC forces thus continued to press their advantage by moving toward the Yalu River, which was on the border of the People’s Republic of China (PRC), by late October, under the belief that Kim Il-sung’s regime in the DPRK would surrender while China would stay out of the conflict. Riding high on optimism, the UNC planned and initiated a “Home-by-Christmas” offensive on November 24.⁷

The UNC was incorrect and unknowingly entered the third and most tumultuous phase of conflict.⁸ At the instruction of Mao Zedong, Chinese forces had clandestinely infiltrated the peninsula in October and November before staging their first major offensive on November 25.⁹ The sudden introduction of over three hundred thousand People’s Volunteer Army (PVA) troops reshaped the battlefield. Over multiple battles culminating in the massive New Year’s Offensive in January 1951, UNC forces were impelled to cede control of Seoul and withdraw to the thirty-eighth parallel, engaging in some of the worst retreats in US military history.¹⁰ By March, the UNC reclaimed the initiative by launching Operation Ripper, which was designed to eliminate as many Communist forces as possible from the vicinity of Seoul and to get UNC troops back to the thirty-eighth parallel. The operation successfully reclaimed Seoul and reached the thirty-eighth parallel but did not take down many Communist forces, who had rapidly retreated. In April and May, the PVA responded with the large Spring Offensive that pressed against the entire front. Chinese forces pushed UNC forces back but were also spread too thin to complete their objective. The UNC’s bombing campaign and counterattack, called Operation Strangle, began on May 20 and reversed the PVA’s gains. By mid-June, lines of control had started to stabilize around the thirty-eighth parallel.

Losses during the first year were severe. Counting all those killed, captured, wounded, and sick, the United States amassed 73,600 casualties, while South Korea sustained about 180,000. The Communist states, however, suffered upward of 1,100,000.¹¹ Facing a clear disadvantage in armaments and air power, North Korea and China compensated through greater willingness to withstand heavier costs than their enemy. Yet by this point, leaders of both sides privately realized that neither side could land a decisive blow that would lead to total victory and thus sought some form of restoration of the prewar status quo.¹²

The Path to Negotiations

Both the United States and the Communist states had contemplated negotiations several months before mid-1951. As my theory highlights, however, all belligerents were deeply concerned that showing interest in talks would create perceptions of weakness, which in turn would embolden the adversary to push harder at both the negotiation table and the battlefield.

During the United States’ darkest moments in the war in December 1950, the Truman administration contemplated whether to start ceasefire talks.¹³ In a memorandum submitted on December 3, George Kennan and Secretary of State Dean Acheson argued vociferously against diplomacy, concluding that any attempt at a negotiated ceasefire “would probably be taken by [Communist] leaders as a bid for peace by us on whatever terms we can get. They would regard this as confirmation that we were faced with the alternative of capitulation, on the one hand, or complete rout and military disaster on the other… . Any approach we make to them without some solid cards in our hand … may simply be exploited by them for purposes of spotlighting out weakness and improving their own position.”¹⁴

During a meeting the following day, additional officials, including Dean Rusk, agreed that this was “the poorest time possible for any negotiations” on the basis that talks would make the United States look like it was suing for peace.¹⁵ Around the same moment, Mao and Chinese prime minister Zhou Enlai conferred with the Soviets about prospects for negotiations. Stalin stridently opposed talks, arguing that they would help the UNC to feel that conditions were improving and to “win time” to fight more aggressively.¹⁶

Third-party pressure to negotiate weighed heavily on all belligerents’ minds during the first year of the war. UN secretary-general Trygve Lie made numerous attempts to initiate talks between the warring parties—talks that were complicated by the fact that the UN was technically an active participant in the war. At the same time, the Nehru government in India exerted significant diplomatic pressure. The first Indian attempt at peace began in July 1950 but ultimately went nowhere. By December 7, India led a coalition of thirteen Arab and Asian states to submit a proposal to begin peace talks. The proposal was soon adopted by the UN General Assembly and led to the creation of a three-party ceasefire committee involving representatives from the UN, India, and Canada. These diplomatic moves concerned both the Communist states and the United States, neither of which wanted to stop fighting in late 1950.

The fact that the two sets of belligerents made efforts to deflect blame, rather than to ignore peace proposals outright or refuse them without explanation, shows that third-party pressures were indeed salient. On December 7, the PRC stated that one of its preconditions for talks was the immediate withdrawal of all foreign troops from the Korean peninsula. This demand, which was formulated in consultation with the Soviet government, was made under the belief that the United States would not agree to it despite how reasonable it would sound to the international community.¹⁷ In a meeting with Lie on December 9, Chinese representatives stressed that China wanted peace and that any blame for continued conflict rested with Washington’s unwillingness to make any useful proposals.¹⁸ In mid-January 1951, the three-party ceasefire committee established a month earlier submitted a report and proposal for peace talks. While the United States accepted the proposal’s terms, China mistakenly believed that global sentiment was still behind the Communists and thus rejected the deal. China’s refusal was a political windfall for Washington.¹⁹ Riding a new swell of international frustration with the PRC, Truman’s administration went on a public diplomatic offensive, culminating with the passing of UN General Assembly Resolution 498, which accused the PRC of aggression and unwillingness to reach peace. Communist leaders were outraged by the development and openly argued that the resolution, which was pushed by the United States, was responsible for killing any prospects for talks. Notably, at the same time that Washington was working to pass this resolution, UNC forces initiated a major counteroffensive called Operation Thunderbolt, which brought Seoul back under UNC control for the remainder of the war.

The emerging and shared reality of a military standoff in June 1951 marked the start of the war’s fourth phase. In the face of a mutually costly stalemate, and reverting back to lines of control that hovered around the thirty-eighth parallel, both sides privately decided that outright military victory was beyond their reach and that negotiations were necessary.²⁰ Oriana Skylar Mastro avers that the costly military stalemate around the thirty-eighth parallel may have demonstrated that any appeal to negotiate should not be interpreted as weakness.²¹ While this may be largely true, a letter written by Mao in June 1951 stated that China ought to wait for the UNC to make the first appeal for talks.²² The White House, while publicly shifting its rhetoric on Korea to a more limited scale, had no immediate plans to make such an offer.

A third party of sorts proved critical to providing political cover and moving the process forward. On June 23, 1951, during a broadcast of the United Nations’ weekly radio program, The Price of Peace, Soviet deputy foreign minister and United Nations delegate Jacob Malik made remarks encouraging both sides to consider “a cease-fire and an armistice providing for the mutual withdrawal of forces from the thirty-eighth parallel.”²³

Tellingly, each side’s response to Malik’s appeal highlights the extent to which all states sought to distance themselves from any interpretation of this proposal being their own idea. Two days after Malik’s radio statement, the People’s Daily, the official newspaper of the Chinese Communist Party, coolly endorsed the Soviet proposal. North Korean radio broadcasts followed suit two days later.²⁴ The United States’ response came after almost a week of internal discussions about how to accede to the negotiation offer without looking weak or generating raw material for Communist propaganda.²⁵ On June 29, General Matthew Ridgway—who had assumed command of all UNC forces after Douglas MacArthur was relieved of command—broadcast a highly vetted message to the Communist states: “I am informed that you may wish a meeting to discuss an armistice providing for the cessation of hostilities… . Upon the receipt of word from you that such a meeting is desired I shall be prepared to name my representative.”²⁶ The Chinese government waited until July 2 to officially agree to talks.²⁷ The two sides soon agreed to meet at Kaesong—a historical capital of the Goryeo dynasty, which was the first kingdom to rule over a unified Korean peninsula between the tenth and fourteenth centuries. At this point in the war, Kaesong was in territory under Communist control but was deemed a neutral zone for purposes of negotiations.

To be sure, third parties may not have been necessary for talks to eventually occur. Both belligerents privately expressed interest in peace multiple times throughout the war, and each would have likely agreed to talk of their own accord due to exhaustion. Truman began to reconsider negotiations in March 1951, and the administration became more interested in a ceasefire over the following months.²⁸ Meanwhile, both Zhou Enlai in the PRC and Kim Il-sung in the DPRK privately noted in 1951 and 1952 that negotiation was necessary because their respective states could not bear the economic and human costs of indefinite war.²⁹ Yet it is very likely that the talks that did come to pass began prematurely, before the two sides learned enough information to reach a common set of expectations about the war’s future. Two pieces of evidence are consistent with this conclusion. First, in preliminary discussions to set up negotiations, both parties agreed that hostilities would continue alongside talks. Neither the UNC nor the Communists trusted that their opponent would truly stop fighting while discussing an armistice, and each harbored significant concerns that the war could tip away from their favor if they inadvertently engaged in a unilateral ceasefire. Even though the costliness of the military stalemate made the warring parties more open to negotiations, it was clear that both sides still had the capability and at least some political will to keep fighting and change the state of the battlefield.

Second, and as will become evident in the next section, the two sides quickly realized the enormous gaps that existed—and would exist to varying degrees for two additional years—between their bargaining positions. In accordance with Hypothesis 2, talks that emerge from heavy third-party pressure occur relatively frequently but often do little in terms of reaching peace.

Fighting and Negotiating in the Second and Third Years

Each set of belligerents assembled a team of individuals with military credentials to negotiate in Kaesong. On the UNC’s side, Vice Admiral C. Turner Joy, commander of the US Naval Forces, Far East, was appointed as the senior delegate.³⁰ He was backed by an array of US officers, as well as Major General Baek Son-yeop from the ROKA. Tellingly, no ROKA official, including Baek, spoke on behalf of the UNC during talks.

The Communist delegation comprised military officials from both China and North Korea. Unlike their UNC counterparts, the Communist representatives also had pasts in civilian politics. Lieutenant General Nam Il of the KPA, who had been deputy minister of education before moving to the Ministry of Defense, was the chief negotiator for the Communists during meetings with the UNC. But the true head of the Communist delegation was Li Kenong from the PRC. Li had rich political and military experience from the Chinese Civil War several years prior. His effectiveness and loyalty to the Chinese Communist Party earned him positions as both vice minister of foreign affairs and director of the army’s Military Intelligence Department by the time the Korean War began. He was thus granted wide authority to coordinate battlefield and negotiation strategy during armistice talks. He managed the Communists’ diplomatic strategy, never entering the negotiation tents himself but supervising day-to-day affairs from a separate tent and reporting to Mao and Zhou in Beijing.³¹

Formal negotiations to arrange an armistice began in Kaesong on July 10, 1951. The UNC delegation was not fully prepared for how the Communists would immediately exploit the trappings of negotiation for propaganda. Because Kaesong technically remained behind Communist lines, the Communist states’ media widely distributed pictures of the UNC delegation entering the city in a car bearing white flags as evidence that the UNC was suing for peace.³² Early flare-ups regarding seating arrangements, the presence of news media, freedom of movement within the neutral zone, and the like typified the constant struggle to control the tone and momentum of talks.

The first order of business was determining the items to be discussed during negotiations. The Communist delegation sought a more politically charged docket, while the UNC delegation wanted to limit talks to strictly military affairs. By July 25, the delegations agreed on the following five agenda items:

• Adoption of the agenda
• Fixing a military demarcation line between both sides so as to establish a demilitarized zone as a basic condition for a cessation of hostilities in Korea
• Concrete arrangements for the realization of a ceasefire and an armistice in Korea, including the composition, authority, and functions of a supervising organization for carrying out the terms of a ceasefire and armistice
• Arrangements relating to prisoners of war
• Recommendations to the governments of the countries concerned on both sides

The delegates then moved on to Item 2. The Communists hoped to designate the thirty-eighth parallel as the demarcation line for an armistice, which was the prewar status quo. The UNC, however, had pushed the actual line of contact between military forces slightly past this point and thus refused to revert to the thirty-eighth parallel.

In August, armistice negotiations began to adopt a tiered structure that would become commonplace for the remainder of talks. Facing a deadlock on Item 2, both delegations sought a different environment where they could be more candid and less performative—in other words, more sincere. Negotiations to this point had all taken place at the plenary level, featuring a formal and more scripted tone that was aimed at public messaging. On August 15, Joy suggested moving talks to a “subdelegation” level that would exclusively address Item 2 in a less stilted setting. The Communists agreed to this proposal. According to the historian Walter G. Hermes, the first subdelegation meeting seemed somewhat promising: “The first subdelegate discussion took place on 17 August, and although no concrete progress resulted, the atmosphere was more relaxed. General Hsieh seemed to like this type of exchange. He spoke frequently and acted as a moderator when the comments became sharp. As the talk flowed back and forth around the small table, there was even a tendency on the part of the Communists to consider the demarcation line on the map.”³³ This shift is very interesting, especially since the same individuals often attended different levels of talks. Discussions continued at the subdelegation level until August 23, when the Communist delegation accused the UNC of flying planes over the neutral zone around Kaesong—an erroneous accusation—and suspended negotiations. The Communists likely did this to express their dissatisfaction with the UNC’s unwillingness to use the thirty-eighth parallel as an armistice line, as well as to nullify the international media’s coalescing view that the Communist delegation was anxious and constantly backing down.³⁴

Consistent with Hypothesis 4, the first period of negotiations was notable for how it temporarily muted the battlefield. In his opening remarks on July 10, C. Turner Joy affirmed what both sides had already accepted: “Hostilities will continue in all areas, except in those neutral zones agreed upon, until such time as there is an agreement.” On July 30, General Nam Il of North Korea reiterated that “hostilities would continue during negotiations.” Yet despite this mutual understanding, hostilities slackened substantially over the first several weeks. This was true for two reasons. First, neither side wanted to needlessly lose lives or resources while assessing whether a peace deal was possible. General Ridgway stated this in his memoir: “While both sides had immediately agreed that hostilities should continue during negotiations, it seemed to me, with a cease-fire faintly visible on the horizon, that I should do all I could to keep our losses at a justifiable minimum. I notified our commanders therefore that we would conduct no major offenses.”³⁵ Second, and at the same time, all sides remobilized and rearmed their troops to address two possible contingencies: that the opponent would exploit talks to prepare a new offensive if negotiations failed and that troops could no longer be moved or replenished if an armistice was reached.

These preparations were brought into action once the Communists suspended talks. The weeks between late August and late October featured several significant UNC offensives, which were designed to force the Communists to return to the bargaining table. Writing about the state of affairs on September 12, General James Van Fleet, commander of the US Eighth Army (which was the commanding formation of the US Army), reported that the Communists “were in bad shape, and we are hurting them more and more. They will want peace before winter before we are through with them.”³⁶ While the UNC suffered significant casualties during these two months, the manifold higher losses the UNC inflicted on the Communists proved crucial to the resumption of talks on October 25.³⁷

Subsequent discussions were moved to a more neutral location—a precondition the UNC demanded for any talks to resume. After some exchanges, the newly chosen site was Panmunjom, which was in decidedly neutral territory several miles southeast of Kaesong. By late November, Item 2, regarding the establishment of a demarcation line, was settled via a series of mutual compromises. The Communists gave up their demand of using the thirty-eighth parallel as the demarcation line. A demilitarized zone around the line of contact at the time, which crept significantly north of the thirty-eighth, would instead be established.³⁸ As a concession, the UNC proposed that the line of contact would be used as the demarcation line for an armistice if the Communist delegation agreed to an overall agreement in the next thirty days; it would be renegotiated if no armistice was struck within one month. As some US officials (including Joy) feared, however, Communist forces merely used this thirty-day period to reinforce their positions and resupply their forces.³⁹ All subsequent fighting became positional in nature, involving limited but strategically and symbolically important changes in territorial control.

Discussions on Item 3, which involved matters regarding the ceasefire, began on November 27. Questions revolved around three main issues: what countries would be part of a Neutral Nations Supervisory Commission (NNSC), what ports of entry and airfields would be permitted for use after the war, and what number of troops would be allowed to rotate in and out of the peninsula on either side. Multiple impasses led the delegations down to the subdelegation level on December 4, and talks moved to another rung that was newer, lower, and more congenial—the staff officer level—on December 20.

The issue of the NNSC vividly demonstrates how progress at the bargaining table not only stalled at points but was even steered backward, standing at odds with the notion of negotiations following a convergence or war-of-attrition dynamic. On February 16, 1952, after almost three months of discussion on the matter, the Communist states abruptly nominated the Soviet Union as one of their choices for a neutral nation that would oversee an armistice. The proposal stunned the UNC. It was no secret that the Soviets were providing material aid and that Stalin was constantly conferring with Mao and Kim about wartime strategy. But by 1951, the Truman administration also knew that Soviet pilots were secretly fighting on behalf of the Communists.⁴⁰ The UNC immediately rejected the nomination of the Soviet Union, surmising that the Communists were trying to stall talks and gain bargaining leverage.⁴¹ This allowed the Communist delegation to counter that “opposition on your part will only inevitably lead to show that your side is attempting not to resolve the question.”⁴² This fabricated deadlock, which would eventually be resolved once the Soviet Union was replaced by India in early May 1953, was soon overshadowed by the next agenda item.

Negotiations ground to their most severe halt in early 1952 over disagreements about how to repatriate prisoners of war (POWs), which fell under Item 4.⁴³ Many POWs detained by the UNC were Nationalist Chinese who had been coerced into serving in the PVA, as well as South Koreans who had been pushed into the North Korean People’s Army when Communist forces had overtaken the southern half of the peninsula. The UNC argued that these POWs would face grave consequences if repatriated to Communist China or North Korea, presented evidence that many POWs did not wish to be sent back,⁴⁴ and thus supported a policy of voluntary repatriation. For his part, President Truman opposed the notion of forced repatriation not only on strategic grounds but also on moral ones.⁴⁵ The Communist delegation angrily refused, accusing the UNC of violating the 1949 Geneva Convention—which mandated that POWs should be repatriated once hostilities end—and brainwashing the prisoners.⁴⁶ Two uprisings of Communist POWs at a camp in Koje-do in February and March resulted in several dozen North Korean deaths, which caused great embarrassment for the UNC. The Communist delegation would spend the next several months vociferously accusing the UNC of wanton slaughter. On April 28, with talks returning to the plenary level after months of stagnation at the subdelegation and staff officer levels, the UNC proposed what it stated was its “final and irrevocable” package proposal for an armistice, which offered to repatriate only a fraction of Communist POWs. The Communist delegation flatly refused the offer, and talks devolved further. After numerous fruitless exchanges where the repatriation question remained the one outstanding issue, the UNC unilaterally ceased negotiations on October 8, 1952.

Historians identify two key factors that may have led both sides to return to the negotiating table in April 1953. One was the January inauguration of President Dwight D. Eisenhower, who as a candidate had openly mentioned the possibility of introducing atomic weapons to the conflict.⁴⁷ The second was the death of Stalin in March 1953. The Soviet Council of Ministers, more concerned with their grip on power than with a proxy war, directly called on North Korea and China to announce their interest in resolving the deadlock.⁴⁸ The UNC and Communists agreed to an exchange of some sick and wounded POWs in late March and performed the exchange in April. Formal talks at the plenary level resumed in Panmunjom on April 26. After a month of ineffectual discussion, the UNC stated its (truly) final position on May 25, offering some concessions to the Communists but ultimately adhering to its refusal to repatriate all POWs.⁴⁹ After expressing vehement objections, the Communist states accepted the offer on June 4. The path to an armistice was temporarily and forcefully derailed on June 18, when ROK president Rhee Syng-man, who was upset with any arrangement that left Korea divided, released about twenty-five thousand anti-Communist POWs from the camps.⁵⁰ The Communist states expressed their anger in Panmunjom and launched a last-minute offensive in mid-July to punish the ROK and make some final territorial gains before agreeing to still move forward with peace. The war came to a close once the final armistice agreement was signed on July 27, 1953.

A Deeper Strategy to Negotiations

Numerous studies of the Korean War have effectively argued that the onset of stalemate in mid-1951 helped both sides choose to come to the bargaining table. The arrival of Eisenhower and the exit of Stalin were critical reasons why negotiations, which had ended in October 1952, resumed in April 1953. I do not dispute these existing claims.

These findings, however, have also invited a relatively cursory analysis of the final two years of the conflict. Reviews of Korean War diplomacy often end or greatly accelerate once they reach the start of talks in July 1951, suggesting that both sides gradually eked out mutual concessions until leadership changes in the United States and Soviet Union propelled talks across the finish line.⁵¹ Moreover, the fact that fighting on the battlefield became more geographically limited over the course of negotiations compared to the dramatic swings over the first year has further encouraged scholars to heed less attention to the impact of fighting on the two sides’ diplomatic behavior. Such characterizations of the war and its associated negotiations may not be wrong in the aggregate, but they still assume away the highly strategic and dynamic manner in which battlefield activities influenced belligerents’ negotiation behavior, as well as the benefits to analyzing it.⁵² Such choices, implicit or not, shortchange our understanding of conflict.

The fact that fighting became more geographically limited after talks began does not mean that it became unimportant or subdued. Both sides actively fought over specific battlefield objectives that had significant strategic or symbolic importance, all in hopes of shaping a diplomatic agreement in their favor.⁵³ Even though clashes became narrower in the spatial sense, they involved significant casualties and thus effort from both sides. Almost half of the fifty deadliest days for the UNC occurred after negotiations began,⁵⁴ and China amassed half its casualties during the same time frame.⁵⁵

Moreover, both sides’ attitudes toward diplomacy and settlement were complex, nonlinear, and not indicative of either a convergence or war-of-attrition dynamic. The Communist delegation’s nomination of the Soviet Union to the NNSC, where progress on Item 3 was intentionally backpedaled, is a case in point. The deadlock regarding POW repatriation also did not trend in a single direction. Even though the issue was not fully resolved until June 1953, the two sides held numerous negotiation sessions that ranged between substantive policy disagreements and acrimonious mudslinging.

This mudslinging was meant not only to attack the opposing delegation but to steer international public opinion regarding the war and the belligerents. The vast majority of negotiations were not private and were thus fair game for media reporting. Each side exploited this situation to frame itself as the magnanimous party interested in peace while portraying its adversary as the real enemy of peace. For example, in the New York Times on June 13, 1952, an article reporting on deadlocked negotiations quoted both delegations at length. The Communists accused the UNC of having an “outrageous attitude in negotiations conducted by both sides on an equal footing,” while the UNC responded that the Communists’ “daily insistence on ‘negotiations’ is nothing but a reiterated demand for further concessions by our side… . When we meet you have nothing to say you haven’t said before.”⁵⁶

Emotional, psychological, and other situational factors must have certainly influenced the approach that delegates took at Kaesong and Panmunjom. Nevertheless, both the UNC and the Communist states committed significant time and effort to sculpting the words they would use during negotiations, as well as how those words would be said. Records indicate that statements made during negotiations—as truculent and bombastic as they would eventually become—were the product of collective decision-making and had a keen awareness of recent battlefield activities. Admiral Joy emphasized this point in his memoir:

The United Nations Command delegation followed a practice of “staffing” all formal statements uttered in an armistice conference by delegates. Each day staff officers prepared a number of proposed statements for use by the delegates. These were considered and discussed by the delegates and staff officers in meetings at our camp in Munsan, before proceeding to [the negotiation site] for the day’s events. The statement finally worked out was almost never the work of any one individual. It was the product of careful editing by all delegates and final approval by the Senior Delegate. Thus the benefit of all the fine intellects to the delegation was used to the fullest.⁵⁷

The Communist delegation had an analogous approach to working out its diplomatic stance. Each night at ten o’clock, Li Kenong and other members of the delegation gathered to discuss the day’s talks, learn about updates from the battlefield from radio operators who were part of the delegation, and determine subsequent negotiation strategies in light of these developments.⁵⁸

The negotiators themselves were also fully aware that armistice talks were not necessarily meant to resolve the conflict but could also help encourage it. In describing the Communist delegation’s tactics, the UNC’s chief psychological warfare adviser, William Vatcher, provided a description of negotiations that my theory would consider to be insincere:

Having been unsuccessful in attaining their objectives on the Korean field of battle, [North Korea and China] turned to the conference table as a means of achieving their ends. Their use of the conference table was obvious indeed: to gain precious time while they rebuilt and strengthened their forces, to obtain every possible benefit from the UNC, and to serve as a sounding board for their propaganda. This they attempted to achieve by haggling over the agenda, demanding the UNC withdraw to the 38th Parallel, manufacturing incidents and pointing to the UNC as instigator, maliciously injecting propaganda into the substance of the meetings in order to create a false impression of UNC perfidy, presenting irrelevant issues for stalling purposes, and frequently acting in a very discourteous and arrogant manner.⁵⁹

Vatcher’s view of Communist intentions could be understandably biased, but archival evidence indicates that the Communists indeed sought to exploit negotiations for political gain. Chinese and Soviet sources released decades after the war indicate that China went into talks in 1951 with the overarching objective of stalling for time in order to bolster the Communist troops’ battlefield positions.⁶⁰ Additionally, until his death in March 1953, Stalin spent the entirety of negotiations encouraging the Communist delegation to adopt a stiff bargaining position because any prolongation of the war would drain US resolve, strain the United States’ relationships with its allies, allow the Soviets to spy on the US military, and give the Communists an avenue to publicly accuse the US of war crimes for its intense bombing of North Korea.⁶¹

The UNC was also guilty of engaging in foot dragging and pursuing side effects.⁶² During a visit to the peninsula in late 1951, Chairman of the Joint Chiefs of Staff Omar Bradley and State Department chief adviser Charles Bohlen discussed the pros and cons of attaining a peaceful settlement. Both concluded that the UNC was doing well on the battlefield and that there was no urgency to reach an agreement with the Communist states. Even so, they also believed that the ongoing talks should continue, largely with the goal of placating the numerous allies of the United States that urged for diplomatic dialogue.⁶³ Private correspondence between Mao and Stalin in November 1951 also observed that “the Americans are dragging out the negotiations” in order to perpetrate “a policy of an advance in the course of negotiations.”⁶⁴

Despite these tactics, the two delegations managed to arrange an armistice agreement that remains in effect to this day. How did the opposing parties ever soften their bargaining positions and exert any diplomatic effort to make an agreement? The belligerents’ answer was simple: gaining leverage by fighting. Both sides commonly understood that changing fortunes on the battlefield, even if slightly more limited and positional, were critical to sculpting bargaining positions and extracting concessions from the other side. Immediately before talks began, Li Kenong told his delegation to constantly pay attention to the battlefield in order to adjust negotiating tactics.⁶⁵ While discussing Item 2 among themselves, Chinese elites argued that the primary way to get more favorable terms was to make gains on the battlefield to generate leverage at the bargaining table.⁶⁶ In accordance with the recommendations of the Joint Chiefs of Staff, the UNC simultaneously planned to amass battlefield successes in order to place pressure on the Communist delegation to more seriously consider an armistice.⁶⁷ Many battles were fought and even more lives were lost during negotiations with these goals in mind.

The summary above reflects the fact that many historical reviews of and anecdotes from the Korean War have already described the general ebbs and flows of hostilities on the peninsula and negotiations at the bargaining table. No study to this point, however, has analyzed the relationship between battlefield activity and bargaining behavior in a fine-grained manner, where one might investigate whether and what developments from fighting on one day were directly followed by any changes in negotiation behavior on subsequent days. Descriptions of negotiation behavior during this war are frequently characterized in terms of phases where the two delegations would or would not engage in insincere behavior. To be sure, talks did undergo stages of sustained acrimony. But times of bitter disdain also featured moments of limited progress; times of productive discussions also featured moments of rancor. These more subtle vacillations can provide tremendous insight into warring parties’ diplomatic strategies. Sincerity exists on a spectrum, but our ability to leverage and understand variation on that spectrum relies on the creation of precise data that track not only negotiation behavior but the battlefield activity that informed it.

How can we systematically capture whether negotiations were sincere or insincere in nature, as well as how contemporaneous hostilities interacted with this choice, in an exact manner over the course of these two years? And to what extent do any of these dynamics align with my theory’s specific predictions about the role of information in negotiation dynamics? I tackle these questions by turning to a novel technique: computational analysis of archival documents from the war.

Measuring Battlefield Activity

On practically every day of the conflict, the UNC operations staff submitted reports about the status of each military unit and the overall progress of the UNC. In many cases, the primary military unit is the regimental combat team (RCT), which was a major infantry unit of the US Army in World War II as well as the Korean War. A total of 1,102 reports were filed, with the majority of reports being between six and ten pages in length.⁶⁹

United Nations Command

One practical aspect of the Korean War is that most military movements went up and down the peninsula. Combined with the fact that the military is already a highly bureaucratized body, the daily operations reports use standardized vocabulary to describe the status of individual military units. A small set of terms are consistently utilized to express gains, losses, or stasis that UNC RCTs experienced each day. Table 6.1 supplies several examples of these entries.

The most common terms used to express gains, losses, and stasis from the perspective of the UNC are summarized below.⁷⁰

Figure 6.2 illustrates what proportion of all status terms each day reflects the UNC’s gains or losses over the entire conflict. The two lines correspond well with our understanding of the broad ebbs and flows of the war as I have previously described.⁷¹ The first spike in gains reflects the Incheon landing of September 1950. The second and third jumps in gains correspond with Operations Ripper and Strangle, respectively, which produced significant gains for the UNC across the entire front.

Table 6.1 Examples of entries from UNC reports

Date	Entry	Type
August 10, 1950	Advanced during early daylight hours against light opposition.	Gain
January 8, 1951	Improved positions on line D in the area north of ANSONG CR4696.	Gain
September 23, 1951	Maintained psns along line CS0492—CS 0794—FS0996 with elms in assy area vic CS0789.	Stasis
October 7, 1952	Outpost elms vic DT0542 received atk from estimated 2 enemy plats at 062200I and withdrew slightly.	Loss
June 2, 1953	No change.	Stasis

Note: Key terms italicized for emphasis.

Table 6.2 Common UNC terms expressing gains, losses, and stasis

Gain	Loss	Stasis
Advance	Enemy capture	Assemble
Capture	Enemy occupy	Maintain
Improve Occupy	Withdraw	No change Remain

The three most prominent waves of losses also reflect the UNC’s grimmest moments during the war. The first surge aligns with North Korea’s sweeping offensive down the peninsula from July to September 1950, which met little resistance before the UNC landed in Incheon. The second major loss, from November 1950 to January 1951, reflects the entry of Chinese forces and the massive New Year’s Offensive. The third spike, in April and May 1951, captures the Chinese Spring Offensive, which Operation Strangle managed to stem.

It is notable that the UNC operations reports contain far more mentions of gains than of losses. Given that the war started and eventually ended at the thirty-eighth parallel, we may expect the number of terms indicating gains and losses to be roughly equivalent. While this may raise issues regarding whether the UNC underreported the absolute magnitude of its losses, the fact that these measures align with broadly understood characterizations of the Korean War increases our confidence that the measures dependably capture everyday and less dramatic battlefield developments.

The conclusion of each UNC operations report also provides a detailed breakdown of the number of troops supplied by each coalition member state, as well as each country’s casualties. Casualties involve three different categories: killed in action (KIA), wounded in action (WIA), and missing in action (MIA). I use these data to create an exponentially weighted moving average of the logged number of UNC casualties over the previous two weeks. Figure 6.3 illustrates this measure, which is central to the forthcoming analysis. Despite the overall stagnation in battlefield movement beginning in 1951, we see frequent swings in casualties over the course of the entire conflict. This divergence speaks to the importance of distinguishing different types of battlefield activity.

Measuring Negotiation Behavior

Stenographers from the UNC were present at all negotiations and transcribed the proceedings. The transcripts total seven thousand pages in length and capture 14,123 statements made by the UNC and Communist delegations over the course of 713 separate meetings. These documents provide the most direct insight possible into both sides’ negotiation behavior.

I perform several steps to create measures of how sincere or insincere each delegation is. First, I split long statements into three-hundred-word segments. This is useful to help capture variation in negotiation behavior during more protracted remarks. Second, I apply standard text preprocessing to each segment. This includes removing punctuation, numbers, words shorter than three letters, and stop words (such as a, the, so, in, and the like) and converting the remaining words into tokens, which are the basic unit of meaningful text. For instance, words like discussion, discussed, discuss, and discussing are all converted into the basic token discuss. I then identify common multiword expressions in the transcripts (such as united nations command, unreasonable demand, raise the question, etc.) and add tokenized versions of these words to the comprehensive list of terms.

Throughout negotiations, 9,639 unique terms are used. Many of these terms are used only once or twice, adding very little to our understanding of talks while inflating the size of the dataset and the intensity of the computation required to analyze them. To make the analysis more tractable without losing much information, I focus on the 1,120 terms that appear most frequently in the negotiations. After removing segments that do not have associated battlefield measures, I end up with a document-term matrix (DTM) with 16,460 rows, each of which reflects a single segment, and 1,120 columns, each of which contains a count of the number of times a unique term was used in a segment. The entire DTM captures about 529,000 words or phrases uttered throughout negotiations.⁷³ I thus utilize what is often called a “bag of words” approach that does not account for grammar or word order aside from the inclusion of some common phrases; I simply count how many times different terms appear in a segment. As will become evident, even this simplified representation of the speech segments produces informative measures of negotiation behavior.

For each segment, the objective is to create a binary coding for whether the segment likely indicates sincere or insincere negotiation behavior. A segment is considered to be sincere when it solely discusses concrete aspects of an agenda item. A segment is coded as insincere when it also or only exhibits overt hostility or obstinacy. Examples of this latter category include accusations of neutral zone violations; denunciations of propagandizing, lying, or delaying during talks; and allegations of mass illegal violence. Note that speech acts that reflect disagreement can still be coded as sincere as long as they discuss actual policy proposals.

Several examples will help to provide useful context. On November 10, 1951, the UNC representative in the Item 2 subdelegation meeting continued discussions of how to draw a demarcation line between opposing forces. The following quotation is coded as indicating sincere negotiation behavior:

First I would like to take up the arguments in regard to Kaesong. General Hsieh mentioned some of the arguments and said that they were all refuted. We do not believe they have been refuted. In addition to the arguments he mentioned there is the one of equivalent adjustment. The adjustment in the Kaesong area which we recommended in other parts of the line. They are compensatory adjustments and are approximately equal in magnitude, equal in withdrawal and equal in range. Now I will take up our 8 November proposal and the difference between it and our 5 November proposal, and why these differences exist. We offered the 8 November proposal in the spirit of compromise. We felt that our original proposal of 25 October was fair, but because we do want to negotiate a military armistice, we were willing to compromise in order to attain that armistice as quickly as could be done and still have reasonable terms.⁷⁴

Even though the segment mentions disagreement regarding how to draw this demarcation line, it involves direct discussion of an agenda item. Conversely, during a plenary session on August 3, 1952, the UNC responded to Communist bluster in the following manner, which is coded as reflecting insincere behavior:

It is noted that you have said nothing new and nothing that can lead toward achievement of an armistice. Some of your language this morning is what we in civilized countries associate with common criminals or persons who through ignorance or stupidity are unable to speak logically and convincingly. In their frustration, they resort to efforts to insult. You should know by now that such talk serves no useful purpose in negotiations and we do not propose to engage in it with you. Therefore, if you expect to conduct discussions, it would be advisable to use more restrained language. As for the issue itself, it is clear. We have nothing to propose.⁷⁵

The Communist delegation’s segments also vary widely. During a staff-level meeting regarding Item 3 on February 14, 1952, the Communists spoke in great detail regarding the wording of a specific proposal, which is coded as being sincere:

Very well. Regarding sub-paragraphs 13c and 13d, and your suggestion in particular to add the phrase “have the right to”—to insert this phrase in the sentence about the conduct of supervision and inspection by the Neutral Nations Supervisory Commission, we still hold that in order to ensure really that the military personnel, and weapons and ammunition, introduced into Korea during the armistice are those permitted by the agreement, the Neutral Nations Inspection Teams shall conduct supervision and inspection, but not just shall “have the right to” conduct supervision and inspection. Regarding sub-paragraph 13i, we have studied your revised draft submitted yesterday. It is our opinion that at present neither side is holding complete data pertaining to the burial of its deceased military personnel in the area of the other side.⁷⁶

On the other hand, during a plenary-level session regarding prisoners of war on June 17, 1952, the Communists began by expressing disagreement on a substantive proposal but quickly veered into vitriol, leading to a statement coded as insincere:

You have only the obligation to repatriate war prisoners but no right to retain war prisoners. You had better pack up forthwith all your nonsense about screening and re-screening. In your attempt to whitewash your criminal acts of slaughtering war prisoners, you trumpeted on the one hand about your so-called humanitarian principles, and resorted on the other hand to fabrication and calumny. I tell you that such efforts of yours are futile. There cannot be on this earth humanitarian principles of slaughtering and retaining war prisoners. Your crimes of murdering war prisoners have already been recorded on the pages of history and are not to be deleted.⁷⁷

Manually coding over sixteen thousand segments for negotiation behavior is impractical for a variety of reasons. A fully manual process is more prone to error and inconsistencies as the number of human coders (or the hours they spend on this task) grows. It also does not provide flexibility if codings need to be adjusted, since every single segment would need to be reviewed once again. Such an approach cannot reasonably scale upward for similar tasks that involve many more observations. For those reasons, I instead code the vast majority of speech segments using a supervised learning approach.⁷⁸ To do so, I first draw a random sample of four hundred segments from each delegation and manually classify these eight hundred total segments for whether each represents sincere (1) or insincere (0) negotiation behavior as defined and illustrated above. Approximately 30 percent of the segments in my sample were coded as reflecting insincere behavior.

These hand-coded classifications, along with the DTMs of the associated segments, are necessary for the statistical learning process. A majority of the hand-coded segments are chosen at random to be part of a training dataset, while the remainder are part of a test dataset. The training data are inputted into a series of candidate statistical learning models, each of which finds the best predictive relationship between the speech act’s composition of tokens and the hand-coded classification. Once these models are trained to predict the relationship between the inputs (token counts) and the outcomes (hand codings of negotiation behavior), they are used to predict the outcomes in the test data. These predictions are then compared to the original hand codings to create an unbiased measure of the model’s performance with new out-of-sample data.

Out of multiple models tested, a balanced random forest model featured the best overall out-of-sample performance.⁷⁹ A random forest is an ensemble learning method where numerous simpler decision trees, which search for relationships between an observation’s features and its target variable, are combined to create a predictive model that can be applied to other data.⁸⁰ This balanced random forest is applied to the entire set of speech segment data to produce a prediction of whether each speech act exhibits sincere or insincere negotiation behavior.⁸¹ Finally, I combine the predictions for all segments from the same speech act to create a single measure of sincerity for each statement. The final dataset includes 13,526 statements, of which 8,266 (61 percent) are sincere and 5,260 (39 percent) are deemed insincere.⁸²

Validating the Measures

A series of predictive metrics based on cross-validation techniques speak favorably to the balanced random forest’s ability to predict sincerity on out-of-sample data not used to train the original model. This is encouraging, but the external validity of these predictions is arguably more important and cannot be assessed using quantitative data alone. Three pieces of evidence indicate that this predicted measure offers a historically credible reflection of negotiations. First, as previously alluded to, qualitative descriptions of the talks suggest that both sides spoke more substantively at lower levels of talks. The most “refreshing” discussions with “an air of serious intent to make progress” on agenda items, as Walter G. Hermes put it, occurred at the staff officer level.⁸³ My quantitative data also bear out this claim. For each separate meeting, I determine what proportion of each delegation’s speeches are sincere. Table 6.3 depicts the distributions of these calculations. The jump in negotiation behavior is most obvious between subdelegation and staff officer meetings. Chi-squared tests show that differences in sincerity between plenary and staff meetings, as well as subdelegation and staff meetings, are highly statistically significant for both the UNC and Communist delegations (p << 0.01). Liaison, control, and investigatory meetings (presented under the umbrella term Other), which were mainly for logistical planning and issues not related to the agenda, are even more agreeable. Importantly, the level of talks was not included as a variable in any of the predictive models. The fact that these differences still appear is compelling evidence that the supervised learning method has created a meaningful measure.

Second, we know that Item 4, relating to POWs, was the largest obstacle to agreement, being the only agenda item left to address between May 1952 and June 1953. The majority of the most embittered days of negotiation should thus involve this issue. My data support this conclusion. Out of the twenty-four days with the highest average level of insincere negotiation behavior (which represent the worst 5 percent of negotiation days), twenty-two feature discussions of Item 4. Eighteen of these involve only Item 4, while the other four also include thorny discussions of Item 3.

Table 6.3 Distribution of sincere negotiation behavior by delegation and meeting level

Meeting level	UNC	Communists
Plenary	0.475	0.358
Subdelegation	0.479	0.469
Staff	0.609	0.587
Other	0.730	0.681

Finally, the general ebbs and flows of negotiation behavior in my predicted data align well with our qualitative accounting of progress made during talks. Figure 6.4 offers a visual representation of the data, tracking what proportion of statements made by each delegation each day were sincere. Three observations are worth mentioning. First, the initial dip and rise in the first round of talks between July 10 and August 23 attest to the delegations’ initial litany of bitter exchanges and posturing at the outset of talks, which eventually gave way to more meaningful discussions regarding Item 2 before the Communists accused the UNC of violating the neutral zone and called off negotiations. Second, the next round of talks exhibits a downward drift beginning around March 1952. This aligns with the period during which Item 4, on POWs, became the single—and, as I have explained, most controversial—issue. The lowest values for sincere negotiation behavior emerge in the weeks immediately prior to the UNC’s unilateral recess. Third, the dramatic upward swing during the final phase of talks in 1953 indicates the two delegations’ ability to reach an agreement regarding POWs by late May. It bears noting again that negotiation behavior does not consistently trend upward in terms of being sincere. While sincerity does not necessarily lead to agreement, the nonmonotonic nature of observed negotiation behavior still undercuts the idea that wartime negotiations inexorably bring the two sides’ divergent positions together.

One potential challenge to my coding procedure, illustrated by a couple of pieces of evidence earlier in this chapter, is that actors may make proposals that seem sincere at face value but are designed to be unacceptable to the enemy. This can help actors better feign interest in settlement and thus accrue more beneficial side effects. When read without deeper context or understanding of political circumstances, a speech act containing this type of proposal could be erroneously classified as sincere when it is made with insincere motives. The aforementioned incident where the Communist delegation nominated the Soviet Union as a “neutral nation” to supervise the armistice is arguably the most prominent example of this tactic.⁸⁴

Note, however, that even if one individual speech act such as this may be misclassified, the broader set of speech acts on this topic promptly discern subsequent changes in negotiation behavior. Recall that I code negotiation behavior based on the presence of hostile or intransigent language, not by whether any mentioned policy is acceptable. As long as speech acts discussing these policy proposals contain such language, which they typically do because the parties will recognize and discuss the other side’s dilatory intentions, they are coded as reflecting insincerity. I indeed see this in discussions about the topic of neutral nations. Prior to the nomination of the Soviets, approximately 59 percent of all speech acts related to neutral nations are classified as being insincere. The figure soars to 81 percent after the nomination.⁸⁵ As such, this act of avoidance bargaining led to insincere behavior, despite the fact that the initial proposal may have been deemed sincere according to my coding criteria and supervised model. Moreover, false negatives like this case are vastly outnumbered by the far more overtly antagonistic statements that the two delegations exchanged, which are more explicit reference points for insincere negotiation behavior.

These correlations between the quantitative data and the historical record bolster the empirical validity of the data, allowing us to proceed to fuller statistical analysis.