ArticlePDF AvailableLiterature Review

Language Style Matching in Psychotherapy: An Implicit Aspect of Alliance

Authors:

Abstract and Figures

In an attempt to operationalize an implicit aspect of the therapeutic alliance, this article proposes the use of the innovative, objective, and time-efficient analysis of language style matching (LSM; Niederhoffer & Pennebaker, 2002). LSM, defined as the degree of similarity in rates of function words in dyadic interactions, is thought to reflect the extent to which conversational partners are automatically coordinating language styles to achieve a common goal. Although LSM has often been researched in the context of everyday conversations, little is known about the matching of clients and therapists' language style in the psychotherapy process. To demonstrate the clinical usefulness of the LSM approach in psychotherapy, 2 exploratory examples of the application of LSM in long-term psychoanalytic treatments are provided. First, LSM analyses per session and per speaking-turn are described for psychotherapy data of 140 sessions of 7 long-term psychoanalytic treatments in relation to outcome measures. Then, a case study is described in which LSM is triangulated with an observer-rated measure of working alliance in relation to outcome measures. These 2 demonstrative empirical examples were explorative in character and illustrate how LSM might tap into an implicit aspect of the therapeutic relationship, different from the working alliance measured by observers, and relevant for treatment outcome. Future larger-scale psychotherapy studies into the relationship between these implicit aspects of the alliance and treatment outcome and relevant clients and therapists' variables are warranted. (PsycInfo Database Record (c) 2020 APA, all rights reserved).
Content may be subject to copyright.
Language Style Matching in Psychotherapy: An Implicit Aspect
of Alliance
Katie Aafjes-van Doorn
Yeshiva University
John Porcerelli
University of Detroit Mercy
Lena Christine Müller-Frommeyer
Technical University of Braunschweig
In an attempt to operationalize an implicit aspect of the therapeutic alliance, this article proposes the use
of the innovative, objective, and time-efficient analysis of language style matching (LSM; Niederhoffer
& Pennebaker, 2002). LSM, defined as the degree of similarity in rates of function words in dyadic
interactions, is thought to reflect the extent to which conversational partners are automatically coordi-
nating language styles to achieve a common goal. Although LSM has often been researched in the context
of everyday conversations, little is known about the matching of clients and therapists’ language style in
the psychotherapy process. To demonstrate the clinical usefulness of the LSM approach in psychother-
apy, 2 exploratory examples of the application of LSM in long-term psychoanalytic treatments are
provided. First, LSM analyses per session and per speaking-turn are described for psychotherapy data of
140 sessions of 7 long-term psychoanalytic treatments in relation to outcome measures. Then, a case
study is described in which LSM is triangulated with an observer-rated measure of working alliance in
relation to outcome measures. These 2 demonstrative empirical examples were explorative in character
and illustrate how LSM might tap into an implicit aspect of the therapeutic relationship, different from
the working alliance measured by observers, and relevant for treatment outcome. Future larger-scale
psychotherapy studies into the relationship between these implicit aspects of the alliance and treatment
outcome and relevant clients and therapists’ variables are warranted.
Public Significance Statement
Language style matching may offer unique opportunities for the examination of therapist– client
interactions within and between sessions, particularly those implicit aspects which elude conscious
awareness of the therapist, client, or outside observer, that nonetheless exert an influence on
treatment outcome.
Keywords: alliance, function words, language style matching, psychoanalytic psychotherapy
Supplemental materials: http://dx.doi.org/10.1037/cou0000433.supp
The therapeutic alliance reflects interactive elements of the
counseling process to which both the client and the therapist
contribute, (e.g., Horvath & Bedi, 2002). It can be thought of as a
measure of fit or match (Kantrowitz et al., 1989), and varies across
time depending on the contribution to it by both parties, something
that is an ongoing result of the interaction between the client
and therapist (Safran & Muran, 2000). Although the exact
active ingredients of the alliance and how they predict treatment
outcome requires further elucidation, it has been established that
higher levels of client–therapist matching are associated with
better relationship outcomes (e.g., Håvås, Svartberg, & Ulvenes,
2015). In counseling, the therapist explicitly tries to build alliance
XKatie Aafjes-van Doorn, Ferkauf Graduate School of Psychology, Yeshiva
University; John Porcerelli, Department of Psychology, University of Detroit
Mercy; XLena Christine Müller-Frommeyer, Institute of Psychology, Depart-
ment for Industrial/Organizational and Social Psychology, Technical University of
Braunschweig.
We are grateful to Amanda Balakirsky, Zhaoyi Chen, Hannah Ei-
dman, Samantha Mahr, Li-Wei Yuan, and Hemrie Zalman for their
support with preparing the transcripts for the LSM and rLSM analyses.
We thank the Psychoanalytic Research Consortium for the use of a
selection of transcribed treatment sessions and thank Leanne Quigley
for her statistical guidance in the revision of this article. Reciprocal
LSM calculations are performed using the Rscript provided by Müller-
Frommeyer, Frommeyer, and Kauffeld (2019a), which can be retrieved
from https://osf.io/arxgu/.
Correspondence concerning this article should be addressed to Katie
Aafjes-van Doorn, Ferkauf Graduate School of Psychology, Yeshiva Uni-
versity, Rousso Building, Room 123, 1165 Morris Park Avenue, Bronx,
NY 10461. E-mail: katie.aafjes@yu.edu
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Journal of Counseling Psychology
© 2020 American Psychological Association 2020, Vol. 67, No. 4, 509–522
ISSN: 0022-0167 http://dx.doi.org/10.1037/cou0000433
509
by matching the client (Baldwin, Wampold, & Imel, 2007), for
example, by mirroring the client’s affective state with facial ex-
pressions of distress to convey sympathy (Blairy, Herrera, & Hess,
1999). In successful treatments, therapists tend to accommodate to
the language of the client more than in unsuccessful treatments
(Hölzer, Mergenthaler, Pokorny, Kächele, & Luborsky, 1996). A
therapist’s matching is meant to communicate that the client has
been seen, is understood, and can be contained (Fonagy, Gergely,
Jurist, & Target, 2002).
Despite the fact that therapists might explicitly try to adapt to an
individual client, there might be several reasons why clients might
try to match their therapist. For example, research on verbal
synchrony in dyads with a power differential indicates that indi-
viduals with a lower social status (i.e., clients) tend to modify their
word choice to match a conversational partner of a higher status
(i.e., therapist; Danescu-Niculescu-Mizil, Lee, Pang, & Kleinberg,
2012). Moreover, when clients present with high levels of symp-
toms/severe psychopathology, they may lack a sense of self-
esteem and therefore easily adapt to the therapist. However, it
could also be argued that these clients’ therapists might feel more
need to scaffold their interventions, and make more effort to merge
with these clients, closely tracking their subjective experiences.
Arguably, the question of who is adapting to whom is irrelevant
given that psychotherapy has been likened to dancing the tango
(Bucci & Maskit, 2007), where continuous interactions make it
purposefully unclear who is leading whom. This mutual adaptation
process in psychotherapy implies no leadership or symmetry per
se, only that influence is bidirectional.
Limitations of Current Measures of Alliance
There is disagreement regarding which modes of assessment
most accurately assess the therapeutic alliance. Previous alli-
ance research depends heavily on self-report measures. Self-
report measures are likely to be affected by the client or
therapist’s ability, awareness, and motivation to report on the
therapeutic process. Also, the results might differ, depending on
who completes the self-report measure (e.g., Hatcher, Barends,
Hansell, & Gutfreund, 1995). To complement the client’s and
therapist’s view of the alliance, several observer-rated process
coding systems have been developed (e.g., Comparative Psycho-
therapy Process Scale; Hilsenroth, Blagys, Ackerman, Bonge, &
Blais, 2005).
Despite the popularity of these measures in counseling, most
observer-rated measures are not easily applied to long-term psy-
chotherapies. Many rely on the availability of video recordings of
sessions. Also, these observer-based coding systems are generally
very time-intensive, which is especially problematic when inves-
tigating therapeutic processes as they unfold over years of long-
term psychoanalytic treatment. Furthermore, observer-based mea-
sures tend to assess explicit behaviors, or narratives as potential
mechanisms of change (McWilliams, 2011), but do not examine
the implicit nature of the relationship patterns, that are important in
psychoanalytic psychotherapy (e.g., Andrade, 2005).
Measuring Implicit Aspects of the Alliance
Implicit automatic aspects of the alliance may affect therapeutic
change on a scale similar to, or even greater than, the conscious or
observable elements of the alliance itself (McWilliams, 2011).
Arguably, what is most relevant in the psychotherapy process is
not verbal content per se, but the manner in which content is
communicated (e.g., Hölzer et al., 1996;Reyes et al., 2008).
Computerized text analyses of the verbal clinical exchange might
offer a method of capturing this implicit aspect of the therapeutic
relationship (e.g., Babcock, Ta, & Ickes, 2014).
Language Style Matching
To operationalize the interactive and implicit aspects of the
alliance in psychoanalytic psychotherapy, the Language Style
Matching (LSM) metric is proposed, which is based on comput-
erized text analyses performed using the software Linguistic In-
quiry and Word Count (Gonzales, Hancock, & Pennebaker, 2010;
Ireland & Pennebaker, 2010). Rather than content-based aspects of
language (e.g., using the client’s description of feeling “livid”
rather than “angry”), LSM represents the degree to which two
people are producing similar rates of function words (e.g., pro-
nouns, prepositions, and conjunctions) in their dialogue (Gonzales
et al., 2010). Moreover, Pennebaker (2011) argues that higher
LSM does not necessarily suggest agreement in terms of the
content of the material being discussed; rather, it suggests that two
people in a dyad automatically mirror or repeat back specific
words and phrases (Babcock et al., 2014). Emphasizing function
words rather than content words allows research to assess dyadic
linguistic coordination irrespective of context (Ireland et al., 2011).
Linguistically, function words help keep track of the mutual
knowledge shared between speakers so that each person under-
stands what the other is referencing (Cannava & Bodie, 2017).
Function words are short (typically 1– 4 letters) and occur at
extremely high frequencies in most day-to-day conversations, ac-
counting for 60% of words that are spoken (Gonzales et al., 2010;
Ireland & Pennebaker, 2010). Function words are processed in
people’s brains faster than content words such as nouns and verbs
(Diaz & McCarthy, 2009), are predictable and help to achieve
communication goals, but are largely undetectable by both speak-
ers and trained observers (Niederhoffer & Pennebaker, 2002). The
similarity in function word usage between people is thought to be
a primarily unconscious coordination process (Ireland & Penne-
baker, 2010). Notably, this is not a reference to the meaning of
unconscious as it is used in the psychoanalytic literature (i.e.,
repressed material often related to primitive drives/instincts in the
client’s mental life) but instead refers to the meaning of uncon-
scious as it is used in the context of language use (i.e., automatic,
less-controlled communication behaviors).
Conceptually, as described by Müller-Frommeyer, Frommeyer,
and Kauffeld (2019a), LSM has many theoretical foundations,
ranging from interpersonal coordination theories (e.g., Lumsden,
Miles, Richardson, Smith, & Macrae, 2012), to communication
accommodation theory (Shephard, Giles, & LePoire, 2001) and
interactive alignment (Muir, Joinson, Cotterill, & Dewdney, 2016).
These theories share conceptual basics, especially the belief that
automatic linguistic coordination represents interpersonal synergy
(Riley, Richardson, Shockley, & Ramenzoni, 2011) and may thus
be a proxy of an implicit aspect of the therapeutic alliance. More
specifically, the level of LSM is thought to map directly onto the
interpersonal coordination of psychological states (Ireland & Pen-
nebaker, 2010) and may be considered an implicit aspect of Rog-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
510 AAFJES-VAN DOORN, PORCERELLI, AND MU
¨LLER-FROMMEYER
ers’ (1957) therapeutic relationship quality of congruence, impor-
tant for engaging in genuine contact.
Overview of Current Implementations of LSM
LSM in Psychology
LSM was introduced in studies in the era of social psychology
(Niederhoffer & Pennebaker, 2002) and was applied to speed
dating and relationship maintenance (Ireland et al., 2011), predict-
ing the degree of mutual romantic interest in speed dating partners,
successful conflict resolution between romantic partners (e.g.,
Bowen, Winczewski, & Collins, 2017), relationship stability in
dating couples, as well as group cohesiveness in face-to-face
communication environments (e.g., Ireland et al., 2011). There is
consistent evidence that LSM increases as two people interact over
time. Even in relatively brief interactions, LSM is predictive of
relationship outcomes (Ireland et al., 2011).
Although most empirical studies on LSM support the idea that
linguistic coordination yields positive outcomes (e.g., relationship
initiation and stability; Ireland et al., 2011), other studies provide
evidence that LSM is context-sensitive (Bowen et al., 2017).
Within romantic partners, LSM does not uniformly signal inter-
personal rapport (i.e., alliance bond) but instead aids communica-
tion by amplifying the positive or negative tone of an interaction
(Bowen et al., 2017). Similarly, Babcock et al. (2014) found that
LSM was highest in conversations when members talked about
other people (using third-person pronouns) and were more disin-
clined to talk to each other. Moreover, studies have shown that
linguistic coordination is context-sensitive, adaptive, and not al-
ways linked to positive outcomes (e.g., Fusaroli et al., 2012).
Building on these empirical findings of linguistic behavior, Fusa-
roli, Ra˛czaszek-Leonardi, and Tylén (2014) outlined a dynamic
framework for studying dialog based on the notion of interpersonal
synergy, and suggested that speakers develop linguistic patterns of
stable interactions to fit the direct goals of the situation (i.e., good
rapport, solution of a problem) through context-sensitive align-
ment and complementary dynamics. In other words, people take
turns in speaking, contribute to each others’ perspectives, and
develop routines to effectively structure the interaction. Consider-
ing this more nuanced context-sensitive perspective, the successful
coordination of function words likely reflects a common under-
standing of the conversational topic and a shared social knowledge
(Meyer & Bock, 1999). Therefore, in this article, LSM is concep-
tualized as interpersonal synergy; an implicit linguistic coordina-
tion aimed at achieving a common goal.
LSM in Psychotherapy
Although LSM has often been researched in the context of
everyday conversations (established benchmarks for low and high
LSM are .60 and. 85, respectively; Cannava & Bodie, 2017), little
is known about the levels of LSM in psychotherapy. To date, two
initial studies applied LSM to the psychotherapy context. First,
Lord, Sheng, Imel, Baer, and Atkins (2015) examined LSM in
individual psychotherapy training sessions. They analyzed 122
transcripts of 20-min motivational interviewing training sessions
with standardized clients who portrayed a recently referred client
with a substance use problem. Lord and colleagues found that
training sessions in which the therapists were rated as high in
empathy showed higher levels of overall LSM (M.52) than
those that were rated as low in empathy (M.42). Second, Borelli
et al. (2019) examined the trajectory of LSM across a 12-session
manualized psychodynamic therapy for seven substance-
dependent mothers. They measured overall LSM in four of the 12
sessions and reported an average LSM of .89 (SD .02), which
remained consistent throughout the early and middle sessions, and
decreased from Session 9 to Session 11 (M.88, SD .02).
Further, lower early LSM in these therapist– client dyads predicted
greater posttreatment psychiatric distress. After controlling for
clients’ pretreatment psychiatric distress and therapist, LSM me-
diated the association between clients’ pretreatment relational
problems and posttreatment psychiatric distress. These two pre-
liminary applications of LSM suggest that more empathetic ther-
apists might match their language style to their clients to a larger
extent, and that LSM might reflect a relationship quality in psy-
chotherapy, which is important for posttreatment symptom reduc-
tion.
Practical Guidelines in the Application of LSM
in Psychotherapy
To complement these two preliminary studies that applied LSM
to single training sessions (Lord et al., 2015) and brief 12-session
manualized treatments (Borelli et al., 2019), this article provides
two empirical examples of the application of LSM in long-term
psychoanalytic treatments. First, LSM is applied to a psychoana-
lytic treatment sample of 140 sessions of seven long-term treat-
ments. Then, its clinical usefulness is reported in a case study in
relation to alliance and outcome measurements.
LSM at Session Level
Following the editing process, client and therapist texts for each
session were examined with the LIWC software (Pennebaker,
Booth, & Francis, 2007). The LIWC2015 software and the
LIWC2015 Operator’s Manual are published by Pennebaker Con-
glomerates, Inc., Austin, Texas. The software comes with a single
application file for either Windows or Macintosh, and an unlimited
academic license for this program costs $89.95 (see website http://
liwc.wpengine.com/). The software calculates a percentage of total
words in a text that fall into 70 different, nonexclusive word
categories, including nine function-word categories that are used in
the calculation of the LSM metric (Niederhoffer & Pennebaker,
2002): auxiliary verbs (e.g., might, would), articles (e.g., the, a,
an), common adverbs (e.g., always, naturally), personal pronouns
(e.g., I, his, their, you), indefinite impersonal pronouns (e.g.,
another, someone), prepositions and relative pronouns (e.g., of,
which, in), negations (e.g., no, not, never), conjunctions (e.g., and,
but, because), and quantifiers (e.g., much, few). After separating
the transcripts by speaker, the percentage of function words (i.e.,
relative score, controlling for differences in the number of words
used by each speaker) is calculated for each of the nine function
word categories and each therapist and client.
To reach the overall LSM score for each session and each
client-therapist dyad, first the absolute value of the difference
between proportions for a client and a therapist for each function
word category is calculated (|(Function Word Client – Function
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
511
LANGUAGE STYLE MATCHING IN PSYCHOTHERAPY
Word Therapist)|). This value is then divided by the combined
function word category proportion for the dyad (Function Word
Client Function Word Therapist .0001). In the denominator
the .0001 is added to prevent empty data sets. This score is
standardized by taking the absolute value and subtracting it from 1,
yielding a range of 0 to 1, per the following equation:
LSMFW 1
|
FWClient FWTherapist
|
FWClient FWTherapist 0.0001 (1)
This calculation is repeated for each of the nine function word
categories for each dyad at each of the therapy sessions. The nine
category-level LSM scores are then averaged to yield a composite
LSM score bounded by 0 and 1, where higher numbers represent
greater LSM between client and therapist.
LSM at the Level of Speaking-Turns
The overall LSM metric describe above is fundamentally a
dyadic index at the overall conversational level (Cannava, 2018).
However, in psychotherapy, there is a temporal sequence of many
conversational turns within a single session. To assess the temporal
reciprocity in LSM, a reciprocal LSM metric has now been devel-
oped (rLSM; Müller-Frommeyer et al., 2019a). Different from the
overall LSM metric that captures similarity in the use of function
words per session, the reciprocal LSM metric captures the accom-
modation of function words unfolding over adjacent speaking-
turns of a conversation (Müller-Frommeyer et al., 2019a). Al-
though no empirical studies on reciprocal LSM have been
published yet,
1
both overall LSM and reciprocal LSM metrics are
based on the same conceptual underpinnings. The calculation of
reciprocal LSM is also based on Equation 1, but applied to each
pair of successive speaking-turns throughout a session. As a result,
one reciprocal LSM score is obtained for each pair of successive
speaking-turns in the session. These scores represent the time-
series of reciprocal LSM within a session. Additionally, scores can
be assigned to individual speakers, thereby deriving a time-series
for the therapist and a time-series for the client, representing how
much the therapist matched the client and vice versa. Reciprocal
LSM calculations are performed using the Rscript provided by
Müller-Frommeyer, Frommeyer, and Kauffeld (2019b), which can
be retrieved from https://osf.io/arxgu/. The example in Table 1 is
an excerpt from a hypothetical psychotherapy transcript that illus-
trates the use of function words by the therapist (T) and the client
(C) as well as their respective levels of reciprocal LSM.
A Treatment Sample
The levels of overall and reciprocal LSM were calculated for a
subsample of seven 20-session treatments from the Psychoanalytic
Research Consortium. Per treatment, 20 transcribed sessions were
available, reflecting eight early sessions (first year of treatment),
four mid sessions (third year of treatment), and eight late-phase
sessions (fifth year of treatment). This resulted in a total of 140
examined sessions. The three treatment phases illustrate the
change over time and their relation to long-term treatment out-
come. The seven treatments were conducted by five American
male psychoanalysts. Three clients were women. The treatments
averaged 3.6 sessions per week, and the average duration was 653
sessions.
The overall goals were twofold: First, to examine the degree to
which psychoanalytic treatments foster interpersonal synergy, ac-
cording to the level of LSM, as this is manifest within and between
sessions in the early, middle, and late phase of treatment. Second,
to assess whether and how the level of overall and reciprocal LSM
predicts treatment outcomes. Given the uniquely intimate nature of
the therapeutic relationship, LSM levels were expected to be in a
similar range as the previously published LSM levels of sessions
of psychotherapy (training), and interactions between romantic
partners and friends (H1). Considering the expected alliance fluc-
tuations between and within sessions indicated by theories of
change (e.g., cyclical nature of the therapeutic discourse; Mc-
Carthy, Mergenthaler, Schneider, & Grenyer, 2011) and recent
empirical studies (Eubanks, Muran, & Safran, 2018), it was hy-
pothesized that overall LSM and reciprocal LSM would show a
nonlinear pattern over time (H2). Moreover, because therapists try
to build alliance by matching the client, it was expected that
therapists adapt more to their clients than vice versa. Reciprocal
LSM dynamics of speaking-turns were thus expected to show that
therapist’s language style would follow the client’s language style
(H3). Based on the alliance-psychotherapy outcome literature,
both LSM metrics were expected to be negatively related with
symptom severity (H4), and both LSM metrics at early phase were
expected to be positively related to symptom reduction, and im-
proved global functioning (H5).
Preparations
For this subsample of 140 sessions, confidentialized transcripts
were already available. To calculate function word usage within
the therapeutic dyad, transcripts of therapy sessions were seg-
mented by speaker, separating therapist utterances from client
utterances into separate documents. Then, the two text files of each
verbatim transcript per session were manually edited according to
the guidelines put forth by Pennebaker and colleagues in the
Linguistic Inquiry and Word Count (LIWC) coding Manual (Pen-
nebaker, Booth, et al., 2015;Pennebaker, Boyd, Jordan, & Black-
burn, 2015; see Supplemental Material A in the online supplemen-
tal materials). For example, filler words (e.g., “like”) and
nonfluencies (e.g., “um”) are marked so that they would be treated
as fillers rather than as content words. The transcript editing of the
140 transcripts took a total of 210 hr (1.5 hr per 45-min session)
and was split between two research assistants, who spent seven
hours per week on transcript editing for the duration of the semes-
ter (15 weeks).
LSM Data Analyses
The distribution of overall and reciprocal LSM data was ex-
plored in SPSS. The use of an existing dataset and computerized
codings meant that there were no missing data. Because of the
exploratory nature of this pilot study of seven treatments, the
1
Müller-Frommeyer et al. (2019a) describe a median reciprocal LSM
score of .2 and median overall LSM score of .8 for speech turns between
Romeo and Benvolio, taken from William Shakespeare’s (1597) tragedy
Romeo and Juliet. They suggest that the current benchmarks for low
(LSM .60) and high (LSM .85) LSM (Cannava & Bodie, 2017) are
not adequate when using the reciprocal LSM metric.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
512 AAFJES-VAN DOORN, PORCERELLI, AND MU
¨LLER-FROMMEYER
reported LSM analyses are provided as a simple demonstrative
illustration of the application of the method and will need further
replication in larger scale research studies.
To address the first, second, and third research question on the
levels of LSM and LSM change over the course of therapy, rLSM
was calculated per client and therapist speaking-turn, and overall
LSM was calculated per session and during early, mid, and late
sessions. The speaking-turns were averaged per session, and ses-
sions were averaged per treatment phase and over treatment.
Reciprocal LSM analyses were based on ore than 14,000 obser-
vations of speaking-turns in 140 sessions within the seven treat-
ments. A linear mixed-effect model was applied to indicate tem-
poral sequences within and between sessions as well as dynamics
between therapist and client over time. Prior to analysis, equally
sized means were created for early, middle, and late rLSM within
each session. To do so, the overall amount of speaking-turns was
divided by three and the mean of the first six speaking-turns in
each phase was calculated. Linear mixed-effects models were used
to analyze the effect of time (early, middle, and late within and
between sessions) on rLSM, addressing the nesting of speaking-
turns, within sessions, within treatments. For the within-session
analyses, session and treatment were included as random effects.
For the between-session analyses, treatment was included as ran-
dom effect. First a linear model and then a quadratic model was
tested. Maximum random effect structure was used (Barr, 2013).
The analysis was performed in R (R Development Core Team,
2016), using the lme4 library (Bates, Maechler, Bolker, & Walker,
2015).
The fourth and fifth research question (whether and how the
level or change of LSM relates to psychopathology and treatment
outcome) were examined by correlating overall and reciprocal
LSM with early treatment symptom levels and treatment change.
Outcome measures were (a) Global Assessment of Functioning
(GAF; Axis V of the Diagnostic and Statistical Manual of Mental
Disorders, fourth edition, text revision [DSM–IV–TR]; American
Psychiatric Association, 2000), a measure of a client’s overall
functioning designed to track clinical progress, and (b) Personality
Health Index (PHI; Waldron et al., 2011), a measure of a client’s
psychological health-sickness, based upon a normative sample of
clients in psychoanalysis, derived from the Shedler-Westen As-
sessment Procedure (SWAP-200; Shedler & Westen, 1998,2007).
Both outcome measures were rated by trained observers on eight
early sessions at the beginning of treatment (year 1) and on eight
late sessions at the end of treatment (year 5). Change scores
between the early and late rating of both PHI and GAF scales were
calculated. Six clients showed significant improvement on the PHI
and four of the seven clients showed significant improvement on
the GAF, measured at the early (eight sessions year one) and late
phase (eight sessions year 5) in treatment.
Findings Between Sessions
Word count for the clients in the seven treatments ranged from
449 to 11,767 per session (M4,255, SD 1,444), and word
count for the therapists ranged from 30 to 7,290 per session (M
1,096, SD 1,070). The LSM data for the 140 sessions were not
normally distributed, so nonparametric tests were used. In line
with the LSM metric based on relative use of function words (% of
total word count per speaker), LSM per session was significantly
related to client word count (r
s
[140] .33, p.001) and therapist
word count (r
s
[140] .41, p.001), with more verbal speech of
each speaker showing higher similarity of language style, but not
related to overall session word count (r
s
[140] ⫽⫺.09, p.86).
The mean overall LSM for the treatment across the sample was
M.84 (SD .07) and ranged from .51 to .91. This range was
just outside the provided low and high overall LSM benchmarks
(.60 and .85, respectively) and similar to the published data on
overall LSM
2
(e.g., .51 for instant messaging and .90 for conver-
sations among romantic partners) and comparable with values
reported in the previous psychotherapy samples (Borelli et al.,
2019;Lord et al., 2015). Throughout the treatments, the level of
overall LSM per session appeared to remain stable, from early
phase (M.84; SD .04; range .79 –.90), mid phase (M.84;
SD .06, range .76 –.91), to late phase (M.84; SD .04,
range .78.90). The overall LSM session-by-session scores for
the eight early phase, four midphase, and eight late-phase indicated
2
Published data on LSM show that average scores in nontherapeutic
settings range from .51 (SD .05) for instant messaging chatting, .90
(SD .07) for written correspondence, .70 (SD .10) for poetry,
.90 (SD .07) for spoken conversations among romantic partners, and .72
(SD .07) for online writing assignments.
Table 1
Illustration of Function Word Use and Reciprocal LSM Calculations in an Excerpt of a Client–Therapist Interaction
Speaker Statement Function words (%) rLSM
THow is it going? 75
CWell, kind of amess. 50 .80
T Kind of amess, alright, so let us check in with that. 64 .88
CWell, something happened. 67 .98
TWhat happened? 50 .86
CI was in a relationship that ended on Friday. 67 .86
T Ah, Iamsosorry ah, ok. 43 .78
CI was like in a major depressed. . ., got off work Monday, and I could not get out of bed,
and have not eaten in three days and lost weight, you know, slowly getting better but
but it is still in a bad place.
58 .85
MrLSM .86
Note. LSM language style matching; rLSM reciprocal LSM. Function words are written in italics. Calculations across speaking-turns in this example
are based on the rLSM metric by Müller-Frommeyer et al. (2019a). The MrLSM score reflects the mean of the reported reciprocal LSM calculations.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
513
LANGUAGE STYLE MATCHING IN PSYCHOTHERAPY
different pathways of change for the seven clients. See Figure 1 for
a graph with individual trajectories of overall LSM per session for
each of the seven clients over time.
Findings Within Sessions
The within-session variability and temporal dynamics of LSM
per speaking-turn were examined with the reciprocal LSM metric.
The mean reciprocal LSM across the sample was M.48 (SD
.05) and ranged from .38 to .54. The rLSM data were not normally
distributed (skewness and kurtosis more than twice the standard
error).
For within-session trajectories of change in the treatment sam-
ple, first a linear model was tested then a quadratic model was
tested. No linear trends (B⫽⫺.004, SE .008, t⫽⫺0.50, p
.62) or quadratic trends (B⫽⫺.2.355, SE 2.032, t⫽⫺.012,
p.99) within session were identified. For the between-session
trajectories, first a linear model was tested (B⫽⫺.016, SE .006,
t⫽⫺2.75, p.05), then a quadratic model was tested
(B⫽⫺.004, SE .002, t⫽⫺2.44, p.05), indicating that the
trajectories of time between sessions followed a negative (high–
low) linear pattern over time.
Examining the individual reciprocal LSM trajectories for the
seven clients in the treatment sample, results showed that the
therapists’ level of rLSM were significantly lower than the clients’
level of rLSM within sessions (B⫽⫺.004, SE .004, t⫽⫺9.73,
p.001), as well as between sessions (B⫽⫺.109, SE .023,
t⫽⫺4.60, p.01). This suggests that clients followed their
therapist’s language style. Figure 2 illustrates the trajectory of
reciprocal LSM within an early, mid and late phase session from
a case from the treatment sample.
Findings in Relation to Symptoms
Early LSM appeared to be positively associated with change on
the PHI (r
s
[7] .86) and on the GAF (r
s
[7] .83), but not
with early treatment functioning (PHI r
s
[7] ⫽⫺.34; GAF
r
s
[7] ⫽⫺.71). Overall treatment LSM was not related to healthy
functioning early in treatment (PHI r
s
[7] ⫽⫺.29) or PHI change
over treatment (r
s
[7] .68), and appeared to be negatively asso-
ciated with GAF early in treatment (r
s
[7] ⫽⫺.86) and positively
associated with GAF change (r
s
[7] .78). In contrast to our
hypothesis, this would indicate higher levels of function word
similarity in dyads with clients with lower global functioning, and
those who show more improvement over treatment.
Mean reciprocal LSM early in treatment did not appear to be
correlated with client functioning on the GAF (r
s
[7] ⫽⫺.04) but
negatively related with healthy functioning on the PHI
(r
s
[7] ⫽⫺.67) early in treatment. Reciprocal LSM early in treat-
ment appeared to be unrelated to change in functioning (GAF
r
s
[7] ⫽⫺.02; PHI r
s
[7] .21). Different from the overall LSM
findings, mean reciprocal LSM over treatment also did not appear
to be related to client functioning early in treatment (PHI
r
s
[7] ⫽⫺.40; GAF r
s
[7] .50) or change on the PHI
(r
s
[7] ⫽⫺.29) or the GAF (r
s
[7] ⫽⫺.56). For a visual inspection
of the correlations scatterplots are provided in Supplemental Ma-
terial B in the online supplemental materials.
In sum, in this treatment sample, early overall LSM appeared to
be positively related to change in client functioning over treatment
(on the PHI and GAF). Overall treatment levels of LSM appeared
to be negatively related to levels of functioning on the GAF in
these same early sessions but positively related to change in GAF.
Notably, given the very small treatment sample (N7 clients),
indications of associations between variables are only exploratory
in nature and need to be replicated in statistical analyses of larger
samples before they can be interpreted with confidence.
A Demonstration of the Clinical Usefulness of LSM
Case Study
To demonstrate the clinical usefulness of the LSM approach
in psychotherapy, the psychoanalytic treatment of Mr. A will be
Figure 1. Individual trajectories of language style matching (LSM) per session for each of the seven clients
over time. See the online article for the color version of this figure.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
514 AAFJES-VAN DOORN, PORCERELLI, AND MU
¨LLER-FROMMEYER
reported (see Porcerelli, Dauphin, Ablon, Leitman, & Bambery,
2007). In this 5-year treatment of a 50-year-old male (Mr. A)
diagnosed with an avoidant personality disorder (AVPD), dif-
ferent aspects of the therapeutic alliance (observer-rated work-
ing alliance by way of agreement of task, goals, and bond and
computerized assessment of LSM) were triangulated, in relation
to treatment outcome. In addition to meeting criteria for AVPD,
Mr. A reported significant psychological distress, object rela-
tions pathology, and moderately severe interpersonal impair-
ment at intake. At the outcome of treatment and at 1-year
follow-up, Mr. A reported clinically significant improvements
in personality functioning, symptom severity, and interperso-
nal functioning. Gains were maintained at 1-year follow-up
(Porcerelli et al., 2007). Fifteen transcribed audio-recorded
sessions (three at the end of each of five treatment years) were
available.
This examination of the transcripts of the 15 recorded sessions
provides the first systematic case study of interpersonal synergy
over the course of a long-term psychoanalytic treatment of AVPD.
To assess the construct validity of LSM, the relationship between
overall LSM per session was also examined with the correspond-
ing observer-rated alliance scores on the Working Alliance Inven-
tory (WAI). Two independent raters completed the WAI-Observer
scale (WAI-O; Tichenor & Hill, 1989) for each of the 15 available
sessions. Then, change in overall LSM and WAI-O was examined
over the five treatment years, analyzing three sessions per year
(total of 15 sessions). Average scores of the raters were used in
further analyses.
As hypothesized for the treatment sample described earlier, the
LSM levels in this case study were expected to be in a similar
range as previously published LSM levels (H1). A nonlinear
sequence of reciprocal LSM was expected within sessions, with a
similar pattern of overall LSM throughout treatment over time
(H2). Moreover, it was expected that the therapist’s language style
would follow Mr A’s language style (H3). This case study did not
allow for examination of the relationship between LSM and psy-
chopathology (H4) or treatment outcome (H5). Given the concep-
tual convergence of LSM with aspects of the alliance, especially
WAI-O goals as proxy of interpersonal synergy, it was hypothe-
sized that the trajectory of change in LSM would be positively
related to observer-rated measurements of alliance (WAI-O) for
goals, task, and bond (H6).
Findings Between Sessions
As to be expected in psychoanalytic treatment, the client con-
tributed more to the conversations than the therapist. Word count
for Mr. A ranged from 1,655 to 3,063 per session (M2,672,
SD 376), whereas the word count for the therapist ranged from
74 to 414 per session (M185, SD 98). The overall LSM data
were normally distributed (skewness & kurtosis less than twice the
standard error). The average overall LSM for Mr. A and his
therapist throughout the 15 examined sessions was .80 (SD .06),
ranging from the lowest score of .69 (Session 7, year 3) to the
highest score of .89 (Session 8, year 3 and Session 13, year 5). This
is in line with previous publications using this overall LSM metric.
Changes in mean WAI-O and overall LSM scores over the five
years of treatment (see Figure 3 and 4for mean WAI-O and overall
LSM scores per year) indicate that, unlike the working alliance,
which remained relatively stable over time, the level of interper-
sonal synergy was highest in the first year of treatment (overall
LSM .84). In this first year of treatment Mr. A also showed the
highest levels of symptom severity and resistance. When Mr. A
started to show an improvement in symptoms and general func-
Figure 2. Trajectories of reciprocal language style matching (LSM) in the
first 10 speaking-turns from a session in the early-phase, middle-phase and
late-phase of a treatment. rLSM reciprocal LSM.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
515
LANGUAGE STYLE MATCHING IN PSYCHOTHERAPY
tioning (after the first year), the levels of interpersonal synergy
reduced.
Looking at the sessions individually (see Figure 5 for overall
LSM scores for the 15 sessions, in the five treatment years), the
overall LSM levels changed substantially between sessions, rang-
ing from an increase of .20 between Session 7 and 8 and a decrease
of .12 between Session 8 and 9. This suggests that there was a lot
of variability in overall LSM within the third treatment year,
something that did not become apparent when analyzing mean
LSM scores per year. It is notable that in Session 7 and 8 change
was particularly large for LSM (from .69 to .89; lowest to highest
in treatment) and the WAI bond (5.25 to 4.67; highest to lowest in
treatment). At this point in the treatment, the topic of the session
became more intense, focusing on Mr. A’s aggressive impulses
and avoidance of his wife instead of his driving phobia. Addition-
ally, the therapist was more inquisitive and directly challenged Mr.
A’s avoidance behaviors. In other words, the initial supportive
tracking of Mr A.’s experience (i.e., building alliance) was re-
placed by more challenging explicit defense interpretations.
Findings Within Sessions
The mean reciprocal LSM across the 15 sessions was M.51
(SD .05) and ranged from .41 to .61. There was an average of
22.7 (SD 8.81) speaking turns per session. The rLSM data were
not normally distributed (skewness and kurtosis more than twice
the standard error). For within-session trajectories of change in the
case study, session and year were included as random effects. For
the between session analysis, year was included as random effect.
First a linear model was tested (B.037, SE .028, t1.309,
Figure 3. Mean levels of Working Alliance Inventory-Observer (WAI-O) per year of treatment. See the online
article for the color version of this figure.
Figure 4. Overall language style matching (LSM) scores.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
516 AAFJES-VAN DOORN, PORCERELLI, AND MU
¨LLER-FROMMEYER
p.26), then a quadratic model was tested (B.007, SE .007,
t.98, p.37). For the between-session trajectories, first a linear
model was tested (B.003, SE .018, t.17, p.88), then a
quadratic model was tested (B.0001, SE .005, t.02, p
.98). No linear or quadratic trends were identified in the data
within and between session.
To test the hypothesis that the therapist adapts to the client’s
language style, the variability of LSM was calculated at the level
of speaking-turns using the rLSM metric. For the case study, the
therapist’s level of rLSM was significantly lower than the client’s
level of rLSM within sessions (B⫽⫺.169, SE .025, t⫽⫺6.84,
p.001), as well as between sessions (B⫽⫺.18, SE .022,
t⫽⫺8.51, p.001). Thus, as was found in the treatment sample,
the client appears to adapt to the therapist within sessions and over
treatment. This temporal dynamic might imply that the therapist
impacts what the client says throughout the conversation. Even
though in these psychoanalytic psychotherapy sessions the thera-
pist only utters very few words, the way in which the therapist
speaks seems to highly influence the client’s response. It is pos-
sible that the therapist, even in a few words, leads the client’s
response, and thus takes charge of the process in the session in an
implicit way. In line with the findings by Danescu-Niculescu-Mizil
et al. (2012), clients (with a lower perceived social status) appear
to modify their word choice to match their therapist (perceived as
having a higher status). Moreover, compared with therapists, cli-
ents tend to have a higher level of symptoms/psychopathology and
lower self-esteem, and therefore might more easily adapt to the
therapist. In contrast to Bucci and Maskit (2007)’s theory of
mutual adaptation, likened to dancing the tango, it appears that in
psychoanalytic interactions it is clear who is leading whom, at least
with regard to language style.
Findings in Relation to Alliance
The overall LSM data and the WAI data met the assumption of
independent errors. The scatterplot of standardized predicted val-
ues (i.e., standardized residuals) showed that the LSM and WAI
data met the assumptions of homogeneity of variance and linearity.
Tests to see whether the data met the assumption of collinearity
indicated that multicollinearity was not a concern (Time and LSM,
Tolerance .92, variance inflation factor 1.08). It is acknowl-
edged that the following statistical analyses of overall LSM and
alliance are based on this sample size of 15 sessions throughout
five treatment years and thus only offer an illustration of the
potential of the LSM method. Further replication in larger scale
research studies is required before conclusions can be drawn.
When holding time constant, LSM was positively correlated to
the WAI-O Goal subscale, r.36, p.21, reaching a moderate
effect size, albeit not significant. LSM was unrelated to the WAI-O
Task subscale, r.01, p.97. Contrary to the hypothesis, LSM
was negatively correlated to the WAI-O Bond subscale, r⫽⫺.54,
p.05, reaching a large effect size. Reciprocal LSM, controlled
for time, was also negatively correlated to the WAI-O Bond
subscale, r⫽⫺.61, p.05, reaching a large effect size, but not
related to the other subscales (all ps.06). See Supplemental
Material B in the online supplemental materials for scatterplots of
all correlations.
These preliminary indications of a positive relationship between
overall LSM and WAI-O Goal combined with the negative rela-
tionship between overall and reciprocal LSM with WAI-O Bond
may suggest that clients with AVPD are more comfortable with
working toward agreed-on goals in psychoanalysis than they are
acknowledging a deepening of the relationship (bond) with the
therapist. In other words, the WAI-O Task and goal subscale may
tap into efforts by therapist and client to agree, showing a positive
correlation with LSM, whereas WAI-O Bond might reflect the
relative ease and comfort of the session experience, which is
negatively related to the hard work and effort of reaching inter-
personal synergy. Thus, different from short interventions where,
arguably, alliance building is more explicitly focused on making
the client feel happy and comfortable, in long-term psychoanalytic
treatments LSM might reflect the initial building of tolerance,
before more challenging interpretations and relationship dynamics
are worked through.
Possibly, clients with AVPD long for closeness with others but
are more comfortable with relating to others in fantasy, because
real relationships can be extremely anxiety-provoking for them. It
is also possible that in psychoanalysis, a positive correlation be-
tween LSM and self-reported working alliance reflects the thera-
pist’s automatic efforts to lessen the client’s fear/resistance,
whereas a negative correlation between LSM and self-reported
alliance bond reflects less fear/resistance on the client’s part and
thus less need for the therapist to provide a sense of safety and
similarity. In other words, when a client appears comfortable/high
functioning (reflected by a high working alliance-bond), the ther-
Figure 5. Language style matching (LSM) per session per treatment year.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
517
LANGUAGE STYLE MATCHING IN PSYCHOTHERAPY
apist might be able to offer more challenging interpretations and
create a therapeutic dissonance to stimulate change (reflected by
lower LSM). Building on the reported case study, future research
with larger samples may help to clarify the role of interpersonal
synergy as a possible implicit aspect of the alliance during long-
term treatment. This case study underscores the need to study
multiple domains of client–therapist relating to understand the
complexity of change processes and suggests that LSM may tap
into a relationship component, not previously captured by ratings
on traditional observer-rated measures of working alliance.
Taken together, these two exploratory examples illustrate how
interpersonal synergy, according to the level of LSM, may mani-
fest within and between sessions over long-term psychoanalytic
treatments and in relation to alliance and treatment outcome. More
specifically, the preliminary findings may be summarized as fol-
lows:
1. The LSM levels appeared to be in similar range as the
previously published LSM levels of sessions of psycho-
therapy (training) and interactions between romantic
partners and friends (H1).
2. Overall, LSM per session and reciprocal LSM within
sessions appeared to remain relatively constant with no
linear or quadratic pattern over time, although trajectories
among the different therapist– client dyads varied widely.
Throughout long-term treatments, a linear high–low se-
quence of reciprocal LSM indicated that LSM within
sessions might reduce over time (H2).
3. The temporal dynamic between therapist and client in
these preliminary session data indicated that client’s lan-
guage style followed the therapist’s language style, im-
plying that clients may adapt to their therapist within
sessions as well as over treatment (H3).
4. In the treatment sample, LSM seemed to be positively
related to symptom severity early in treatment (H4).
5. In the treatment sample, early phase overall LSM per
session appeared to be positively related to symptom
reduction and improved global functioning. Reciprocal
LSM was unrelated to psychopathology and treatment
outcome (H5).
6. In the case study, overall and reciprocal LSM appeared to
be negatively related to self-reported alliance bond (H6).
Wider Applicability and Common Pitfalls
Applicability of LSM in the Field of Counseling
In line with Borelli et al. (2019), LSM appears to hold promise
as a metric of interpersonal synergy, an implicit aspect of the
therapeutic alliance. The LSM metric has potential for research
and clinical applications. First, the practical advantages of com-
puterized quantitative speech analysis of function words, including
its reliability, objectivity, and cost-effectiveness (Tausczik & Pen-
nebaker, 2010), might make it a promising tool for counseling and
psychotherapy research. Automated text analysis programs such as
LIWC may potentially provide a means of analyzing dyadic dy-
namics when self-report data cannot be collected. For example, it
could be applied as post hoc process analysis on a previously
collected sample of therapy transcripts. This means that existing
psychotherapy data sets may be analyzed retrospectively, allowing
for testing of the LSM construct validity relatively easily.
Second, LSM may be a useful complement to self-report data.
The overall and reciprocal LSM metrics may provide a novel way
to study some of the more elusive aspects of the therapeutic
alliance and the therapeutic process, not captured by self-report
measures of attunement, alliance, or empathy or physiological
measures of body movements. LSM is unlikely to be influenced
consciously and may be too subtle to be perceived by coders; it
thus allows examination beyond the verbal content of the interac-
tion to the content-free functional verbal interaction. Nonverbal
synchrony has shown to facilitate the working alliance, which in
turn promotes the client’s emotion-regulatory skills (Koole &
Tschacher, 2016). This means that the LSM metric could poten-
tially be applied in conjunction with measures of affect experienc-
ing in-session and help elucidate the role of implicit aspects of
alliance in affect regulation and tolerance in psychotherapy (Håvås
et al., 2015).
Third, LSM measurements may allow for useful comparisons
between different treatment models. The LSM metrics rely on the
ratio of function words relative to general word count per speaker,
which means that different types of psychotherapies in which
therapists tend to be more or less verbally active (e.g., psycho-
analysis vs. CBT) might still result in similar levels of LSM. Once
this pan-theoretical LSM metric is validated, it may help to trans-
late intuitive processes that are familiar to psychoanalytic clini-
cians, into a measurable construct that can be communicated to
counselors more broadly. In future, the reciprocal LSM metric
could help identify trajectories of relational change over time
(either within-session, session-by-session or from early, mid and
late phase of treatment), illustrating how meaningful interaction is
achieved by speakers on a turn-by-turn basis in different treatment
models. LSM may be both predicted by prior relationship experi-
ences and predictive of the quality of future interpersonal pro-
cesses occurring between therapist and client. Given the consistent
relationship between quality of alliance and treatment outcome,
future research may explore whether LSM levels in early sessions
prove to have significant prognostic value, potentially predicting
the self-reported working alliance and eventual treatment outcome,
and signaling cases that are unlikely to do well.
If the overall and reciprocal LSM metrics are validated in these
research applications, a possible future clinical application lies in
improving clinical training programs. Therapists vary substantially
in clinical effectiveness, and at least some of these variations are
attributable to their different abilities in forming a strong alliance
(Del Re, Flückiger, Horvath, Symonds, & Wampold, 2012).
Whether or not levels of function words are automatically coordi-
nated, it is the therapists’ responsibility to notice when they are
matched to clients and when they are not. In the future, the LSM
metrics might offer a medium in which to provide reliable feed-
back to counselors regarding their client’s ability to match their
language style. In this manner, LSM could help counselors to build
and strengthen their clinical assessment expertise. The next step
could then be to provide alliance trainings, in which tracking LSM
is explicitly taught. It should be emphasized, however, that attuned
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
518 AAFJES-VAN DOORN, PORCERELLI, AND MU
¨LLER-FROMMEYER
responses, especially the implicit ones, reflect essentially an intu-
itive, spontaneous, and immediate activity and are to a lesser
degree driven by cognitive processes. It is thus reasonable to
assume that this may limit the extent to which interpersonal
synergy can be fully learned (Håvås et al., 2015).
Avoiding Common Pitfalls in Applying LSM and
Interpreting the Findings
Several potential pitfalls were identified in applying LSM to
psychotherapy treatments. As is the case with other text-analysis
programs, computerized analysis with the LIWC program offers
the option of datamining; exploring the relationships between all
70 possible word categories (different content words, themes and
stylistic differences) to search for significant findings, not based on
predetermined hypotheses (Rasmussen, Malchow-Møller, & An-
dersen, 2011). Also, the fact that one could potentially run the
LIWC program with raw transcripts that are not edited, without an
error message, makes it possible to generate invalid results and cut
corners that might not be picked up by reviewers or collaborators.
Besides these potential practical pitfalls, there are several other
important pitfalls that must be avoided when interpreting the LSM
findings. First, the output and interpretation of high LSM scores
might be somewhat misleading. If both people use the same
percentage of function words, their level of matching is deemed
100%. This implies that a high LSM score can represent people not
matching their language style; people can match by not matching.
This could be a problem for how LSM is conceptualized, and LSM
results are interpreted and how it functions to predict treatment
outcomes (Müller-Frommeyer et al., 2019a).
A second potential pitfall in interpreting LSM findings is that it
currently remains unclear how exactly LSM is related to the
operationalized constructs of therapeutic alliance, synchrony and
attunement. Snyder and Silberschatz (2017) conclude in their
validity study that attunement may well be a more specific sub-
component of the therapeutic alliance. Thus, in line with this case
study illustration, the next research step should be to test the
construct validity of LSM in larger treatment samples. This would
help to determine the degree of convergence between LSM and an
established measure of the alliance such as the WAI and test
whether the LSM metric predicts the alliance and subsequent
treatment outcome. Similarly, future research on the WAI and
LSM predicting outcome would allow for the testing of the incre-
mental validity of LSM.
A third possible pitfall in interpreting the LSM results is draw-
ing conclusions without having a benchmark for the overall and
reciprocal LSM metrics in good and bad outcome treatments. Too
few empirical studies have been conducted so far to be conclusive
about the meaning of the LSM trajectories, within sessions, and
over treatment, and the differences between treatments. Also, it is
possible that LSM research that only includes a subsample of
segments within sessions or a subsample of sessions within treat-
ment does not validly represent the nonlinear process of interper-
sonal synergy in therapy. Mergenthaler (1996)’s Therapeutic
Cycles Model, for example, suggests that gains accumulate
throughout treatments and that linguistic change is nonlinear,
occurring in one or more cycles per treatment (e.g., McCarthy et
al., 2011). Moreover, the LSM metric does not currently address
any cultural/class differences in use of function words, implied by
the fact that not all of the function word categories are available in
the dictionaries of all languages.
Like many other new approaches, the application of LSM relies
on knowledge from other disciplines, which has been imported
into the field of counseling and psychotherapy research only
recently. Therefore, little systematic work has been done on the
unique issues that arise in the process of integrating LSM in the
examination of the client–therapist dyad. It remains unclear to
what extent the previous studies conducted in other social sciences
offer a relevant benchmark for LSM in psychotherapy. The two
published studies (Borelli et al., 2019;Lord et al., 2015), together
with the two psychoanalytic pilot studies reported in this article,
offer an initial key point of reference for future empirical investi-
gation. Based on the initial data reported here, it can be concluded
that the Ireland and Pennebaker (2010) overall LSM metric only
captures a balanced use of function words between two conversa-
tional partners rather than dynamic coordination. Hence, in psy-
chotherapy, where conversational dependency is assumed, the new
reciprocal LSM metric is arguably most clinically relevant.
Remaining Questions and Directions for
Future Research
The LSM metrics may create several unique opportunities in the
future for the examination of therapist– client interactions, partic-
ularly those aspects which elude conscious awareness of the ther-
apist, client, or outside observer, that nonetheless exert an influ-
ence on treatment outcome. First, in larger scale studies, therapist
trait and state variables that may influence the level of implicit
alliance, such as countertransference and attachment style (Mi-
kulincer & Shaver, 2008), could be examined as moderating fac-
tors. Similarly, further research into specific strategies that might
change the level of interpersonal synergy in sessions could be
important.
Also, given the reported pilot data, clients’ levels of function
word use early in treatment might reflect their psychopathology.
Interpersonal synergy is thus expected to be higher in the treat-
ments of clients with more severe psychopathology. The appro-
priate separation from the therapist may then be reflected by lower
LSM scores at the end of successful treatments. Perhaps clients
who are more sensitive to interpersonal rejection (such as clients
with personality disorders) might be more sensitive to momentary
fluctuations in the therapist’s language style, whereas those clients
with more robust pretreatment psychological functioning might
show less need for interpersonal synergy. The same could be true
for therapists. Future large-scale research is needed in which the
moderating effect of client psychopathology and therapist factors
can be examined.
Moreover, the level or range of function words clients and
therapists use might reflect their culture, educational level, and
upbringing, more than the dynamic response to the current situa-
tion of the conversation. Therefore, researchers should also exam-
ine aspects of therapist— client fit (e.g., racial/ethnic/gender
matching) in terms of their contribution to LSM. Also, future
research is needed to examine how function word usage might
differ in different types of discourse within psychotherapy, ranging
from concrete language (e.g., symptom descriptions, early mem-
ories, dreams, and relationship episodes) to philosophical dis-
course (e.g., interpretations or insight). In line with expected
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
519
LANGUAGE STYLE MATCHING IN PSYCHOTHERAPY
differences in alliance between treatment orientations and clients
(Zilcha-Mano & Errázuriz, 2015), further larger-scale research is
warranted that allows for comparisons of levels of LSM between
treatments, therapists and clients. This will allow researchers to
disentangle trait-like from state-like characteristics of interper-
sonal synergy.
Furthermore, in future research, it may be useful to adopt a
construct validation framework (e.g., Cronbach & Meehl, 1955)
for assessing the overall and reciprocal LSM metrics as a measure
of alliance. LSM could, for example, be compared with other
computerized programs of language style analysis, such as the
Discourse Attributes Analysis Program (Bucci & Maskit, 2006). In
addition to triangulation with verbal process measures, LSM anal-
ysis could be usefully complemented by nonverbal indicators of
implicit relational behavior, including acoustic (Imel, Steyvers, &
Atkins, 2015), vocal rhythm coordination (Håvås et al., 2015),
movement synchrony (Ramseyer & Tschacher, 2011), and physi-
ological measures of heartrate and skin conductance (Marci, Ham,
Moran, & Orr, 2007). Using these approaches will increase meth-
odological pluralism (e.g., Barber & Sharpless, 2009) and will help
provide a more comprehensive description of the implicit thera-
peutic process.
In sum, the intention in this article was to describe and test the
overall and reciprocal LSM metrics, conceptualized as a proxy of
interpersonal synergy, for its potential use in counseling and psy-
chotherapy research. A theoretical framework was provided in
which to situate the potential utility of the LSM construct, and
hypotheses were explored within a pilot treatment sample and a
single case study of long-term psychoanalytic treatment. These
two empirical examples illustrate how LSM could yield further
insight into the extent to which and mechanisms by which implicit
aspects of the alliance, different from observer-rated alliance-
bond, predict treatment outcome. Future larger-scale research into
the relationship between LSM, clients and therapists’ variables, as
well as LSM differences between treatments over time is war-
ranted.
References
American Psychiatric Association. (2000). Diagnostic and statistical man-
ual of mental disorders (DSM–IV–TR). Washington, DC: Author.
Andrade, V. M. (2005). Affect and the therapeutic action of psychoanal-
ysis. The International Journal of Psychoanalysis, 86, 677– 697. http://
dx.doi.org/10.1516/YHJK-63WN-QX6X-KYP2
Babcock, M. J., Ta, V. P., & Ickes, W. (2014). Latent semantic similarity
and language style matching in initial dyadic interactions. Journal of
Language and Social Psychology, 33, 78 – 88. http://dx.doi.org/10.1177/
0261927X13499331
Baldwin, S. A., Wampold, B. E., & Imel, Z. E. (2007). Untangling the
alliance-outcome correlation: Exploring the relative importance of ther-
apist and patient variability in the alliance. Journal of Consulting and
Clinical Psychology, 75, 842– 852. http://dx.doi.org/10.1037/0022-006X
.75.6.842
Barber, J. P., & Sharpless, B. A. (2009). New methods—more questions:
A commentary on interdisciplinary dialogues. Psychotherapy Research,
19, 644 – 648. http://dx.doi.org/10.1080/10503300903045131
Barr, D. J. (2013). Random effects structure for testing interactions in
linear mixed-effects models. Frontiers in Psychology, 4, 328. http://dx
.doi.org/10.3389/fpsyg.2013.00328
Bates, D., Maechler, M., Bolker, B., & Walker, S. (2015). lme4: Linear
mixed-effects models using Eigen and S4 (R package version 1.1–7)
[Computer software]. Retrieved from http://CRAN.R-project.org/
packagelme4
Blairy, S., Herrera, P., & Hess, U. (1999). Mimicry and the judgment of
emotional facial expressions. Journal of Nonverbal Behavior, 23, 5– 41.
http://dx.doi.org/10.1023/A:1021370825283
Borelli, J. L., Sohn, L., Wang, B. A., Hong, K., DeCoste, C., & Suchman,
N. E. (2019). Therapist– client language matching: Initial promise as a
measure of therapist– client relationship quality. Psychoanalytic Psy-
chology, 36, 9 –18.
Bowen, J. D., Winczewski, L. A., & Collins, N. L. (2017). Language style
matching in romantic partners’ conflict and support interactions. Journal
of Language and Social Psychology, 36, 263–286. http://dx.doi.org/10
.1177/0261927X16666308
Bucci, W., & Maskit, B. (2006). A Weighted Referential Activity Diction-
ary. In J. G. Shanahan, Y. Qu, & J. Wiebe (Eds.), Computing attitude
and affect in text: Theory and applications (pp. 49 60). Dordrecht, the
Netherlands: Springer: http://dx.doi.org/10.1007/1-4020-4102-0_6
Bucci, W., & Maskit, B. (2007). Beneath the surface of the therapeutic
interaction: The psychoanalytic method in modern dress. Journal of the
American Psychoanalytic Association, 55, 1355–1397. http://dx.doi.org/
10.1177/000306510705500412
Cannava, K. (2018). Language style matching (LSM), (Ireland & Penne-
baker, 2010). In D. L. Worthington & G. D. Bodie (Eds.), The source-
book of listening research: Methodology and measures (pp. 348 –353).
Hoboken, NJ: Wiley.
Cannava, K., & Bodie, G. D. (2017). Language use and style matching in
supportive conversations between strangers and friends. Journal of So-
cial and Personal Relationships, 34, 467– 485. http://dx.doi.org/10
.1177/0265407516641222
Cronbach, L. J., & Meehl, P. E. (1955). Construct validity in psychological
tests. Psychological Bulletin, 52, 281–302. http://dx.doi.org/10.1037/
h0040957
Danescu-Niculescu-Mizil, C., Lee, L., Pang, B., & Kleinberg, J. (2012).
Echoes of power: Language effects and power differences in social
interaction. Proceedings of the 21st International Conference on World
Wide Web (pp. 699 –708). New York, NY: Association Computational
Machinery. http://dx.doi.org/10.1145/2187836.2187931
Del Re, A. C., Flückiger, C., Horvath, A. O., Symonds, D., & Wampold,
B. E. (2012). Therapist effects in the therapeutic alliance-outcome rela-
tionship: A restricted-maximum likelihood meta-analysis. Clinical Psy-
chology Review, 32, 642– 649. http://dx.doi.org/10.1016/j.cpr.2012.07
.002
Diaz, M. T., & McCarthy, G. (2009). A comparison of brain activity
evoked by single content and function words: An fMRI investigation of
implicit word processing. Brain Research, 1282, 38 – 49. http://dx.doi
.org/10.1016/j.brainres.2009.05.043
Eubanks, C. F., Muran, J. C., & Safran, J. D. (2018). Alliance rupture
repair: A meta-analysis. Psychotherapy, 55, 508 –519. http://dx.doi.org/
10.1037/pst0000185
Fonagy, P., Gergely, G., Jurist, E., & Target, M. (2002). Mentalization,
affective regulation and the development of the self. New York, NY:
Basic Books.
Fusaroli, R., Bahrami, B., Olsen, K., Roepstorff, A., Rees, G., Frith, C., &
Tylén, K. (2012). Coming to terms: Quantifying the benefits of linguistic
coordination. Psychological Science, 23, 931–939. http://dx.doi.org/10
.1177/0956797612436816
Fusaroli, R., Ra˛czaszek-Leonardi, J., & Tylén, K. (2014). Dialog as inter-
personal synergy. New Ideas in Psychology, 32, 147–157. http://dx.doi
.org/10.1016/j.newideapsych.2013.03.005
Gonzales, A. L., Hancock, J. T., & Pennebaker, J. W. (2010). Language
style matching as a predictor of social dynamics in small groups.
Communication Research, 37, 3–19. http://dx.doi.org/10.1177/
0093650209351468
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
520 AAFJES-VAN DOORN, PORCERELLI, AND MU
¨LLER-FROMMEYER
Hatcher, R. L., Barends, A., Hansell, J., & Gutfreund, M. J. (1995).
Patients’ and therapists’ shared and unique views of the therapeutic
alliance: An investigation using confirmatory factor analysis in a nested
design. Journal of Consulting and Clinical Psychology, 63, 636 – 643.
http://dx.doi.org/10.1037/0022-006X.63.4.636
Håvås, E., Svartberg, M., & Ulvenes, P. (2015). Attuning to the unspoken:
The relationship between therapist nonverbal attunement and attachment
security in adult psychotherapy. Psychoanalytic Psychology, 32, 235–
254. http://dx.doi.org/10.1037/a0038517
Hilsenroth, M. J., Blagys, M. D., Ackerman, S. J., Bonge, D. R., & Blais,
M. A. (2005). Measuring psychodynamic-interpersonal and cognitive-
behavioral techniques: Development of the Comparative Psychotherapy
Process Scale. Psychotherapy: Theory, Research, Practice, Training, 42,
340 –356. http://dx.doi.org/10.1037/0033-3204.42.3.340
Hölzer, M., Mergenthaler, E., Pokorny, D., Kächele, H., & Luborsky, L.
(1996). Vocabulary measures for the evaluation of therapy outcome:
Re-studying transcripts from the Penn Psychotherapy Project. Psy-
chotherapy Research, 6, 95–108. http://dx.doi.org/10.1080/
10503309612331331618
Horvath, A. O., & Bedi, R. P. (2002). The alliance. In J. C. Norcross (Ed.),
Psychotherapy relationships that work: Therapist contributions and
responsiveness to patients (pp. 37– 69). Oxford, UK: Oxford University
Press.
Imel, Z. E., Steyvers, M., & Atkins, D. C. (2015). Computational psycho-
therapy research: Scaling up the evaluation of patient-provider interac-
tions. Psychotherapy, 52, 19 –30. http://dx.doi.org/10.1037/a0036841
Ireland, M. E., & Pennebaker, J. W. (2010). Language style matching in
writing: Synchrony in essays, correspondence, and poetry. Journal of
Personality and Social Psychology, 99, 549 –571. http://dx.doi.org/10
.1037/a0020386
Ireland, M. E., Slatcher, R. B., Eastwick, P. W., Scissors, L. E., Finkel,
E. J., & Pennebaker, J. W. (2011). Language style matching predicts
relationship initiation and stability. Psychological Science, 22, 39 – 44.
http://dx.doi.org/10.1177/0956797610392928
Kantrowitz, J. L., Katz, A. L., Greenman, D. A., Morris, H., Paolitto, F.,
Sashin, J., & Solomon, L. (1989). The patient-analyst match and the
outcome of psychoanalysis: A pilot study. Journal of the American
Psychoanalytic Association, 37, 893–919. http://dx.doi.org/10.1177/
000306518903700402
Koole, S. L., & Tschacher, W. (2016). Synchrony in psychotherapy: A
review and an integrative model of the therapeutic alliance. Frontiers in
Psychology, 7, 862. http://dx.doi.org/10.3389/fpsyg.2016.00862
Lord, S. P., Sheng, E., Imel, Z. E., Baer, J., & Atkins, D. C. (2015). More
than reflections: Empathy in motivational interviewing includes lan-
guage style synchrony between therapist and client. Behavior Therapy,
46, 296 –303. http://dx.doi.org/10.1016/j.beth.2014.11.002
Lumsden, J., Miles, L. K., Richardson, M. J., Smith, C. A., & Macrae,
C. N. (2012). Who syncs? Social motives and interpersonal coordina-
tion. Journal of Experimental Social Psychology, 48, 746 –751. http://
dx.doi.org/10.1016/j.jesp.2011.12.007
Marci, C. D., Ham, J., Moran, E., & Orr, S. P. (2007). Physiologic
correlates of perceived therapist empathy and social-emotional process
during psychotherapy. Journal of Nervous and Mental Disease, 195,
103–111. http://dx.doi.org/10.1097/01.nmd.0000253731.71025.fc
McCarthy, K. L., Mergenthaler, E., Schneider, S., & Grenyer, B. F. (2011).
Psychodynamic change in psychotherapy: Cycles of patient-therapist
linguistic interactions and interventions. Psychotherapy Research, 21,
722–731. http://dx.doi.org/10.1080/10503307.2011.615070
McWilliams, N. (2011). The psychodynamic diagnostic manual: An effort
to compensate for the limitations of descriptive psychiatric diagnosis.
Journal of Personality Assessment, 93, 112–122. http://dx.doi.org/10
.1080/00223891.2011.542709
Mergenthaler, E. (1996). Emotion-abstraction patterns in verbatim proto-
cols: A new way of describing psychotherapeutic processes. Journal of
Consulting and Clinical Psychology, 64, 1306 –1315. http://dx.doi.org/
10.1037/0022-006X.64.6.1306
Meyer, A. S., & Bock, K. (1999). Representations and processes in the
production of pronouns: Some perspectives from Dutch. Journal of
Memory and Language, 41, 281–301. http://dx.doi.org/10.1006/jmla
.1999.2649
Mikulincer, M., & Shaver, P. R. (2008). Adult attachment and affect
regulation. In J. Cassidy & P. R. Shaver (Eds.), Handbook of attach-
ment: Theory, research, and clinical application (pp. 503–531). New
York, NY: Guilford Press.
Muir, K., Joinson, A., Cotterill, R., & Dewdney, N. (2016). Characterizing
the linguistic chameleon: Personal and social correlates of linguistic
style accommodation. Human Communication Research, 42, 462– 484.
http://dx.doi.org/10.1111/hcre.12083
Müller-Frommeyer, L. C., Frommeyer, N. A. M., & Kauffeld, S. (2019a).
Introducing rLSM: An integrated metric assessing temporal reciprocity
in language style matching. Behavior Research Methods, 51, 1343–
1359. http://dx.doi.org/10.3758/s13428-018-1078-8
Müller-Frommeyer, L. C., Frommeyer, N. A., & Kauffeld, S. (2019b).
RScript for the calculation of rLSM scores [OSFHOME FILES web-
page]. Retrieved from https://osf.io/arxgu/
Niederhoffer, K. G., & Pennebaker, J. W. (2002). Linguistic style matching
in social interaction. Journal of Language and Social Psychology, 21,
337–360. http://dx.doi.org/10.1177/026192702237953
Pennebaker, J. W. (2011). The secret life of pronouns: What our words say
about us. New York, NY: Bloomsbury Publishing. http://dx.doi.org/10
.1016/S0262-4079(11)62167-2
Pennebaker, J. W., Booth, R. J., Boyd, R. L., & Francis, M. E. (2015).
Linguistic inquiry and word count: LIWC2015 operator’s manual. Mah-
wah, NJ: Erlbaum.
Pennebaker, J. W., Booth, R. J., & Francis, M. E. (2007). Linguistic inquiry
and word count: LIWC [Computer software]. Austin, TX: Liwc. Net.
Pennebaker, J. W., Boyd, R. L., Jordan, K., & Blackburn, K. (2015). The
development and psychometric properties of LIWC2015. Retrieved from
https://repositories.lib.utexas.edu/handle/2152/31333
Porcerelli, J. H., Dauphin, V. B., Ablon, J. S., Leitman, S., & Bambery, M.
(2007). Psychoanalysis with avoidant personality disorder: A systematic
case study. Psychotherapy: Theory, Research, Practice, Training, 44,
1–13. http://dx.doi.org/10.1037/0033-3204.44.1.1
Ramseyer, F., & Tschacher, W. (2011). Nonverbal synchrony in psycho-
therapy: Coordinated body movement reflects relationship quality and
outcome. Journal of Consulting and Clinical Psychology, 79, 284 –295.
http://dx.doi.org/10.1037/a0023419
Rasmussen, O. D., Malchow-Møller, N., & Andersen, T. B. (2011). Walk-
ing the talk: The need for a trial registry for development interventions.
Journal of Development Effectiveness, 3, 502–519. http://dx.doi.org/10
.1080/19439342.2011.605160
R Development Core Team. (2016). R: A language and environment for
statistical computing. Vienna, Austria: R Foundation for Statistical
Computing. Retrieved from http://www.R-project.org
Reyes, L., Aristegui, R., Krause, M., Strasser, K., Tomicic, A., Valdés, N.,
. . . Ben-Dov, P. (2008). Language and therapeutic change: A speech acts
analysis. Psychotherapy Research, 18, 355–362. http://dx.doi.org/10
.1080/10503300701576360
Riley, M. A., Richardson, M. J., Shockley, K., & Ramenzoni, V. C. (2011).
Interpersonal synergies. Frontiers in Psychology, 2, 38.
Rogers, C. R. (1957). The necessary and sufficient conditions of therapeu-
tic personality change. Journal of Consulting Psychology, 21, 95–103.
http://dx.doi.org/10.1037/h0045357
Safran, J. D., & Muran, J. C. (2000). Resolving therapeutic alliance
ruptures: Diversity and integration. Journal of Clinical Psychology, 56,
233–243. http://dx.doi.org/10.1002/(SICI)1097-4679(200002)56:
2233::AID-JCLP93.0.CO;2-3
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
521
LANGUAGE STYLE MATCHING IN PSYCHOTHERAPY
Shedler, J., & Westen, D. (1998). Refining the measurement of axis II: A
Q-sort procedure for assessing personality pathology. Assessment, 5,
333–353. http://dx.doi.org/10.1177/107319119800500403
Shedler, J., & Westen, D. (2007). The Shedler-Westen Assessment Proce-
dure (SWAP): Making personality diagnosis clinically meaningful.
Journal of Personality Assessment, 89, 41–55. http://dx.doi.org/10.1080/
00223890701357092
Shephard, C., Giles, H., & LePoire, B. (2001). Communication accommo-
dation theory. In W. P. Robinson & H. Giles (Eds.), The new handbook
of language and social psychology (pp. 33–56). Chichester, UK: Wiley.
Snyder, J., & Silberschatz, G. (2017). The patient’s experience of attun-
ement and responsiveness scale. Psychotherapy Research, 27, 608 – 619.
http://dx.doi.org/10.1080/10503307.2016.1147658
Tausczik, Y. R., & Pennebaker, J. W. (2010). The psychological meaning
of words: LIWC and computerized text analysis methods. Journal of
Language and Social Psychology, 29, 24 –54. http://dx.doi.org/10.1177/
0261927X09351676
Tichenor, V., & Hill, C. E. (1989). A comparison of six measures of
working alliance. Psychotherapy: Theory, Research, Practice, Training,
26, 195–199. http://dx.doi.org/10.1037/h0085419
Waldron, S., Moscovitz, S., Lundin, J., Helm, F. L., Jemerin, J., & Gorman,
B. (2011). Evaluating the outcomes of psychotherapies: The Personality
Health Index. Psychoanalytic Psychology, 28, 363–388. http://dx.doi
.org/10.1037/a0024559
Zilcha-Mano, S., & Errázuriz, P. (2015). One size does not fit all: Exam-
ining heterogeneity and identifying moderators of the alliance-outcome
association. Journal of Counseling Psychology, 62, 579 –591. http://dx
.doi.org/10.1037/cou0000103
Received April 14, 2019
Revision received February 14, 2020
Accepted February 19, 2020
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
522 AAFJES-VAN DOORN, PORCERELLI, AND MU
¨LLER-FROMMEYER
... Studies have shown that AI systems can accurately predict patient-reported alliance and symptom ratings based on behavioral markers extracted from video recordings of psychotherapy sessions. Verbal features (Aafjes-van Doorn et al., 2020;Vail et al., 2022;Goldberg et al., 2020), body and facial movements (Cohen et al., 2021), have all been considered as directly associated with self-reported measures of the working alliance. ...
Preprint
As mental health issues rise among college students, there is an increasing interest and demand in leveraging Multimodal Language Models (MLLM) to enhance mental support services, yet integrating them into psychotherapy remains theoretical or non-user-centered. This study investigated the opportunities and challenges of using MLLMs within the campus psychotherapy alliance in China. Through three studies involving both therapists and student clients, we argue that the ideal role for MLLMs at this stage is as an auxiliary tool to human therapists. Users widely expect features such as triage matching and real-time emotion recognition. At the same time, for independent therapy by MLLM, concerns about capabilities and privacy ethics remain prominent, despite high demands for personalized avatars and non-verbal communication. Our findings further indicate that users' sense of social identity and perceived relative status of MLLMs significantly influence their acceptance. This study provides insights for future intelligent campus mental healthcare.
... Recent proof-of-concept studies have sought to measure the working alliance using automated computational approaches of nonverbal communication analysis between patient and clinician. Direct associations with self-reported working alliance measurements were found for a set of monadic and dyadic linguistic features, including patients' language content, pronoun use, non-fluency, and reciprocal linguistic style matching and reciprocal linguistic content matching between clinician and patient (Aafjes-van Doorn et al., 2020;Goldberg et al., 2020;Martinez et al., 2019;Ryu et al., 2021;Ryu et al., 2023;Vail et al., 2022); patients' head gesture features (nods, shakes) and turn-taking behaviors (turn length, wait time) (Vail et al., 2021); and patient-clinician synchrony of facial movements and expressions (Yokotani et al., 2020), and body movement (Cohen et al., 2021). Although these studies of different patient, clinician, or dyadic behavioral markers provide important starting points, the best achieved performances were suboptimal, see Table I. ...
Article
Objective.Session-by-session tracking of the working alliance enables clinicians to detect alliance deterioration and intervene accordingly, which has shown to improve treatment outcome, and reduce dropout. Despite this, regular use of alliance self-report measures has failed to gain widespread implementation. We aimed to develop an automated alliance prediction using behavioral features obtained from video-recorded therapy sessions. Method: A naturalistic dataset of session recordings with patient-ratings of working alliance was available for 252 in-person and teletherapy sessions from 47 patients treated by 10 clinicians. Text and audio-based features were extracted from all 252 sessions. Additional video-based feature extraction was possible for a subsample of 80 sessions. We developed a modeling pipeline for audio and text and for audio, text and video to train machine learning regression models that fuse multimodal features. Results: Best results were achieved with a Gradient Boosting architecture, when using audio, text, and video features extracted from the patient (ICC = 0.66, Pearson r = 0.70, MAE = 0.33). Conclusion: Automated alliance prediction from video-recorded therapy sessions is feasible with high accuracy. A datadriven multimodal approach to feature extraction and selection enables powerful models, outperforming previous work.
... From a psychoanalytic treatment perspective, these models could be used to assess the degree of the referential process (MCT; Bucci, 1997), Reflective Functioning (RF; Fonagy et al., 1998) or expand the work from cognitive distortions to defense mechanisms (Bowins, 2004;Cramer, 2015). From a common factors perspective, this would entail the assessment of alliance (e.g., Aafjes-van Doorn et al., 2020b) or therapeutic self-disclosure (e.g., Shapira & Alfi-Yogev, 2024). ...
Article
Full-text available
Background Based on previous work on the identification of cognitive distortions in language data, we aim to predict patient depression symptoms by identifying cognitive distortions in their psychotherapy transcripts. Further, using explainable AI we want to create a machine learning model that can be used to identify important distortions, allowing us to assess the most important distortions across the whole dataset and for individual patients. Methods We identified 14 cognitive distortions by modifying lists of representative n-grams created by Bathina et al. (Nat Hum Behav 5(4):458–466, 2021, https://doi.org/10.1038/s41562-021-01050-7). Based on these n-grams, the relative frequency of the distortions was calculated across 552 transcripts of 124 patients and employed to predict patient depression symptoms. Further, we joined all distortions in an exploratory explainable AI model, combining various machine learning algorithms in a nested cross-validation framework for the prediction of depression symptoms. Results Depression severity and occurrence were predicted by the distortions personalizing (r=.13), dichotomous thinking (r=.11), and overgeneralizing (r=.10), while occurrence was additionally predicted by mental filtering. The joined machine learning model achieved a moderate performance of r=.29. Using explainable AI, we identified the distortions with the highest feature impact in the combined model (i.e., dichotomous reasoning, minimization, personalizing, mindreading, and mental filtering) and were able to explain the prediction for individual transcripts. Conclusions This approach illustrates how language-based measures can identify relevant processes that predict depression symptoms. This may improve our understanding of the effects of cognitive distortions and may be utilized to create feedback to therapists. Limitations prevail due to small effect sizes and the exploratory approach of this study.
Article
Full-text available
Background: Interpersonal synchrony entails the coordination of verbal and nonverbal features underlying communicative patterns. However, the strength of the association between the synchronisation of nonverbal features (movements and vocal tones) and therapeutic alliance and outcome remains unclear. Objectives: This meta-analysis aimed to investigate the strength of (1) the association between nonverbal synchrony and perception of alliance reported by the patients, (2) the association between nonverbal synchrony and therapeutic outcome, and (3) the moderating role of the type of psychotherapeutic approach. Methods: This meta-analysis included 11 studies. Inclusion criteria were studies that utilised a quantitative measure of nonver-bal synchrony in association with measures of therapeutic alliance and outcome. Results: Random effects model indicated that nonverbal synchrony was significantly associated with the alliance perceived by the patient (r = 0.19; 95% CI: 0.02-0.35; z = 2.18, p = 0.02); however, it was not significantly associated with the therapeutic outcome (r = 0.22; 95% CI: −0.04 to 0.47; z = 1.65, p = 0.09). No moderating effects were observed for the type of therapeutic approach. Conclusions: These findings support the view that nonverbal synchrony is a central aspect of psychotherapy and highlight the possible interdependence between nonverbal synchrony and therapeutic alliance.
Article
Purpose The purpose of this paper is to highlight current utilizations of advanced technology within the digital mental health platform Keel Mind. Design/methodology/approach Keel Mind, based out of Ontario, Canada is a therapeutic delivery platform that dually functions to train clinicians and deliver virtual therapy to clients. Findings From personalizing client and therapist matching to using natural language processing (NLP) and artificial intelligence (AI)-driven features Keel Mind supports clinician decision-making to meet the growing mental health needs of the population. Keel Mind functions to provide accessible service, efficiently train high-quality clinicians and enhance clinical outcomes for clients. In a preliminary study within post-secondary students from Ontario, Canada, 90.1% of clients reported liking the platform, and 83.4% reported that they experienced positive mental health outcomes as a result of using Keel Mind. Practical implications Keel Mind is a leader within the rapidly developing field of telepsychology and aims to promote positive mental health outcomes worldwide. Originality/value This digital showcase is the first published piece of work highlighting the technological capabilities of Keel Mind as a digital therapeutic platform. The intention of this work is to highlight current uses of digital technology within the field of therapeutic practice and to promote technological growth within the field.
Article
Objective: In an attempt to operationalize an implicit aspect of the therapeutic relationship, this study assesses reciprocal linguistic style entrainment (rLSM) between the patient and therapist. rLSM is defined as the dynamic adjustment of function word usage to synchronize or to be in rhythm with another person as they change over time. Method: In this exploratory study, levels of rLSM per talk turn were analyzed for 540 sessions of 27 long-term psychoanalytic treatments in relation to treatment outcomes. Results: Within sessions, rLSM appeared to decrease by the end of sessions and followed a negative linear trajectory, βlinear = −0.0002, SE < .001, t = −13.04, p < .001. Between sessions, rLSM showed significant variability such that neither a linear, nor a quadratic, nor a cubic trend line fit the session-by-session change over treatment. On average, therapist talk turns had significantly lower rLSM than patient talk turns, while accounting for the nested nature of the data using multilevel models βSpeakerT = −0.033, SE = 0.009, t = −3.65, p < .001. Levels of rLSM did not relate to treatment outcome. Conclusion: Most of the rLSM variance was at the within-patient and within-session level. rLSM was not an indicator of psychoanalytic treatment outcomes.
Article
Full-text available
The goal of the study was to develop a computer-aided system that is able to identify key moments in transcripts from psychoanalytic sessions and to provide an adequate theory of change. The term key moment refers to 1 or more sessions of a treatment or to segments of a session that are seen as clinically important and often considered to be a turning point or breakthrough and that mirror points of insight as they occur in the course of the psychotherapeutic process. It will be shown that patterns built of combinations of the content analysis variables “emotion tone” and “abstraction” allow for describing therapeutic cycles including key moments. The method is shown successfully for a single case and for a sample of improved and not improved patients.
Article
Full-text available
A rupture is a deterioration in the therapeutic alliance, manifested by a disagreement between the patient and therapist on treatment goals, a lack of collaboration on therapeutic tasks, or a strain in their emotional bond. We present the most frequently used measures of alliance ruptures and clinical examples to illustrate their repair. To examine the relation of rupture repairs to outcome, and the impact of rupture resolution training on outcome, we conducted two meta-analyses. In the first meta-analysis, we examined 11 studies (1,314 patients) that examined the relation between rupture repair episodes and patient treatment outcomes. Results yielded an effect size of r .29, d .62, 95% confidence interval [.10, .47], p .003, indicating a moderate relation between rupture resolution and positive patient outcome. Our second meta-analysis examined the impact of rupture resolution training or supervision on patient outcome. We examined 6 studies (276 trainees/supervisees) that compared the outcomes of trainees who received rupture resolution training with a comparison group. Results did not find a significant relation, r-.11, d .22, 95% confidence interval [09, .30], p .28. Moderator analyses indicated that the relation between training and outcome was stronger when the sample included fewer patients with personality disorders, when the training was more closely aligned with cognitive behavioral therapy than psychodynamic therapy, and when the treatment was brief. The article concludes with limitations of the research, diversity considerations, and research-informed therapeutic practices for repairing ruptures in ways that contribute to good treatment outcome.
Article
Full-text available
While research suggests that the therapeutic alliance is important in predicting outcomes of psychotherapy, relatively little is known about the development of the alliance or the moment-to-moment components of the relationship and how they combine to create an alliance, which may represent a serious limitation in existing methods of measurement. Language style matching (LSM), or the degree to which unconscious aspects of an interactional partner’s language mimic that of the other partner, is a promising, unobtrusive measure of interaction quality that could provide novel insight into the therapist–client alliance. In this article, we present a theoretical argument regarding the trajectory of therapist–client LSM across therapy sessions, as well as potential precursors and consequences of LSM. We then report on a pilot test of our hypotheses that examined how LSM, clients’ relational histories, and clients’ symptoms were associated within a therapeutic context. Using a small sample of substance dependent mothers (N = 7, 100% Caucasian women) enrolled in a randomized controlled trial of psychodynamic psychotherapy lasting 12 sessions, we examined client and therapist LSM across 4 of the 12 sessions. We found that, on average, LSM decreases over the course of treatment. Furthermore, greater client interpersonal problems prospectively predict lower early LSM in therapist–client dyads, which in turn predicts greater posttreatment psychiatric distress. Results generate questions for future research and support further investigations of LSM as one index of the quality of interactions between therapist and client.
Article
Full-text available
Article
Full-text available
This study examined the association between language style matching (LSM), subjective perceptions of interaction quality (perceived responsiveness and affect), and partner behavior in two communication contexts: conflict and social support. Romantic couples (N = 91) engaged in a video-recorded discussion of either a relationship stressor or one partner’s personal stressor (a social support discussion). LSM was associated with unique outcomes in each communication setting. Higher LSM was associated with lower subjective perceptions of responsiveness and less positive emotion for partners discussing relationship stressors but more positive emotion for partners in social support discussions. Furthermore, higher LSM was associated with more critical and negative interpersonal behavior and less responsive and caring behavior during discussions of relationship stressors but was unrelated to behavior in support discussions. Findings suggest that LSM does not uniformly signal interpersonal rapport (as often assumed) and may instead amplify the positive or negative tone of an interaction.
Article
Full-text available
During psychotherapy, patient and therapist tend to spontaneously synchronize their vocal pitch, bodily movements, and even their physiological processes. In the present article, we consider how this pervasive phenomenon may shed new light on the therapeutic relationship– or alliance– and its role within psychotherapy. We first review clinical research on the alliance and the multidisciplinary area of interpersonal synchrony. We then integrate both literatures in the Interpersonal Synchrony (In-Sync) model of psychotherapy. According to the model, the alliance is grounded in the coupling of patient and therapist’s brains. Because brains do not interact directly, movement synchrony may help to establish inter-brain coupling. Inter-brain coupling may provide patient and therapist with access to another’s internal states, which facilitates common understanding and emotional sharing. Over time, these interpersonal exchanges may improve patients’ emotion-regulatory capacities and related therapeutic outcomes. We discuss the empirical assessment of interpersonal synchrony and review preliminary research on synchrony in psychotherapy. Finally, we summarize our main conclusions and consider the broader implications of viewing psychotherapy as the product of two interacting brains.
Chapter
Dyadic coordination is a fundamental feature of all human interaction. From the early work on motor mimicry, scholars have devoted tremendous energy to discover patterns of behavioral adaptation and the impact those patterns have on individual and relational outcomes. With growing technology, researchers have tools capable of analyzing certain elements of human communication quickly and efficiently. Specifically, automated textual analysis software programs such as Linguistic Inquiry and Word Count (LIWC) (Pennebaker, Booth, Boyd, & Francis, 2015; Pennebaker, Booth, & Francis, 2007) can be used to create indices of language behavior. This profile focuses on a measure of linguistic coordination called Language Style Matching (LSM) (Ireland & Pennebaker, 2010), which indexes the degree to which two or more language outputs converge. LSM is said to be an indicator of psychological synchrony or the degree to which two language users (writers or speakers) are thinking in similar ways.In general, LSM might suggest that conversational partners are listening to one another on a fundamental level; it would make sense that people who are engaged with each other on a topic would speak about the topic in the same way.