Perception of Personality and Naturalness through Dialogues by
Native Speakers of American English and Arabic
Maxim Makatchev
Robotics Institute
Carnegie Mellon University
Pittsburgh, PA, USA
mmakatch@cs.cmu.edu
Reid Simmons
Robotics Institute
Carnegie Mellon University
Pittsburgh, PA, USA
reids@cs.cmu.edu
Abstract
Linguistic markers of personality traits have
been studied extensively,
but few cross-
cultural studies exist. In this paper, we eval-
uate how native speakers of American English
and Arabic perceive personality traits and nat-
uralness of English utterances that vary along
the dimensions of verbosity, hedging, lexical
and syntactic alignment, and formality. The
utterances are the turns within dialogue frag-
ments that are presented as text transcripts to
the workers of Amazon’s Mechanical Turk.
The results of the study suggest that all four di-
mensions can be used as linguistic markers of
all personality traits by both language commu-
nities. A further comparative analysis shows
cross-cultural differences for some combina-
tions of measures of personality traits and nat-
uralness, the dimensions of linguistic variabil-
ity and dialogue acts.
1
Introduction
English has been used as a lingua franca across the
world, but the usage differs. The variabilities in En-
glish introduced by dialects, cultures, and non-native
speakers result in different syntax and words ex-
pressing similar meanings and in different meanings
attributed to similar expressions. These differences
are a source of pragmatic failures (Thomas, 1983):
situations when listeners perceive meanings and af-
fective attitudes unintended by speakers. For exam-
ple, Thomas (1984) reports that usage of Illocution-
ary Force Indicating Devices (IFIDs, such as “I warn
you”, (Searle, 1969)) in English by native speak-
ers of Russian causes the speakers to sometimes
appear “inappropriately domineering in interactions
with English-speaking equals.” Dialogue systems,
just like humans, may misattribute attitudes and mis-
interpret intent of user’s utterances. Conversely, they
may also cause misattributions and misinterpreta-
tions on the user’s part. Hence, taking into account
the user’s dialect, culture, or native language may
help reduce pragmatic failures.
This kind of adaptation requires a mapping from
utterances, or more generally, their linguistic fea-
tures, to meanings and affective attributions for each
of the target language communities. In this paper
we present an exploratory study that evaluates such
a mapping from the linguistic features of verbosity,
hedging, alignment, and formality (as deﬁned in
Section 3.1) to the perceived personality traits and
naturalness across the populations of native speak-
ers of American English and Arabic.
Estimating the relationship between linguistic
features and their perception across language com-
munities faces a number of methodological difﬁcul-
ties. First, language communities shall be outlined,
in a way that will afford generalizing within their
populations.
Deﬁning language communities is a
hard problem, even if it is based on the “mother
tongue” (McPherson et al., 2000). Next, linguistic
features that are potentially important for the adap-
tation must be selected.
These are, for example,
the linguistic devices that contribute to realization of
rich points (Agar, 1994), i.e. the behaviors that sig-
nal differences between language communities. To
be useful for dialogue system research, the selected
linguistic features should be feasible to implement in
natural language generation and interpretation mod-
arXiv:1105.4582v1  [cs.CL]  23 May 2011
ules. Then, a corpus of stimuli that span the variabil-
ity of the linguistic features must be created. The
stimuli should reﬂect the context where the dialogue
system is intended to be used. For example, in case
of an information-giving dialogue system, the stim-
uli should include some question-answer adjacency
pairs (Schegloff and Sacks, 1973). Finally, scales
should be chosen to allow for scoring of the stimuli
with respect to the metrics of interest. These scales
should be robust to be applied within each of the lan-
guage communities.
In the remainder of this paper, we describe each of
these steps in the context of an exploratory study that
evaluates perception of English utterances by native
speakers of American English and Arabic. Our ap-
plication is an information-giving dialogue system
that is used by the robot receptionists (roboception-
ists) in Qatar and the United States (Makatchev et
al., 2009; Makatchev et al., 2010). In the next sec-
tion, we continue with an overview of the related
work. Section 3 introduces the experiment, includ-
ing the selection of stimuli, measures, design, and
describes the recruitment of participants via Ama-
zon’s Mechanical Turk (MTurk). We discuss results
in Section 4 and provide a conclusion in Section 5.
2
Related work
2.1
Cross-cultural variability in English
Language is tightly connected with culture (Agar,
1994). As a result, even native speakers of a lan-
guage use it differently across dialects (e.g. African
American Vernacular English and Standard Amer-
ican English), genders (see, for example, (Lakoff,
1973)) and social statuses (e.g. (Huspek, 1989)),
among other dimensions.
Speakers of English as a second language display
variabilities in language use that are consistent with
their native languages and backgrounds. For exam-
ple, Nelson et al. (1996) reports that Syrian speakers
of Arabic tend to use different compliment response
strategies as compared with Americans.
Aguilar
(1998) reviews types of pragmatic failures that are
inﬂuenced by native language and culture. In partic-
ular, he cites Davies (1987) on a pragmatic failure
due to non-equivalence of formulas: native speakers
of Moroccan Arabic use a spoken formulaic expres-
sion to wish a sick person quick recovery, whereas in
English the formula “get well soon” is not generally
used in speech. Feghali (1997) reviews features of
Arabic communicative style, including indirectness
(concealment of wants, needs or goals (Gudykunst
and Ting-Toomey, 1988)), elaborateness (rich and
expressive language use, e.g. involving rhetorical
patterns of exaggeration and assertion (Patai, 1983))
and affectiveness (i.e. “intuitive-affective style of
emotional appeal” (Glenn et al., 1977), related to
the patterns of organization and presentation of ar-
guments).
In this paper, we are concerned with English us-
age by native speakers of American English and na-
tive speakers of Arabic. We have used the features
of the Arabic communicative style outlined above
as a guide in selecting the dimensions of linguistic
variability that are presented in Section 3.1.
2.2
Measuring pragmatic variation
Perception of pragmatic variation of spoken lan-
guage and text has been shown to vary across
cultures
along
the
dimensions
of
personality
(e.g. (Scherer, 1972)), emotion (e.g. (Burkhardt et
al., 2006)), deception (e.g. (Bond et al., 1990)),
among others. Within a culture, personality traits
such as extraversion, have been shown to have
consistent markers in language (see overview in
(Mairesse et al., 2007)).
For example, Furnham
(1990) notes that in conversation, extraverts are less
formal and use more verbs, adverbs and pronouns.
However, the authors are not aware of any quantita-
tive studies that compare linguistic markers of per-
sonality across cultures. The present study aims to
help ﬁll this gap.
A mapping between linguistic dimensions and
personality has been evaluated by grading es-
says and conversation extracts (Mairesse et al.,
2007), and by grading utterances generated automat-
ically with a random setting of linguistic parame-
ters (Mairesse and Walker, 2008). In the exploratory
study presented in this paper, we ask our participants
to grade dialogue fragments that were manually cre-
ated to vary along each of the four linguistic dimen-
sions (see Section 3.1).
3
Experiment
In the review of related work, we presented some ev-
idence supporting the claim that linguistic markers
of personality may differ across cultures. In this sec-
tion, we describe a study that evaluates perception
of personality traits and naturalness of utterances by
native speakers of American English and Arabic.
3.1
Stimuli
The selection of stimuli attempts to satisfy three ob-
jectives. First, our application: our dialogue system
is intended to be used on a robot receptionist. Hence,
the stimuli are snippets of dialogue that include four
dialogue acts that are typical in this kind of em-
bodied information-giving dialogue (Makatchev et
al., 2009): a greeting, a question-answer pair, a dis-
agreement (with the user’s guess of an answer), and
an apology (for the robot not knowing the answer to
the question).
Second, we would like to vary our stimuli along
the linguistic dimensions that are potentially strong
indicators of personality traits. Extraverts, for exam-
ple, are reported to be more verbose (use more words
per utterances and more dialogue turns to achieve
the same communicative goal), less formal (Furn-
ham, 1990) (in choice of address terms, for exam-
ple), and less likely to hedge (use expressions such
as “perhaps” and “maybe”) (Nass et al., 1995). Lex-
ical and syntactic alignment, namely, the tendency
of a speaker to use the same lexical and syntactic
choices as their interlocutor, is considered, at least
in part, to reﬂect the speaker’s co-operation and will-
ingness to adopt the interlocutor’s perspective (Hay-
wood et al., 2003). There is some evidence that the
degree of alignment is associated with personality
traits of the speakers (Gill et al., 2004).
Third, we would like to select linguistic dimen-
sions that potentially expose cross-cultural differ-
ences in perception of personality and naturalness.
In particular, we are interested in the linguistic de-
vices that help realize rich points (the behaviors that
signal differences) between the native speakers of
American English and Arabic. We choose to real-
ize indirectness and elaborateness, characteristic of
Arabic spoken language (Feghali, 1997), by vary-
ing the dimensions of verbosity and hedging. High
power distance, or inﬂuence of relative social status
on the language (Feghali, 1997), can be realized by
the degrees of formality and alignment.
In summary, the stimuli are dialogue fragments
where utterances of one of the interlocutors vary
across (1) dialogue acts: a greeting, question-answer
pair, disagreement, apology, and (2) four linguistic
dimensions: verbosity, hedging, alignment, and for-
mality. Each of the linguistic dimensions is parame-
terized by 3 values of valence: negative, neutral and
positive. Within each of the four dialogue acts, stim-
uli corresponding to the neutral valences are repre-
sented by the same dialogue across all four linguistic
dimensions. The four linguistic dimensions are real-
ized as follows:
• Verbosity is realized as number of words within
each turn of the dialogue. In the case of the
greeting, positive verbosity is realized by in-
creased number of dialogue turns.1
• Positive valence of hedging implies more ten-
tative words (“maybe,” “perhaps,” etc.) or ex-
pressions of uncertainty (“I think,” “if I am
not mistaken”). Conversely, negative valence
of hedging is realized via words “sure,” “deﬁ-
nitely,” etc.
• Positive valence of alignment corresponds to
preference towards the lexical and syntactic
choices of the interlocutor. Conversely, neg-
ative alignment implies less overlap in lexical
and syntactic choices between the interlocu-
tors.
• Our model of formality deploys the follow-
ing linguistic devices: in-group identity mark-
ers that target positive face (Brown and Levin-
son, 1987) such as address forms, jargon and
slang, and deference markers that target nega-
tive face, such as “kindly”, terms of address,
hedges. These devices are used in Arabic po-
liteness phenomena (Farahat, 2009), and there
is an evidence of their pragmatic transfer from
Arabic to English (e.g. (Bardovi-Harlig et al.,
2007) and (Ghawi, 1993)).
The complete set of stimuli is shown in Tables 2–6.
Each dialogue fragment is presented as a text on
an individual web page. On each page, the partici-
1The multi-stage greeting dialogue was developed via
ethnographic studies conducted at Alelo by Dr.
Suzanne
Wertheim. Used with permission from Alelo, Inc.
pant is asked to imagine that he or she is one of the
interlocutors and the other interlocutor is described
as “a female receptionist in her early 20s and of
the same ethnic background” as that of the partici-
pant. The description of the occupation, age, gender
and ethnicity of the interlocutor whose utterances
the participant is asked to evaluate should provide
minimal context and help avoid variability due to the
implicit assumptions that subjects may make.
3.2
Measures
In order to avoid a possible interference of scales,
we ran two versions of the study in parallel.
In
one version, participants were asked to evaluate the
receptionist’s utterances with respect to measures
of the Big Five personality traits (John and Srivas-
tava, 1999), namely the traits of extraversion, agree-
ableness, conscientiousness, emotional stability, and
openness, using the ten-item personality question-
naire (TIPI, see (Gosling et al., 2003)). In the other
version, participants were asked to evaluate the re-
ceptionist’s utterances with respect to their natu-
ralness on a 7-point Likert scale by answering the
question “Do you agree that the receptionist’s utter-
ances were natural?” The variants of such a natural-
ness scale were used by Burkhardt et al. (2006) and
Mairesse and Walker (2008).
3.3
Experimental design
The experiment used a crossed design with the fol-
lowing factors: dimensions of linguistic variability
(verbosity, hedging, alignment, or formality), va-
lence (negative, neutral, or positive), dialogue acts
(greeting, question-answer, disagreement, or apol-
ogy), native language (American English or Arabic)
and gender (male or female).
In an attempt to balance the workload of the par-
ticipants, depending on whether the participant was
assigned to the study that used personality or nat-
uralness scales, the experimental sessions consisted
of one or two linguistic variability conditions—12
or 24 dialogues respectively. Hence valence and dia-
logue act were within-subject factors, while linguis-
tic variability dimension were treated as an across-
subject factor, as well as native language and gen-
der. Within each session the items were presented in
a random order to minimize possible carryover ef-
fects.
Language
Country
N
Arabic
Algeria
1
Bahrain
1
Egypt
56
Jordan
32
Morocco
45
Palestinian Territory
1
Qatar
1
Saudi Arabia
5
United Arab Emirates
13
Total
155
American English
United States
166
Table 1: Distribution of study participants by country.
3.4
Participants
We used Amazon’s Mechanical Turk (MTurk) to re-
cruit native speakers of American English from the
United States and native speakers of Arabic from
any of the set of predominantly Arabic-speaking
countries (according to the IP address).
Upon completion of each task, participants re-
ceive monetary reward as a credit to their MTurk ac-
count. Special measures were taken to prevent mul-
tiple participation of one person in the same study
condition: the study website access would be re-
fused for such a user based on the IP address, and
MTurk logs were checked for repeated MTurk user
names to detect logging into the same MTurk ac-
count from different IP addresses. Hidden questions
were planted within the study to verify the ﬂuency
in the participant’s reported native language.
The distribution of the participants across coun-
tries is shown in Table 1. We observed a regional
gender bias similar to the one reported by Ross et al.
(2010): there were 100 male and 55 female partici-
pants in the Arabic condition, and 63 male and 103
female participants in the American English condi-
tion.
4
Results
We analyzed the data by ﬁtting linear mixed-effects
(LME) models (Pinheiro and Bates, 2000) and per-
forming model selection using ANOVA. The com-
parison of models ﬁtted to explain the personality
and naturalness scores (controlling for language and
gender), shows signiﬁcant main effects of valence
and dialogue acts for all pairs of personality traits
(and naturalness) and linguistic features. The results
also show that for every personality trait (and nat-
uralness) there is a linguistic feature that results in
a signiﬁcant three-way interaction between its va-
lence, the native language, and the dialogue act.
These results suggest that (a) for both language com-
munities, every linguistic dimension is associated
with every personality trait and naturalness, for at
least some of the dialogue acts, (b) there are differ-
ences in the perception of every personality trait and
naturalness between the two language communities.
To further explore the latter ﬁnding, we conducted
a post-hoc analysis consisting of paired t-tests that
were performed pairwise between the three values of
valence for each combination of language, linguis-
tic feature, and personality trait (and naturalness).
Note, that comparing raw scores between the lan-
guage conditions would be prone to ﬁnd spurious
differences due to potential culture-speciﬁc tenden-
cies in scoring on the Likert scale: (a) perception
of magnitudes and (b) appropriateness of the inten-
sity of agreeing or disagreeing. Instead, we compare
the language conditions with respect to (a) the rela-
tive order of the three valences and (b) the binarized
scores, namely whether the score is above 4 or be-
low 4 (with scores that are not signiﬁcantly different
from 4 excluded from comparison), where 4 is the
neutral point of the 7-point Likert scale.
The selected results of the post-hoc analysis are
shown in Figure 1.
The most prominent cross-
cultural differences were found in the scoring of
naturalness across the valences of the formality di-
mension. Speakers of American English, unlike the
speakers of Arabic, ﬁnd formal utterances unnatu-
ral in greetings, question-answer and disagreement
dialogue acts.
Formal utterances tend to also be
perceived as indicators of openness and conscien-
tiousness by Arabic speakers, and not by American
English speakers, in disagreements and apologies
respectively. Finally, hedging in apologies is per-
ceived as an indicator of agreeableness by American
English speakers, but not by speakers of Arabic.
Interestingly, no qualitative differences across
language conditions were found in the perception
of extraversion and stability. It is possible that this
cross-cultural consistency conﬁrms the view of the
extraversion, in particular, as one of most consis-
tently identiﬁed dimensions (see, for example, (Gill
and Oberlander, 2002)).
It could also be possi-
ble that our stimuli were unable to pinpoint the
extraversion-related rich points due to a choice of
the linguistic dimensions or particular wording cho-
sen. A larger variety of stimuli per condition, and an
ethnography to identify potentially culture-speciﬁc
linguistic devices of extraversion, could shed the
light on this issue.
5
Conclusion
We presented an exploratory study to evaluate a set
of linguistic markers of Big Five personality traits
and naturalness across two language communities:
native speakers of American English living in the
US, and native speakers of Arabic living in one
of the predominantly Arabic-speaking countries of
North Africa and Middle East. The results suggest
that the four dimensions of linguistic variability are
recognized as markers of all ﬁve personality traits by
both language communities. A comparison across
language communities uncovered some qualitative
differences in the perception of openness, conscien-
tiousness, agreeableness, and naturalness.
The results of the study can be used to adapt nat-
ural language generation and interpretation to native
speakers of American English or Arabic. This ex-
ploratory study also supports the feasibility of the
crowdsourcing approach to validate the linguistic
devices that realize rich points—behaviors that sig-
nal differences across languages and cultures.
Future work shall evaluate effects of regional di-
alects and address the issue of particular wording
choices by using multiple stimuli per condition.
Acknowledgments
This publication was made possible by the support
of an NPRP grant from the Qatar National Research
Fund. The statements made herein are solely the re-
sponsibility of the authors.
The authors are grateful to Ameer Ayman Abdul-
salam, Michael Agar, Hatem Alismail, Justine Cas-
sell, Majd Sakr, Nik Melchior, and Candace Sidner
for their comments on the study.
References
[Agar1994] Michael Agar. 1994. Language shock: Un-
derstanding the culture of conversation. William Mor-
row, New York.
[Aguilar1998] Maria Jose Coperias Aguilar. 1998. Inter-
cultural (mis)communication: The inﬂuence of L1 and
C1 on L2 and C2. A tentative approach to textbooks.
Cuadernos de Filolog´ıa Inglesa, 7(1):99–113.
[Bardovi-Harlig et al.2007] Kathleen
Bardovi-Harlig,
Marda Rose, and Edelmira L. Nickels. 2007. The use
of conventional expressions of thanking, apologizing,
and refusing.
In Proceedings of the 2007 Second
Language Research Forum, pages 113–130.
[Bond et al.1990] Charles F. Bond, Adnan Omar, Adnan
Mahmoud, and Richard Neal Bonser. 1990. Lie de-
tection across cultures. Journal of Nonverbal Behav-
ior, 14:189–204.
[Brown and Levinson1987] P. Brown and S. C. Levinson.
1987. Politeness: Some universals in language usage.
Cambridge University Press, Cambridge.
[Burkhardt et al.2006] F.
Burkhardt,
N.
Audibert,
L. Malatesta, O. Trk, Arslan, L., and V Auberge.
2006.
Emotional prosody—does culture make a
difference? In Proc. Speech Prosody.
[Davies1987] Eirlys E. Davies. 1987. A contrastive ap-
proach to the analysis of politeness formulas. Applied
Linguistics, 8(1):75–88.
[Farahat2009] Said Hassan Farahat.
2009.
Politeness
phenomena in Palestinian Arabic and Australian En-
glish: A cross-cultural study of selected contemporary
plays (PhD thesis).
Australian Catholic University,
Australia.
[Feghali1997] Ellen Feghali. 1997. Arab cultural com-
munication patterns.
International Journal of Inter-
cultural Relations, 21(3):345–378.
[Furnham1990] A. Furnham. 1990. Language and per-
sonality. In H. Giles and W. Robinson, editors, Hand-
book of Language and Social Psychology, pages 73–
95. Wiley.
[Ghawi1993] Mohammed Ghawi. 1993. Pragmatic trans-
fer in Arabic learners of English.
El Two Talk,
1(1):39–52.
[Gill and Oberlander2002] A. Gill and J. Oberlander.
2002. aking care of the linguistic features of extraver-
sion. In Proceedings of the 24th Annual Conference of
the Cognitive Science Society, pages 363–368.
[Gill et al.2004] A. Gill, A. Harrison, and J. Oberlander.
2004.
Interpersonality: Individual differences and
interpersonal priming.
In Proceedings of the 26th
Annual Conference of the Cognitive Science Society,
pages 464–469.
[Glenn et al.1977] E. S. Glenn, D. Witmeyer, and K. A.
Stevenson. 1977. Cultural styles of persuasion. Inter-
national Journal of Intercultural Relations, 1(3):52–
66.
[Gosling et al.2003] Samuel D. Gosling, Peter J. Rent-
frow, and Jr. William B. Swann. 2003. A very brief
measure of the Big-Five personality domains. Journal
of Research in Personality, 37:504–528.
[Gudykunst and Ting-Toomey1988] W.
B.
Gudykunst
and S. Ting-Toomey. 1988. Culture and interpersonal
communication. Sage, Newbury Park, CA.
[Haywood et al.2003] S. Haywood, M. Pickering, and
H. Branigan. 2003. Co-operation and co-ordination
in the production of noun phrases. In Proceedings of
the 25th Annual Conference of the Cognitive Science
Society, pages 533–538.
[Huspek1989] Michael Huspek. 1989. Linguistic vari-
ability and power: An analysis of you know/I think
variation in working-class speech. Journal of Prag-
matics, 13(5):661 – 683.
[John and Srivastava1999] Oliver P. John and Sanjay Sri-
vastava.
1999.
The Big Five trait taxonomy: His-
tory, measurement, and theoretical perspectives.
In
Lawrence A. Pervin and Oliver P. John, editors, Hand-
book of Personality: Theory and Research, pages 102–
138.
[Lakoff1973] Robin Lakoff.
1973.
Language and
woman’s place. Language in Society, 2(1):45–80.
[Mairesse and Walker2008] Francois Mairesse and Mari-
lyn Walker.
2008.
Trainable generation of big-ﬁve
personality styles through data-driven parameter esti-
mation. In Proc. of 46th Annual Meeting of the Asso-
ciation for Computational Linguistics (ACL).
[Mairesse et al.2007] F. Mairesse, M. A. Walker, M. R.
Mehl, and R. K. Moore. 2007. Using linguistic cues
for the automatic recognition of personality in conver-
sation and text. Journal of Artiﬁcial Intelligence Re-
search, 30:457–500.
[Makatchev et al.2009] Maxim Makatchev, Min Kyung
Lee, and Reid Simmons. 2009. Relating initial turns
of human-robot dialogues to discourse. In Proc. of the
Int. Conf. on Human-Robot Interaction (HRI), pages
321–322. ACM.
[Makatchev et al.2010] Maxim Makatchev, Imran Aslam
Fanaswala, Ameer Ayman Abdulsalam, Brett Brown-
ing, Wael Mahmoud Gazzawi, Majd Sakr, and Reid
Simmons. 2010. Dialogue patterns of an arabic robot
receptionist.
In Proc. of the Int. Conf. on Human-
Robot Interaction (HRI), pages 167–168. ACM.
[McPherson et al.2000] M. McPherson, L. Smith-Lovin,
and J. M. Cook.
2000.
What is a language com-
munity?
American Journal of Political Science,
44(1):142–155.
[Nass et al.1995] Clifford Nass, Y. Moon, B. Fogg, and
B. Reeves.
1995.
Can computer personalities be
human personalities?
Journal of Human-Computer
Studies, 43:223–239.
[Nelson et al.1996] Gaylel Nelson, Mahmoud Al-Batal,
and Erin Echols. 1996. Arabic and english compli-
ment responses: Potential for pragmatic failure. Ap-
plied Linguistics, 17(4):411–432.
[Patai1983] R. Patai.
1983.
The Arab mind.
Charles
Scribner’s Sons, New York.
[Pinheiro and Bates2000] J. C. Pinheiro and D. M. Bates.
2000.
Mixed-Effects Models in S and S-PLUS.
Springer.
[Ross et al.2010] Joel Ross, Lilly Irani, M. Six Silberman,
Andrew Zaldivar, and Bill Tomlinson. 2010. Who are
the crowdworkers?: shifting demographics in mechan-
ical turk. In Proceedings of the 28th of the interna-
tional conference extended abstracts on Human fac-
tors in computing systems, CHI EA ’10, pages 2863–
2872, New York, NY, USA. ACM.
[Schegloff and Sacks1973] Emanuel A. Schegloff and
Harvey Sacks. 1973. Opening up closings. Semiot-
ica, 8(4):289–327.
[Scherer1972] Klaus R. Scherer. 1972. Judging person-
ality from voice: A cross-cultural approach to an old
issue in interpersonal perception. Journal of Personal-
ity, 40:191–210.
[Searle1969] John Searle. 1969. Speech acts: An essay
in the philosophy of language. Cambridge University
Press.
[Thomas1983] Jenny Thomas.
1983.
Cross-cultural
pragmatic failure. Applied Linguistics, 4(2):91–112.
[Thomas1984] Jenny Thomas. 1984. Cross-cultural dis-
course as ’unequal encounter’: Towards a pragmatic
analysis. Applied Linguistics, 5(3):226–235.
Dimension and
valence
Greeting
Question-Answer
Disagreement
Apology
Control
(neutral)
A: Good morning.
B: Good morning. How may
I help you?
A: Could you tell me where
the library is?
B: It’s at the end of the hall-
way on your left.
A: Could you tell me where
the library is?
B: It’s on the second ﬂoor.
A: I thought it was on the
ﬁrst ﬂoor.
B: No, there is no library on
the ﬁrst ﬂoor.
A: Could you tell me where
the library is?
B: Sorry, I don’t know.
Table 2: Stimuli corresponding to the neutral control condition, shared by all dimensions of linguistic variability.
Dimension and
valence
Greeting
Question-Answer
Disagreement
Apology
Verbosity
negative
A: Good morning.
B: Morning. May I help you?
A: Could you tell me where
the library is?
B: End of the hallway on
your left.
A: Could you tell me where
the library is?
B: Second ﬂoor.
A: I thought it was on the
ﬁrst ﬂoor.
B: No, it is not.
A: Could you tell me where
the library is?
B: I don’t know.
Verbosity
positive
A: Good morning.
B: Good morning. How are
you today?
A: I am doing well, thanks.
You?
B: Very well,
thank you.
How’s your family?
A: Everyone is doing ﬁne,
thanks. How about yours?
B: Mine is doing well too.
How may I help you?
A: Could you tell me where
the library is?
B: Yes, just follow this hall-
way until it ends and you
will ﬁnd the library on your
left hand side.
A: Could you tell me where
the library is?
B: Yes, the library is on the
second ﬂoor.
A: I thought it was on the
ﬁrst ﬂoor.
B: No, there is no library on
the ﬁrst ﬂoor of this build-
ing.
A: Could you tell me where
the library is?
B: I am so sorry, I don’t re-
ally know where the library
is in this building.
Table 3: Stimuli along the verbosity dimension.
Dimension and
valence
Greeting
Question-Answer
Disagreement
Apology
Hedging
negative
A: How are you?
B: Deﬁnitely good,
and
you?
A: Good, thank you.
B: How may I help you?
A: Could you tell me where
the library is?
B: Sure, it is at the end of
the hallway on your left.
A: Could you tell me where
the library is?
B: The library is on the sec-
ond ﬂoor.
A: I thought it was on the ﬁrst
ﬂoor.
B: No, there is deﬁnitely no li-
brary on the ﬁrst ﬂoor.
A: Could you tell me where
the library is?
B: Sorry, I have no idea
where it is.
Hedging
positive
A: How are you?
B: Good, I guess, and you?
A: Good, thank you.
B: Could I help you?
A: Could you tell me where
the library is?
B: I think it is at the end of
the hallway on your left.
A: Could you tell me where
the library is?
B: I think the library is on the
second ﬂoor.
A: I thought it was on the ﬁrst
ﬂoor.
B: No, if I am not mistaken,
there is no library on the ﬁrst
ﬂoor.
A: Could you tell me where
the library is?
B: Sorry, I am not sure.
Table 4: Stimuli along the hedging dimension.
Dimension and
valence
Greeting
Question-Answer
Disagreement
Apology
Alignment
negative
A: How are you?
B: Good morning.
How
may I help you?
A: Could you tell me where
the bathroom is?
B: The restroom is at the
end of the hallway on your
left.
A: Could you tell me where
the bathroom is?
B: The restroom is on the sec-
ond ﬂoor.
A: I thought the bathroom was
on the ﬁrst ﬂoor.
B: No, there is no restroom on
the ﬁrst ﬂoor.
A: Could you tell me where
the bathroom is?
B: Sorry, I don’t know about
the restroom.
Alignment
positive
A: How are you?
B: Good,
how are you?
How may I help you?
A: Could you tell me where
the bathroom is?
B: The bathroom is at the
end of the hallway on your
left.
A: Could you tell me where
the bathroom is?
B: The bathroom is on the
second ﬂoor.
A: I thought the bathroom was
on the ﬁrst ﬂoor.
B: No, there is no bathroom
on the ﬁrst ﬂoor.
A: Could you tell me where
the bathroom is?
B: Sorry,
I don’t know
where the bathroom is.
Table 5: Stimuli along the lexical and syntactic alignment dimension.
Dimension and
valence
Greeting
Question-Answer
Disagreement
Apology
Formality
negative
A: Good morning.
B: What’s up?
Need any-
thing?
A: Could you tell me where
the library is?
B: Just go to the end of the
hallway, you can’t miss it.
A: Could you tell me where
the library is?
B: Go to the second ﬂoor.
A: I thought it was on the ﬁrst
ﬂoor.
B: No, honey, there is none on
the ﬁrst ﬂoor.
A: Could you tell me where
the library is?
B: Sorry about that, I have
no idea.
Formality
positive
A: Good morning.
B:
Good
morning,
sir
(madam).
Would
you
allow me to help you with
anything?
A: Could you tell me where
the library is?
B: Kindly follow this hall-
way and you will encounter
the entrance on your left.
A: Could you tell me where
the library is?
B: Yes, you may ﬁnd the li-
brary on the second ﬂoor.
A: I thought it was on the ﬁrst
ﬂoor.
B: I am afraid that is not cor-
rect, there is no library on the
ﬁrst ﬂoor.
A: Could you tell me where
the library is?
B: I have to apologize, but I
don’t know.
Table 6: Stimuli along the formality dimension.
greeting
qa
disagree
apology
American English, formality, naturalness
1
2
3
4
5
6
7
** **
**
**
** **
** **
greeting
qa
disagree
apology
Arabic, formality, naturalness
1
2
3
4
5
6
7
** *
**
****
greeting
qa
disagree
apology
American English, formality, openness
1
2
3
4
5
6
7
**
****
greeting
qa
disagree
apology
Arabic, formality, openness
1
2
3
4
5
6
7
**
**
*
greeting
qa
disagree
apology
American English, formality, conscienciousness
1
2
3
4
5
6
7
** *
**
*
*
greeting
qa
disagree
apology
Arabic, formality, conscienciousness
1
2
3
4
5
6
7
* *
**
*
greeting
qa
disagree
apology
American English, hedging, agreeableness
1
2
3
4
5
6
7
* **
**
****
**
**
**
greeting
qa
disagree
apology
Arabic, hedging, agreeableness
1
2
3
4
5
6
7
**
**
Figure 1: A subset of data comparing scores on the Big Five personality traits and naturalness as given by native
speakers of American English (left half of the page) and Arabic (right half of the page). Blue, white, and pink bars
correspond to negative, neutral, and positive valences of the linguistic features respectively. Dialogue acts listed along
the horizontal axis are a greeting, question-answer pair, disagreement, and apology. Error bars the 95% conﬁdence
intervals, brackets above the plots correspond to p-values of paired t-tests at signiﬁcance levels of 0.05 (∗) and 0.01
(∗∗) after Bonferroni correction.