JOURNAL ARTICLE REVIEW – Multiparty Negotiation

Each student will select one of the key terms listed  below and conduct a search of our University’s online Library  resources to find 1 recent peer-reviewed article (ATTACHED TO THIS POST) that closely relate to the concept. Your paper must include the following information in the following format:

Key Terms:

Multiparty Negotiation

DEFINITION: a brief definition of the key term followed by the APA reference for the term; this
does not count in the word requirement.

SUMMARY: Summarize the article in your own words- this should be in the 150-200-word
range. Be sure to note the article’s author, note their credentials, and why we should put any
weight behind his/her opinions, research, or findings regarding the key term. Name the title of
the article and the author(s).

ANALYSIS(300-350 words): 

1. Explain why you selected this particular article among all the articles you could have chosen on your selected term.

2. Explain why you agree or disagree with the author’s key positions in the article. 

3. Explain how the article was easy or difficult to understand and why? 

4. What did the author do well in your opinion? Explain.

5. Describe what you believe the author could have done better in your opinion?

6. What else should the author have included in the article and would the article benefit from a different perspective (such as from a different nationality or different industry or experience perspective). Explain. 

7. What other sources or methods could the author have used to improve the research in the article?  (Hint: look up the types of qualitative and types of quantitative research methods). 

8. What information / in-depth study / or further research should the author focus on as a follow up to this article and why? 

9. Explain what audience would gain the most benefit from your selected article and how they could apply it in their professional lives. 

10. What did you personally gain from this article and how has it shaped your thinking on the topic?  

11. What are the conflicting or alternative viewpoints of the author’s position? Or  What additional research backs up and confirms or adds to the author’s position?  (Hint: this will require you to find another peer-reviewed article that challenges, confirms, or adds to, or provides a different perspective to your chosen article.)

Cognitive Science 41 (2017) 259–271
Copyright © 2015 Cognitive Science Society, Inc. All rights reserved.
ISSN: 0364-0213 print / 1551-6709 online
DOI: 10.1111/cogs.12325

Language Use and Coalition Formation in Multiparty
Negotiations

Eyal Sagi,
a
Daniel Diermeier

b

aPsychology Department, University of St. Francis
bHarris School of Public Policy, University of Chicago

Received 6 March 2014; received in revised form 19 May 2015; accepted 15 September 2015

Abstract

The alignment of bargaining positions is crucial to a successful negotiation. Prior research has

shown that similarity in language use is indicative of the conceptual alignment of interlocutors.

We use latent semantic analysis to explore how the similarity of language use between negotiating

parties develops over the course of a three-party negotiation. Results show that parties that reach

an agreement show a gradual increase in language similarity over the course of the negotia

tion.

Furthermore, reaching the most financially efficient outcome is dependent on similarity in lan-

guage use between the parties that have the most to gain from such an outcome.

Keywords: Negotiation; Coalition formation; Linguistic entrainment; Psycholinguistics; Latent
semantic analysis

1. Introduction

Negotiations are usually conducted through the use of language. For negotiations to be

successful the negotiating parties need to build common ground and represent the issues

of the negotiation similarly. Pickering and Garrod (2004) suggest that such a convergence

in representation is often the result of successful dialog accompanied by a convergence in

patterns of language use. Language similarity, however, can also be the consequence of

strategic behavior. In this article, we examine how similarity in language use evolves

throughout the course of three-party negotiations and compare the language used by par-

ties that reach agreement and those that are excluded from it.

Negotiations between multiple participants are more complex than two-party

negotiations (Bazerman, Curhan, Moore, & Valley, 2000). This is especially true when

Correspondence should be sent to Eyal Sagi, Psychology Department, University of St. Francis, 500

Wilcox St., Joliet, IL 60435. E-mail: esagi@stfrancis.edu

an agreement requires only a subset of the negotiating parties, in which case being

excluded from an agreement is an ongoing concern for each negotiator. Partial coalition

agreements, however, are often less desirable than agreements that involve the group if

they do not allocate all of the available resources. However, even being part of a partial

agreement is preferable to not reaching an agreement or being excluded from an agree-

ment reached by others (Diermeier, Swaab, Medvec, & Kern, 2008).

The added complexity of multiparty negotiation has been shown to affect the patterns

of language use in such negotiations. Following the framework of communication accom-

modation theory (cf. Giles, Willemyns, Gallois, & Anderson, 2007), Huffaker, Swaab,

and Diermeier (2011) demonstrate how the formation of a coalition is affected by specific

aspects of the language used by the negotiating parties. Specifically, they find that part-

ners to a coalition show more similarity in language use than participants who were not

part of a coalition. The use of assents was also found to correlate positively with partici-

pating in the coalition agreement. In contrast, the use of negative emotion words was a

detrimental predictor to being a part of a coalition. These results are congruent with

empirical findings in psycholinguistics which show that in successful dialogs the repre-

sentations and language used by dialog partners tend to converge over time (e.g., Brennan

& Clark, 1996; Pickering & Garrod, 2004). Even intentionally created similarity in lan-

guage use can facilitate agreement. For example, Swaab, Maddux, and Sinaceur (2011)

asked participants to mimic the distinctive linguistic patterns of their negotiation partners.

In particular, participants were asked to use the same jargon, abbreviation, and emoticons

used by the other participant in the negotiation. Swaab et al. (2011) found that such

intentional mimicry was effective when used early in the negotiation, but not in later

stages.

Consequently, there are two possible patterns of language similarity that might lead to

an eventual coalition: First, if strategic mimicry is an effective tool for reaching agree-

ment, we might expect that early similarity in language use might lead to the forming of

a coalition later on. Second, if similarity in language use is a consequence of a gradual

alignment of representations and the emergence of a mutual understanding, then language

similarity should increase over the course of the negotiation and ultimately result in the

parties reaching an agreement.

It is also possible that both of these factors play a role. In that case, we would expect

to find not only that eventual coalition partners show more similar language early on than

non-coalition partners, but also that this difference increases over time.

While Huffaker et al. (2011) show that similarity in language use is correlated with

the outcome of the negotiation, they use the entire negotiation as their unit of analysis.

Consequently, their results do not explore whether such similarity exists early in the

negotiation or whether it develops throughout its course. In this article, we measure lan-

guage similarity between adjacent dialog moves, a measure that allows us to track

changes in language similarity over the course of the negotiation.

As mentioned earlier, reaching an optimal agreement in a multiparty negotiation often

requires the formation of a grand coalition, a coalition that includes all of the parties.

However, reaching this kind of agreement can be difficult and may require a high level

260 E. Sagi, D. Diermeier / Cognitive Science 41 (2017)

of cooperation between the parties (Diermeier et al., 2008). Consequently, we are inter-

ested not only in when the parties use similar language, but also which parties’ exhibit

similarity in language use when more than two of them participate in the final coalition.

Specifically, cooperation between participants increases their bargaining power and allows

participants with weak bargaining positions, who might otherwise be excluded from the

final coalition, to strengthen their position. Consequently, we hypothesize that participants

in weaker bargaining positions have the most to gain from cooperating in order to

achieve a grand coalition. Such a partial coalition should result in increased similarity in

language use between these weaker parties.

1.1. Measuring similarity in language use

The measure of language similarity we use in this article is based on the latent seman-

tic analysis (LSA) cosine similarity of a pair of utterances. Such a measure has been used

in the past as a measure of textual coherence (Foltz, Kintsch, & Landauer, 1998; McNa-

mara, Cai, & Louwerse, 2007) and as a measure of linguistic entrainment (Huffaker, Jor-

gensen, Iacobelli, Tepper, & Cassell, 2006). Measures based on LSA have also been used

to explore a diverse set of psychological phenomena related to texts and language, such

as semantic priming (Chwilla & Kolk, 2002; Landauer & Dumais, 1997), the acquisition

of knowledge from texts (Wolfe et al., 1998), the representation of moral concerns (Sagi

& Dehghani, 2014), and conceptual framing (Sagi, Diermeier, & Kaufmann, 2013).

Latent semantic analysis vectors for individual words are generated based on the co-

occurrence patterns of words in large corpora. These vectors identify points in a high-

dimensional space.
1
The more likely two words are to co-occur with similar words, the

closer they will be in the space. For example, the vectors for sun and moon are fairly
close together and show a cosine similarity of .53, whereas man and moon are not very
similar and show a cosine similarity of .03. Moreover, when several word vectors from a

single utterance are combined together, as was done in this study, the result identifies a

point in space that represents the overall topic of the utterance.

It is important to note that this kind of automatic measure disregards certain linguistic

elements that a human coder might use. For instance, the use of negation is generally

ignored, while sarcasm and metaphors are often misrepresented. However, since we are

interested in the convergence of language use—that is, whether participants are using
similar and/or related words and terms to convey their (sometimes conflicting) ideas, this

type of analysis seems sufficient.

2. Method

2.1. The corpus

The data used in this article come from a study reported by Huffaker et al. (2011).

They patterned their study after a coalition game developed by Raiffa (1982). In that

E. Sagi, D. Diermeier / Cognitive Science 41 (2017) 261

study, 180 MBA students were divided into 60 three-person groups. The participants in

the study had some experience in managerial decision making and negotiation. Within

each group, participants were assigned to one of three roles (A, B, C) and instructed that

they were to use an online chat room to negotiate a split of that payoff among them-

selves. Participants were unaware of the identities of the other participants in the negotia-

tion.

All participants were provided with the payoff table in advance of the negotiation (see

Table 1) and informed of their role in the negotiation and the role of the other partici-

pants. As is evident from the table, different coalition formations received different pay-

offs, and if no coalition was formed all participants received a payoff of zero. The

participants were allowed to negotiate how the payoff was distributed between them.

These payoff options provided incentives for the participants to join up with another par-

ticipant in a partial coalition and then jointly take advantage of the resulting weak bar-

gaining position of the third participant. However, the payoff table was designed so that

the third player left out of a partial coalition could always make an attractive offer to one

of the members of the initial coalition to induce a defection from the preliminary agree-

ment. Consequently, participants were incentivized not only to be a part of a forming

coalition, but also to ensure that it is a stable coalition and that their partner(s) would not

defect. Only then were they able to use their bargaining power to extract a favorable out-

come from the remaining third party. If members of a two-party coalition were unsure

about the reliability of their respective coalition partner, they may have refrained from

engaging discussions with the third party for fear of giving the “out party” an opening to

split the current coalition. In this case, the negotiation resulted in a partial coalition leav-

ing some money on the table, compared to the most efficient coalition, which allocated

the largest monetary award among all parties. Consequently, the ideal strategy for each

participant was to quickly form a stable two-party coalition and squeeze the remaining

party into accepting an unfavorable deal. Maintaining such a position of strength, how-

ever, required maintaining the loyalty of the coalition partner in the process.

Each session of the experiment lasted an hour. Participants in the experiment were

placed at computers in different rooms so that their only means of communication with

Table 1

Payoff table in the negotiation game from Huffaker et al. (2011)

Possible Agreements Total Payoff

A alone $0

B alone $0

C alone $0

A and B $118,000

A and C $84,000

B and C $50,000

A, B, and C $121,000

Notes. A, B, and C represent the participants in the negotiation. The payoff is split between the parties
that reach the described final agreement according to their agreement, not necessarily evenly.

262 E. Sagi, D. Diermeier / Cognitive Science 41 (2017)

each other was through the provided chat software. They logged into a public chat room

to begin the negotiation process. The negotiation ended when at least two participants

reached an agreement and declared it as final, or the allotted time had elapsed.

The software allowed participants to move from the public chat room to three private

chat rooms. That is, participant A could move into one of the private chat rooms together

with participant B so that they could negotiate without participant C being privy to the

content of the negotiation. However, all participants were alerted whenever a participant

entered or exited a chat room so that the excluded participant was always aware that the

two other participants might be negotiating in private. This mimics some of the real-

world aspects of a negotiation, where parties are often able to communicate in private,

but the fact that they communicate in private is common knowledge. A private exchange

of information can also provide an indication that the two parties are forming

a coalition.

While it is theoretically possible for the two parties to exclude a third party completely

from the negotiation process, such exclusionary behavior rarely occurred. In particular,

utterances in the public chat room were fairly evenly distributed across the quarters of

the negotiation and there were generally more utterances in the public chat rooms than in

the private rooms.

Sessions lasted 33–318 turns (M = 110.10, SD = 53.31). The most common coalition
was AB (24 times). No agreement was reached in five sessions. Most sessions lasted

about 40–60 min, while the shortest session lasted about 20 min and concluded with a
grand coalition.

2.2. Semantic analysis

The analysis in this article is based on the transcripts of the negotiation experiment

conducted by Huffaker et al. (2011). Each submitted line in the chat was considered a

separate utterance. An LSA vector was computed for each individual utterance by using

vector addition to combine the vectors of all of the content-bearing words in the utter-

ance. When an utterance did not include any content-bearing words, a null vector was

used to represent it. The vector space used for this analysis was generated by Infomap

(Sch€utze, 1997; Takayama, Flournoy, Kaufmann, & Peters, 1998) using the written por-
tion of the British National Corpus (BNC Consortium, 2007).

2
We then computed the

correlation of vectors representing non-null adjacent utterances that occurred in the same

chat context (i.e., when they occurred in the same chat room, whether public or private).
3

Fig. 1 presents some examples of utterance pairs and their correla

tions.

The first few utterances were of an introductory nature, mostly with participants report-

ing “I am present.” These utterances consequently resulted in identical vectors for all

three participants that inflated the correlations at the beginning of many sessions. Conse-

quently, we elected to discard utterances that included the words “present,” “hello,” “hi,”

“morning,” and “welcome.”
4
There were 253 such utterances (3.9% of the corpus), all of

which were greetings and occurred within the first 10% of each session. Moreover, 196

(77.5%) of these occurred within the first five utterances of a session and 124 (49%) are

the formulaic utterance “I am present.”

E. Sagi, D. Diermeier / Cognitive Science 41 (2017) 263

In order to measure language similarity, we categorized the utterance pairs based on

the two participants that contributed to them. We predicted that participants who were

included in the final coalition would use more similar language with each other than with

participants who were excluded from the coalition. For example, if an AB coalition was

ultimately reached, utterance pairs between A and B would be predicted to have more

similar language use (i.e., utterance-to-utterance correlation) than those between A and C

or B and C. Consequently, we divided the utterance pairs into those in which both partic-

ipants were included in the final coalition (intracoalition utterance pairs) and those in
which at least one of the participants was excluded from the coalition (extracoalition
utterance pairs). Importantly, when the final coalition included all parties, all of the utter-
ance pairs were considered to be intracoalition pairs. In contrast, when no agreement was

reached, all of the utterances were considered to be extracoalition pairs.

We were also interested in examining how the difference in language similarity

between intracoalition and extracoalition utterance pairs evolved over time. Therefore, we

divided the utterance pairs based on their position in the negotiation. We used a relatively

Fig. 1. Sample utterance pairs from three sessions and the computed correlations between them.

264 E. Sagi, D. Diermeier / Cognitive Science 41 (2017)

coarse grain division of time (quarters) because some of the discussions consisted of rela-

tively few utterances (under 50).
5
We categorized each utterance pair based on the quar-

ter of the negotiation in which the first utterance of the pair occurred.

3. Results

We tested three distinct hypotheses:

1. Following accounts of linguistic entrainment (e.g., Pickering & Garrod, 2004), we

hypothesized that coalition formation will be accompanied by the alignment of lan-

guage use. On the basis of this hypothesis we predicted that as a coalition comes

together, the language used by its participants will increase in similarity. Because

forming a coalition takes time, we expected that this increase in similarity was more

likely to occur later in the negotiation. Moreover, this increase in language similarity

should not affect language use by excluded parties. This hypothesis therefore pre-

dicted that as the negotiation proceeds, the similarity of intracoalition utterance

pairs, but not extracoalition ones, should increase.

2. Following the literature on the effectiveness of strategic mimicry in negotiations

(e.g., Swaab et al., 2011), we hypothesized that a coalition might be more likely to

form when one of the participants intentionally created linguistic similarity with

another participant during the initial stages of the beginning of a negotiation. While

this type of linguistic similarity would have been intentional, it might have led the

second party to assume that it indicated a shared perspective and made a protocoali-

tion between the two parties more likely. If this type of intentional manipulation is

an effective tool in these negotiations, intracoalition utterance pairs should have

been more similar than extracoalition ones early in the negotiation.

3. On the basis of Diermeier et al.’s (2008) theory of coalition formation, we hypothe-

sized that a protocoalition involving participant A will be less likely to result in a

grand coalition (ABC) than one that involves only participants B and C. In the nego-

tiation game used by Huffaker et al. (2011), a protocoalition that involves partici-

pants B and C can increase its payoff by $71,000 by adding participant A to the

final coalition. In contrast, protocoalitions that involve participant A have a substan-

tially lower incentive to make the effort to incorporate the remaining participant

because the increase in their payoff is substantially lower ($3,000 for an AB proto-

coalition, and $37,000 for an AC protocoalition). After all, protocoalitions have to

balance the marginal payoff increase from adding the third player with the risk that

the third player may try to break up the protocoalition. Consequently, a grand coali-

tion is more likely to result from a BC protocoalition than either an AB or AC one.

We therefore predicted that the pattern of language similarity for a grand coalition

should be similar to that of a BC coalition. That is, a grand coalition should demon-

strate a pattern where BC utterance pairs show a higher degree of similarity than

AB or AC ones.

E. Sagi, D. Diermeier / Cognitive Science 41 (2017) 265

To test the first two hypotheses, we used a mixed model with the type of utterance pair

(intracoalition vs. extracoalition) and its position in the session (first through fourth quar-
ters) as the independent variables. The experimental session was included as a random

variable. The dependent measure was the average utterance-to-utterance correlation.

Fig. 2 plots the mean utterance similarity over time.

As expected, intracoalition utterance pairs (M = .19, SD = .17) showed more language
similarity than extracoalition ones (M = .10, SD = .13; F(1, 286) = 20.93, MSE = 0.02,
p < .0001). Overall utterance pair similarity (combining over utterance pairs from all par- ties) increased over time (F(1, 286) = 9.21, MSE = 0.02, p < .01). Most important, the difference in language similarity between intra- and extracoalition utterance pairs

increased over time (F(3, 286) = 13.71, MSE = 0.02, p < .001). To further explore this interaction, Table 2 presents the results of planned comparisons of language similarity for

intra- and extracoalition utterances for each quarter. Moreover, intracoalition utterance

pairs showed an increase in language similarity over time (F(3, 145) = 18.11,
MSE = .024, p < .0001), whereas no such increase was observed for extracoalition utter- ance pairs (F(3, 102) < 1, n.s.).

The gradual increase in language similarity over the course of the negotiation that was

observed for intracoalition pairs is congruent with accounts in which a gradual alignment

in language use and semantic representation is related to the likelihood of forming a

coalition (Hypothesis 1). However, we found no evidence to support accounts in which

early similarity in language use leads to the formation of a coalition (Hypothesis 2).

Fig. 2. Similarity of language use by utterance pair type and time for negotiation. Error bars represent

standard error.

Table 2

Planned comparisons of similarity in language use between intra- and extracoalition utterance pairs

Quarter df F MSE p

1 1, 29 <1 .007 .6831 2 1, 26 3.99 .010 .0564

3 1, 25 7.28 .013 .0123

4 1, 28 14.66 .037 .0007

266 E. Sagi, D. Diermeier / Cognitive Science 41 (2017)

Huffaker et al. (2011) found that the use of assents, but not negations, was related to

the formation of a coalition. Since such use might also affect linguistic similarity, we

tested whether assents and negations affected similarity using a mixed model with the

individual utterance-to-utterance similarity as the dependent measure. The independent

measures were position, type of pair, and the use of assents and/or negations. We deter-

mined whether assents or negations were used by identifying whether words from the

appropriate LIWC dimensions (Linguistic Inquiry and Word Count; Pennebaker, Chung,

Ireland, Gonzales, & Booth, 2007) were present in the utterance pair.
6
The experimental

session was included as a random variable. The similarity of language use also increased

when assents were used (F(1, 1570) = 21.16, MSE = 0.06, p < .0001) and marginally decreased when negations were used (F(1, 1570) = 2.85, MSE = 0.06, p = .09). In the case of assents, this effect increased over time (F(1, 1570) = 3.86, MSE = 0.06, p < .05). There was no significant interaction between the type of pair and the use of either negations or assents. Most important, like the previous analysis, this model identi-

fied an interaction between the type of pair and the position of the utterance even after

controlling for the effects of assents and negations (F(1, 1570) = 11.28, MSE = 0.06,
p < .001).

We now turn to Hypothesis 3. To test this hypothesis, we compared the similarity in

language use of the various pairs in sessions that resulted in a grand coalition to those that

resulted in an AB coalition. We discarded sessions that resulted in other coalitions because

there were not enough data to use them to conduct a meaningful analysis—there were only
10 sessions that resulted in an AC coalition and 4 that resulted in a BC coalition.

We used a mixed model with the participants of utterance pair (AB, AC, or BC), the

utterance pair’s position the negotiation session (first through fourth), and the resulting

coalition (AB or ABC) as the independent variables. As stated earlier, the dependent

measure was the average utterance-to-utterance correlation and the session was included

as a random variable. Fig. 3 shows the mean utterance similarity for each participant pair

for an AB coalition, and Fig. 4 shows the mean utterance similarity for each participant

pair for an ABC coalition.

Fig. 3. Similarity of language use by utterance pair participants and time for negotiations that resulted in an

AB coalition. Error bars represent standard error.

E. Sagi, D. Diermeier / Cognitive Science 41 (2017) 267

The analysis yielded two statistically significant interactions—First, the three-way
interaction between utterance pair participants, position within the negotiation, and the

resulting coalition was significant (F(2, 314) = 9.96, MSE = 0.03, p < .0001). More important, the expected two-way interaction between the utterance pair participants and

the resulting coalition was also significant (F(2, 314) = 7.56, MSE = 0.03, p < .001). Planned comparisons revealed that these interactions were most likely due to an increase

in the language similarity between participants B and C during the later stages of negotia-

tions that resulted in an ABC coalition (F(1, 66) = 10.60, MSE = 0.03, p < .01) and an increase in the language similarity between participants A and B in the later stages of

negotiations that resulted in sessions that resulted in an AB coalition (F(1, 106) = 8.56,
MSE = 0.03, p < .01). Moreover, the initial positions of the three possible pairs were not statistically distinguishable (F < 1 in both cases). No difference in language similarity between A and C was found between negotiations that resulted in ABC coalition and

those that resulted in an AB coalition (F(1, 67) = 1.10, n.s.).

4. Discussion

The analysis presented here, based on data collected by Huffaker et al. (2011), demon-

strates that as agreement forms during a negotiation, the language used by the parties

involved becomes more similar. Moreover, parties that are excluded from the coalition do

not show this pattern of convergence. This result has implications for theories of linguis-

tic entrainment as it demonstrates that merely being party to a linguistic exchange is not

enough. This suggests that linguistic similarity is representative of the alignment of repre-

sentation rather than merely a result of exposure to linguistic input.

Our results are also congruent with Diermeier et al.’s (2008) model of coalition for-

mation, in which building trust between parties plays a pivotal role, provided that lan-

guage similarity, as a measure of shared perspective, is indicative of trust between the

parties. Following this interpretation, trust appears to gradually emerge over the course

Fig. 4. Similarity of language use by utterance pair participants and time for negotiations that resulted in an

ABC coalition. Error bars represent standard error.

268 E. Sagi, D. Diermeier / Cognitive Science 41 (2017)

of a negotiation, and it is a reliable predictor of the ultimate outcome of the negotia-

tion. Moreover, our results are consistent with an interpretation that trust between the

participants with the most to gain from a grand coalition is the key to achieving such

a coalition.

We found no evidence for the effectiveness of early similarity in language use in

directing the course of a negotiation. However, this could be a consequence of the

design of the negotiations we used. Research that demonstrates the effectiveness of early

similarity in language use often relies on an explicit and strategic use of repetition and

mimicry by the participants (e.g., Swaab et al., 2011). In contrast, the participants whose

negotiations we analyzed in this article were not instructed to use such strategies. While

strategic use of mimicry may be effective, we found no evidence that this strategy was

used.

It is important to note that the negotiation task we analyzed in this article focuses on

the split of a monetary payoff. While this task is commonly used to simulate negotiations

in economics, many real-world negotiations involve various non-monetary stakes (e.g.,

work practices) resulting in a more complex negotiation task. Nevertheless, the fact that

our results show evidence of linguistic alignment even in negotiations with minimal con-

tent suggests that the effects should be even more pronounced in negotiations that involve

more diverse interests. Interestingly, as the statement from participant B in the AB coali-

tion in Fig. 1 shows, some participants did make arguments that used non-monetary

values (e.g., “key technology”), although they were infrequent.

Finally, our analysis illustrates that multiparty negotiations, while more complex than

two-party negotiations and dialogs, follow many of the same patterns as their simpler

counterparts. However, the added dynamics of such a negotiation allows researchers to

examine topics, such as the role of similarity in language use in reaching agreement and

the influence of the availability of private communication channels on the course of a

negotiation, that are often difficult to explore when only two parties are involved in a

linguistic exchange.

Acknowledgments

We thank David Huffaker and Roderick Swaab for sharing the data from their study.

Notes

1. In accordance with the accepted practices for Infomap (http://infomap-nlp.source-

forge.net/), we used 100 dimensions in our analysis.

2. This corpus was selected because it represents a well-balanced sample of the Eng-

lish language. Nevertheless, the choice of corpus does not make a large difference

and a qualitatively and quantitatively similar pattern of results was obtained using

a space derived from a corpus of New York Times articles (Sandhaus, 2008).

E. Sagi, D. Diermeier / Cognitive Science 41 (2017) 269

http://infomap-nlp.sourceforge.net/

http://infomap-nlp.sourceforge.net/

3. Because the first dimension of LSA vector spaces tends to correlate with the fre-

quency and length of the text, it was dropped from the analysis (cf. Hu, Cai, Wie-

mer-Hastings, Graesser, & McNamara, 2007).

4. These are common and formulaic terms used at the beginning of the conversation

to announce arrival. As such, they are generally used by all participants and do not

constitute mimicry as defined by Swaab et al. (2011). Furthermore, including these

terms inflates the correlations in the first part of the negotiations but does not other-

wise change the results of the analysis presented below.

5. This coarse division is adequate for our hypotheses. It is possible to utilize smaller time

units in an analysis of this type to gain further insight into the temporal progression of

the negotiation. Essentially, the choice of temporal units for analysis represents a trade-

off between precision and statistical power. Nevertheless, similar results can be obtained

by analyzing individual utterance pairs based on their relative position in the session

rather than aggregating them based on the quarter in which they appear.

6. LIWC is a tool that uses the frequency of occurrence of various words to estimate

related linguistic and psychological variables, such as the use of assents and nega-

tions.

References

Bazerman, M. H., Curhan, J. R., Moore, D. A., & Valley, K. L. (2000). Negotiation. Annual Review of
Psychology, 51(1), 279–314. doi:10.1146/annurev.psych.51.1.279

BNC Consortium. (2007). The British National Corpus, version 3 (BNC XML Edition). Oxford University
Computing Services. Available at http://www.natcorp.ox.ac.uk/. Accessed May 19, 2015.

Brennan, S. E., & Clark, H. H. (1996). Conceptual pacts and lexical choice in conversation. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 22(6), 1482–1493. doi:10.1037/0278-
7393.22.6.1482

Chwilla, D. J., & Kolk, H. H. J. (2002). Three-step priming in lexical decision. Memory & Cognition, 30(2),
217–225. doi:10.3758/BF03195282

Diermeier, D., Swaab, R. I., Medvec, V. H., & Kern, M. C. (2008). The micro-dynamics of coalition

formation. Political Research Quarterly, 61(3), 484–501. doi:10.1177/1065912908316981
Foltz, P. W., Kintsch, W., & Landauer, T. K. (1998). The measurement of textual coherence with latent

semantic analysis. Discourse Processes, 25(2–3), 285–307. doi:10.1080/01638539809545029
Giles, H., Willemyns, M., Gallois, C., & Anderson, M. C. (2007). Accommodating a new frontier: The

context of law enforcement. In K. Fiedler (Ed.), Social communication (pp. 129–162). New York:
Psychology Press.

Hu, X., Cai, Z., Wiemer-Hastings, P., Graesser, A. C., & McNamara, D. S. (2007). Strengths, limitations, and

extensions of LSA. In T. K. Landauer, D. S. McNamara, S. Dennis, & W. Kintsch (Eds.), Handbook of
latent semantic analysis (pp. 401–426). Mahwah, NJ: Erlbaum.

Huffaker, D., Jorgensen, J., Iacobelli, F., Tepper, P., & Cassell, J. (2006). Computational measures for

language similarity across time in online communities. In E. Hovy, K. Zechner, & L. Zhou (Eds.),

Proceedings of the HLT-NAACL 2006 workshop on analyzing conversations in text and speech (pp.
15–22). New York: Association for Computational Linguistics.

Huffaker, D. A., Swaab, R., & Diermeier, D. (2011). The language of coalition formation in online

multiparty negotiations. Journal of Language and Social Psychology, 30(1), 66–81. doi:10.1177/
0261927X10387102

270 E. Sagi, D. Diermeier / Cognitive Science 41 (2017)

http://dx.doi.org/10.1146/annurev.psych.51.1.279

http://www.natcorp.ox.ac.uk/

http://dx.doi.org/10.1037/0278-7393.22.6.1482

http://dx.doi.org/10.1037/0278-7393.22.6.1482

http://dx.doi.org/10.3758/BF03195282

http://dx.doi.org/10.1177/1065912908316981

http://dx.doi.org/10.1080/01638539809545029

http://dx.doi.org/10.1177/0261927X10387102

http://dx.doi.org/10.1177/0261927X10387102

Landauer, T. K., & Dumais, S. T. (1997). A solution to Plato’s problem: The latent semantic analysis theory

of acquisition, induction, and representation of knowledge. Psychological Review, 104(2), 211–240.
doi:10.1037/0033-295X.104.2.211

McNamara, D. S., Cai, Z., & Louwerse, M. M. (2007). Optimizing LSA measures of cohesion. In T. K.

Landauer, D. S. McNamara, S. Dennis, & W. Kintsch (Eds.), Handbook of latent semantic analysis
(pp. 379–400). Mahwah, NJ: Erlbaum.

Pennebaker, J., Chung, C., Ireland, M., Gonzales, A., & Booth, R. (2007). The development and psychometric
properties of LIWC2007. Austin, TX: LIWC. Available at www.liwc.net. Accessed May 19, 2015.

Pickering, M. J., & Garrod, S. (2004). Toward a mechanistic psychology of dialogue. Behavioral and Brain
Sciences, 27(2), 169–189.

Raiffa, H. (1982). The art and science of negotiation. Cambridge, MA: Harvard University Press.
Sagi, E., & Dehghani, M. (2014). Measuring moral rhetoric in text. Social Science Computer Review, 32(2),

132–144. doi:10.1177/0894439313506837
Sagi, E., Diermeier, D., & Kaufmann, S. (2013). Identifying Issue Frames in Text. PLoS ONE, 8(7), e69185.
Sandhaus, E. (2008). The New York Times annotated corpus. Philadelphia, PA: Linguistic Data Consortium.
Sch€utze, H. (1997). Ambiguity resolution in language learning: Computational and cognitive models.

Stanford, CA: CSLI Publications.

Swaab, R. I., Maddux, W. W., & Sinaceur, M. (2011). Early words that work: When and how virtual

linguistic mimicry facilitates negotiation outcomes. Journal of Experimental Social Psychology, 47(3),
616–621. doi:10.1016/j.jesp.2011.01.005

Takayama, Y., Flournoy, R., Kaufmann, S., & Peters, S. (1998). Information mapping: Concept-based
information retrieval based on word associations. Stanford, CA: CSLI Publications.

Wolfe, M. B. W., Schreiner, M. E., Rehder, B., Laham, D., Foltz, P. W., Kintsch, W., & Landauer, T. K.

(1998). Learning from text: Matching readers and texts by latent semantic analysis. Discourse Processes,
25(2–3), 309–336. doi:10.1080/01638539809545030

E. Sagi, D. Diermeier / Cognitive Science 41 (2017) 271

http://dx.doi.org/10.1037/0033-295X.104.2.211

http://www.liwc.net

http://dx.doi.org/10.1177/0894439313506837

http://dx.doi.org/10.1016/j.jesp.2011.01.005

http://dx.doi.org/10.1080/01638539809545030