Abstract

Developing empathy toward a user is a vital part of user-centered design. Several approaches developed in psychology, sociology, and neuroscience have been adopted in measuring empathy in design. However, these approaches have limitations, such as subjective bias, time consumption, the need for specialized equipment, and expensive laboratories. To address these shortcomings, there is a need for a quick, easy-to-implement, and automatic measure of empathy in design. We explore empathy measurement using transcripts from a user interview. More specifically, we explored whether language style matching (LSM), a measure of unconscious verbal mimicry, can be an indicator of empathic mental processes. We further investigated its relationship with the designer's empathic understanding of the user and the expressed emotion similarity between the designer and the user. The results show that verbal mimicry exists between the designer and the user. However, this mimicry, as detected with LSM, was not correlated with empathic understanding. Instead, we found that LSM has a significant correlation with the similarity between the designer's and the user's expressed emotions during the interview. Verbal mimicry using LSM shows the potential to measure the designer's empathic understanding of the user, which is both cognitive and affective. Further research should explore other measures of empathic understanding.

1 Introduction

1.1 Empathy.

Empathy toward the user has been emphasized as a key component in successful product design. Although empathy is used throughout the design process, it is most evident in the user understanding phases [1]. Empathy development in the early design stages is important because user involvement is often more limited in the later stages.

Heylighen and Dong [2] emphasized the importance of adopting a comprehensive perspective when addressing empathy in design research. To achieve this, it is essential to better understand how the development of empathy influences design decisions and actions. It is useful to evaluate empathy development and its various dimensions. Accordingly, there has been an increase in empathy research in engineering design. Empathy frameworks have been introduced in design, borrowing the concepts of empathy from psychology, philosophy, neuroscience, and sociology [3,4]. To implement empathy in design, Kouprie and Visser [5] conceptualized empathy as an intentional action of stepping into others' lives as the designer, using four steps: discovery, immersion, connection, and detachment. In engineering education, Hess and Fila [6] conceptualized empathy using two major axes: cognitive processes—affective experience and self-oriented—other-oriented. Dong et al. [4], on the other hand, presented three domains (affection and cognition, subject-oriented and object-oriented, and attitude and technique) to conceptualize empathy. Beyond just empathy between the designer and the user, Smeenk et al. [7] also presented an empathic transfer framework for designers.

Surma-Aho and Hölttä-Otto [8] reviewed past empathy research in design to conceptualize and identify means to measure empathy in design. Five major design-related concepts of empathy were presented, grouping them based on the area in which the knowledge is constructed. They found a general agreement in the literature that the goal of empathic design is to gain empathic user understanding. Empathic design action (the user-centered activities designers carry out in the design process) and empathic design research (the methods implemented in understanding the user) are external constructs related to empathic understanding, while the empathic mental processes (empathic internal representation of the user's situation) and the empathic orientation (conscious intention to understand the user) are internal constructs that lead to empathic understanding. Figure 1 shows the five conceptualizations of empathy and their relationships. Studies already show that empathic understanding leads to better empathic design actions [9] and empathic design research can improve empathic understanding [10]. However, the relationship between empathic understanding and the internal constructs consisting of empathic mental processes and empathic orientation is still unclear and a current research area in design. In this study, we focus on empathic mental processes, which are one of the internal constructs of empathy.

Fig. 1
Current state of the relationship between the constructs of empathy related to design (white text). The potential approaches to measure empathy constructs are in pink text.
Fig. 1
Current state of the relationship between the constructs of empathy related to design (white text). The potential approaches to measure empathy constructs are in pink text.
Close modal

1.2 Measuring Empathy in Design Research.

Researchers have implemented several qualitative and quantitative methods from other disciplines to measure empathy in design and its impact on design outcomes. Trait measures like interpersonal reactivity index (IRI) [1113] and empathy quotient [14,15] have been used to measure empathic orientation. For example, Alzayed et al. [11] used IRI to study the factors that contribute to the building of empathy in engineering design education. Surma-Aho et al. [1] also implemented the IRI measure to understand the relationship between disposition about empathic ability and innovation. In addition, Sleeswijk Visser and Kouprie [15] implemented their approach to empathy in design using the empathy quotient to measure empathy. This measure of empathy is described as evaluating self-belief empathy [16]. However, this approach may be influenced by subjective bias, and reliance on respondents' self-assessment of their empathic ability could be considered a potential limitation.

Empathic understanding has also been measured using performance-based methods such as empathic accuracy (EA) [17,18] and quick empathic accuracy (QEA) [19,20]. Chang-Arana et al. [18] used the empathic accuracy method [17] in interview sessions between designers and users to explore the relationship between empathic tasks and physiological measures. They showed that designers were able to accurately identify approximately 50% of the users' reported content. However, there was no significant relationship between empathic accuracy and physiological measures.

Furthermore, Physiological measures such as electromyography (EMG) [18] and heart rate synchrony [21] have also been used to study empathy in design research. These automatic measures of empathy often require specialized equipment and advanced laboratory setups, making them costly to implement on an industrial scale in design. Additionally, recruiting participants for the user studies can be both challenging and expensive.

Researchers have also developed design-specific metrics and measures for empathy in design research. Hess et al. [22] designed an engineering-specific measure of empathy consisting of four components building up from Hess and Fila's [6] empathy model framework. Also, Drouet et al. [23] developed a scale to measure empathy in design, specifically service design, building on frameworks for empathy in design.

However, in design, there is still a need for a measure that will solve the subjective bias of the trait measurement of empathy. This measure should be automatic, reduce the time consumption of the current performance-based measures, and reduce the expensive laboratory and specialized equipment requirements of the physiological measures.

In the most recent work to achieve this, Salmi et al. [20] used video recordings to analyze smiling and frowning mimicry as a measure of empathic mental processes in design. Nguyen et al. [24] used natural language processing to automatically predict the empathic accuracy score and reduce the labor and time involved in the manual annotation process. They found fine-tuning the model before prediction to be the best in performance. Also, Fabunmi et al. [25] compared approaches to predict empathic accuracy and emphasized the potential shortcomings of implementing Large Language Models in empathic design and how they can be mitigated. They presented an empathic accuracy dataset for benchmarking and model choice to encourage data-driven empathic design. The use of natural language has been well explored outside of design in other disciplines, such as linguistics [26,27] and psychotherapy [28]. Methods like emotion detection [28], empathy detection [26,27], similarity in the use of specific groups of words [29,30], and language coordination [31,32] have been used to measure and study empathy.

Building on previous work, we explore language style matching (LSM) [33], a novel verbal mimicry approach to measuring empathic mental processes as operationalized by Surma-Aho and Hölttä-Otto [8]. LSM is a phenomenon in dyadic conversations based on the discovery that people's matching in their language styles is an indicator of harmony in their psychological worlds [34]. Previous studies have shown that similarity in using function words (such as pronouns, articles, prepositions, conjunctions, and negations) indicates the state of the relationship between people [33,34]. In the growing body of research, LSM is used as an approach to interpersonal coordination of language styles [35]. LSM is described as taking place when dyads approach a shared psychological state [36]. This method uses the degree to which the dyads coordinate their use of function words regardless of the context of the word. Ireland and Pennebaker [33] define function words as “placeholders,” of which the meaning is unique to the partners in communication, and understanding the meaning requires shared knowledge between the partners in the conversation.

Language styles and patterns have been used to study gender differences among engineering teams when performing design tasks [37]. Disciplines such as therapy and psychology have also used verbal mimicry measures like LSM to study social rapport [3840] and empathy [31,32]. Furthermore, previous studies have investigated language style similarity over time between dyads to study social interactions such as friendship status [41] and perceived emotional support [42]. Theories in psycholinguistics like the communication accommodation theory [43] have related convergence in communication behaviors with positive outcomes.

In this article, we investigated whether unconscious verbal mimicry, detected as LSM, can be used as a measure of empathic mental processes and is related to empathic understanding. Previous studies have examined mutual understanding between dyads using empathic accuracy [44,45]. LSM has also been used to study the same, suggesting that there could be a relationship between LSM and empathic accuracy. Additionally, LSM and emotion similarity have been used in the study and measure of empathy, indicating a possible relationship between them [32,46]. LSM, over a period of time, has also been used in studying social interaction and outcomes [40,41]. We aim to answer the following research questions (RQs) by evaluating the hypotheses (H) attached to them:

  • RQ1: Is there verbal mimicry between the designer and the user?

    • H1: Language style matching shows that verbal mimicry exists between the designer and the user.

  • RQ2: Is there a relationship between verbal mimicry and empathic understanding?

    • H2a: There is a positive correlation between language style matching and the designer's empathic accuracy.

    • H2b: There is a positive correlation between language style matching and the designer's emotional tone accuracy (ETA).

  • RQ3: Is there a relationship between verbal mimicry and similarity in expressed emotions?

    • H3: There is a positive correlation between language style matching and similarity in expressed emotions.

  • RQ4: Does the verbal mimicry between the designer and user increase as the interview progresses?

    • H4: There will be increased similarity in the language style between the designer and user as the interview progresses.

2 Materials and Methods

2.1 Data.

The data used are anonymized transcripts from the semistructured user interviews of Salmi et al.'s [20] previous work. Interviews were professionally transcribed and edited to remove all identifying names and any specific location, such as a suburb or a small town, to protect the privacy of the participants. The interviews aimed to understand users in difficult driving scenarios with limited visibility due to weather or surroundings. The same interview guide was followed in all interviews. The conversation was about the user's driving experience and questions on specific driving situations, such as driving in city centers and crossroads and challenging scenarios like rain and low visibility.

The empathic and emotional tone accuracy scores for each interview were obtained from the previous study. Three designers interviewed 46 users in the study. The details of the users can be found in the previous study [20]. In this current study, 42 user interview transcripts were used (4 of the transcripts were unavailable).

2.2 Data Preprocessing.

The transcripts were cleaned before analysis. Data preprocessing included removing information such as the transcript title, length of the interview recording, and transcription notes to ensure that the data only contained conversations between the designer and the user. The cleaned transcript was then split into two parts to distinguish between the user's and the designer's parts in the conversation. The transcript was analyzed as transcribed from the user interview without grammatical correction, sentence completion, or additional preprocessing.

The data analysis process carried out in this study is described in Fig. 2. Each method implemented is explained in the next section.

Fig. 2
The data analysis process used in the current study
Fig. 2
The data analysis process used in the current study
Close modal

2.3 Methods

2.3.1 Language Style Matching.

In answering the research questions, the Linguistic Inquiry Word Count (LIWC) software formula [47] was used to calculate the LSM score for each user interview transcript. The formula for similarity in language style presented by Ireland and Pennebaker [33] in Eq. (1) was used for eight function words (preposition, article, auxiliary verbs, adverbs, conjunction, personal pronouns, impersonal pronouns, and negate), then averaged using Eq. (2). The value 0.0001 is included to prevent a division by zero:
(1)
(2)

For RQ1–3, LSM was calculated for the entire transcript. While for RQ4, each transcript was divided into three interview sessions, and LSM was calculated for each session individually. The three sections include Phase 1: the introductory section, which focuses on getting to know the user; Phase 2: the context-setting session, centered on discussions about the user's driving experience; and Phase 3: the session discussing specific driving scenarios and the need for a driving assistant device.

2.3.2 Empathic Understanding.

The empathic understanding measure is a performance-based measure of empathy. It is a measure of understanding the mental contents of the target. It can be measured as the accuracy in inferring the thoughts and/or feelings of a perceived target (user).

The EA and ETA scores used in this work are from Salmi et al. [20]. They were collected via a process where the user replayed the videos and carried out the empathic accuracy task of pausing and writing out their thoughts and/or feelings. This process was also carried out with the designers but with the task of inferring the thoughts or feelings of the user at the same timestamp where the user paused the video. Implementing the EA measure in design shifts quantifying empathy from the trait measures to assess it as an objective task, focusing on how accurately the designer identifies the user's thoughts and feelings.

The designers implemented the QEA approach [19], a shorter version of the empathic accuracy [17] measure to infer the users' thoughts, feelings, and emotional tones [positive (+), neutral (0), and negative (−)]. A sample of the user's entry is described in Table 1. The similarity between the designer's guess and the user's reported thoughts and feelings was rated by humans with a cross-rater's Cronbach's alpha of 0.823 [20]. The raters scored the similarity as either 0 (i.e., low), 1 (i.e., medium), or 2 (i.e., high), depending on how similar the designer's guess is to the user's thoughts and/or feelings. The designer's EA score was calculated by averaging the three individual raters' scores of the similarity between the users' thoughts and/or feelings for each time stamp, as shown in Table 2. The scores were then averaged to obtain the designer's overall EA score. The accuracy of the emotional tone for each entry was scored with a binary scoring system: true (1) if the user's recorded emotional tone matched the designer's guessed emotional tone, and false (0) otherwise. These binary scores indicated the accuracy of the guessed emotional tone, which were then averaged to evaluate the designer's overall ETA score.

Table 1

A sample of the thoughts and/or feelings entry by the user

TimestampThoughts and/or feelingsEmotional tone (+, 0, −)
2:38I was: “Don’t like driving and tried to find out the reason”0
2:41I was: “Nervous, and I didn’t know how to express”
10:30I was: “I think with smart cars, I would actually like using a car more”+
TimestampThoughts and/or feelingsEmotional tone (+, 0, −)
2:38I was: “Don’t like driving and tried to find out the reason”0
2:41I was: “Nervous, and I didn’t know how to express”
10:30I was: “I think with smart cars, I would actually like using a car more”+
Table 2

Example of the manual annotation for the designer's guess

User's entryDesigner’s inferenceRaterEA score
123
I was: “Don’t like driving and tried to find out the reason”He/she was: “Thinking that luckily he/she has experience of driving in a foreign country but he/she was unsure if it is useful for this interview”0000.00
I was: “I think with smart cars, I would actually like using car more”He/she was: “Imagine themselves in a smart car and what that could do”1211.33
I was: “Nervous, and I didn’t know how to express”He/she was: “Feeling nervous with his/her English and debating if he/she could complete the interview or not”1121.33
User's entryDesigner’s inferenceRaterEA score
123
I was: “Don’t like driving and tried to find out the reason”He/she was: “Thinking that luckily he/she has experience of driving in a foreign country but he/she was unsure if it is useful for this interview”0000.00
I was: “I think with smart cars, I would actually like using car more”He/she was: “Imagine themselves in a smart car and what that could do”1211.33
I was: “Nervous, and I didn’t know how to express”He/she was: “Feeling nervous with his/her English and debating if he/she could complete the interview or not”1121.33

Three variations of the mental inference of the user's thoughts, feelings, and emotional tone were used for this analysis. As explained above, overall empathic and emotional tone accuracy was obtained. Design-related empathic and emotional tone accuracy was obtained by adopting the approach used by Li et al. [48]. Users' thoughts and feelings directly or indirectly related to the design brief, product, or accessory were selected as design-related thoughts and/or feelings (see examples in Table 3). The conversation-related empathic and emotional tone accuracy was obtained by removing the user's thoughts and feelings that were not related to what was being said in the interview (see examples in Table 4).

Table 3

Design-related and nondesign-related user's thoughts and/or feelings example

User's entryCategory
I was: “Definitely, one of the things that bothers me the most on the road is other drivers. The assistant device should definitely provide an improvement on that”Design related
I was: “Think with smart cars, I would actually like using car more”Design related
I was: “Relieved that I found an answer”Nondesign related
I was: “Realized my loophole for the previous answer”Nondesign related
User's entryCategory
I was: “Definitely, one of the things that bothers me the most on the road is other drivers. The assistant device should definitely provide an improvement on that”Design related
I was: “Think with smart cars, I would actually like using car more”Design related
I was: “Relieved that I found an answer”Nondesign related
I was: “Realized my loophole for the previous answer”Nondesign related
Table 4

Conversation-related and nonconversation-related user's thoughts and/or feelings example

Conversation contextUser's entryCategory
A question on what driving is to the user.I was: “Concentrating. Lots of brainwork”Conversation related
A conversation about driving in an intersection, the traffic light turning green, and what information the user would need.I was: “Remembering more stressful driving situations. One situation that gets me particularly angry is traffic lights with a very short “green” time”Conversation related
A conversation on driving in the dark and the user remembering a near miss in November.I was: “Feeling a need to get relief, to get something off my mind, to get some acceptance because I’m feeling that I’m doing this thing wrong sometimes”Nonconversation related
Conversation on trusting the device and at what point the user will trust the device.I was: “Buying time again”Nonconversation related
Conversation contextUser's entryCategory
A question on what driving is to the user.I was: “Concentrating. Lots of brainwork”Conversation related
A conversation about driving in an intersection, the traffic light turning green, and what information the user would need.I was: “Remembering more stressful driving situations. One situation that gets me particularly angry is traffic lights with a very short “green” time”Conversation related
A conversation on driving in the dark and the user remembering a near miss in November.I was: “Feeling a need to get relief, to get something off my mind, to get some acceptance because I’m feeling that I’m doing this thing wrong sometimes”Nonconversation related
Conversation on trusting the device and at what point the user will trust the device.I was: “Buying time again”Nonconversation related

2.3.3 Expressed Emotion Similarity.

Previous studies that used natural language to measure empathy, conceptualized empathy as emotional empathy. This can be described as a shared feeling, a measure of empathic mental processes [8]. The same approach was used in this article, operationalizing empathic mental processes as similarity in feeling between the designer and the user.

The expressed emotions score was evaluated using the rule-based VADER [49] sentiment analyzer pretrained model. This model was chosen due to its uniqueness compared to other rule-based sentiment analyzers that only consider the polarity of emotions (positive, neutral, and negative). The VADER pretrained model takes both the emotion's polarity and the intensity of each word (how much is positive, negative, or neutral).

The positive, negative, and neutral expressed emotion scores were evaluated separately for the designer and user for each user–designer pair of the transcript. Then, the similarity was calculated using cosine similarity (Eq. (3)). The designer and user are represented as a three-dimensional vector each [i.e., designer = (positive expressed emotion score, negative expressed emotion score, neutral expressed emotion score) and user = (positive expressed emotion score, negative expressed emotion score, neutral expressed emotion score)], having the positive, negative, and neutral expressed emotion scores of the conversation as components of one vector. The expressed emotion similarity score is the average of the designer's response and user (user–designer) pair similarity, as shown in Eq. (4):
(3)
(4)
The similarity of individual emotion components was also evaluated for each user–designer pair. The positive, negative, and neutral scores for the user and designer in each user–designer pair were used to calculate the similarity. The same formula used in calculating the similarity in function words was used, as shown in Eq. (5). The individual emotion component similarity is the average of each emotion similarity:
(5)

2.3.4 Statistical Analysis.

The data in this study were both parametric and nonparametric. That is, some passed the normality test while others failed. The Shapiro–Wilk (S–W) test was used to check for normality. The test for statistical difference was carried out using the Wilcoxon signed rank test.

For the RQ1, we calculated the LSM score for all the interview pairs, as explained above. To test the statistical significance of an existing verbal mimicry between the designer and user, we compared the actual verbal mimicry score and a randomly generated user's conversation with the same user's word count for all the transcripts while keeping the designer's words the same. We randomly selected words from an open-source corpus [50] used for word embeddings to replace the user's conversation for all the transcripts. For RQ2, Pearson's correlation was used for the LSM and empathic understanding measure when they both passed the parametric test. Otherwise, Spearman's correlation was used. For RQ3, the Spearman or Pearson correlation was used based on data normality.

3 Results

3.1 Verbal Mimicry Exists Between the Designer and the User.

RQ1 aimed to determine whether there is evidence of verbal mimicry between the designer and the user. As described above, the LSM score was calculated for all the interview pairs. The user's conversation was replaced with random words. However, the designer's conversation was retained to test whether the verbal mimicry measure was statistically significant and not a random occurrence. The random data for each user's conversation was the exact word count as the original users' word count. We expected the actual LSM score to be greater than that of the random data since the random data was not produced in coordination with the designer, even though it has function words.

We investigated the data distribution of the actual LSM score [M = 0.886, standard deviation (SD) = 0.029] and random data LSM score (M = 0.835, SD = 0.040) with the S–W test. Normal distribution was assumed for the actual LSM score (p = 0.184) but not for the random data LSM score (p = 0.035). The Wilcoxon signed rank test score showed that the actual LSM score was statistically significantly higher than the random LSM score (z = 4.953, p < 0.001). This is also shown in Fig. 3.

Fig. 3
Box plot of the actual and random LSM scores
Fig. 3
Box plot of the actual and random LSM scores
Close modal

3.2 Verbal Mimicry Is Not Correlated With the Designer's Empathic Accuracy and Emotional Tone Accuracy Scores.

RQ2 evaluates the relationship between verbal mimicry, measured as LSM, and empathic understanding, measured as the designers' empathic and emotional tone accuracy. As described above, we implemented three variations for empathic and emotional tone accuracy scores.

For the evaluation of the relationship between LSM and the empathic accuracy score (M = 0.231, SD = 0.118), the S–W test assumed nonnormality for the data distribution (p = 0.013). Spearman's correlation was insignificant (r = −0.046, p = 0.771).

The design-related thoughts and/or feelings entries were 16% of the total user's thoughts and/or feelings entries. The small proportion for design-related thoughts and/or feelings is similar to past literature [48,51] when evaluating how well the designer guessed design-related thoughts and /or feelings.

The correlation between LSM and the other empathic accuracy variations was also insignificant, as shown in Table 5.

Table 5

The correlation of LSM and empathic understanding (EA and ETA)

MeanSDrp Value
EA0.2310.118−0.0460.771
Design-related EA0.2580.2930.0920.622
Conversation-related EA0.4440.1300.0500.755
ETA0.4520.173−0.2150.172
Design-related ETA0.1570.207−0.2430.188
Conversation-related ETA0.5550.258−0.2040.195
MeanSDrp Value
EA0.2310.118−0.0460.771
Design-related EA0.2580.2930.0920.622
Conversation-related EA0.4440.1300.0500.755
ETA0.4520.173−0.2150.172
Design-related ETA0.1570.207−0.2430.188
Conversation-related ETA0.5550.258−0.2040.195

We also investigated the relationship between LSM and the ETA score (M = 0.452, SD = 0.173) using Pearson's correlation. The S–W test (p = 0.098) assumed a normal distribution. The correlation was insignificant (r = −0.215, p = 0.172).

The relationship of LSM with the design-related emotional tone accuracy and conversation-related emotional tone accuracy was investigated using Spearman's correlation, and they were both insignificant, as shown in Table 5.

3.3 Verbal Mimicry Is Correlated With Similarity in Expressed Emotion Between the Designer and User.

To answer RQ3, we investigated empathic mental processes, specifically shared feeling, which can be measured as similarity in expressed emotions during the user interview. The similarity in expressed emotions measures how well the designer expressed the same feelings as the user when responding to the user during the interview. The transcripts contained user–designer pairs where at least one expressed emotion component (i.e., positive, negative, or neutral) has a zero (0) score for both the user and the designer. The similarity in expressed emotions was evaluated using two approaches: one that included all user–designer pairs, and another that excluded pairs in which both the designer and user had at least a zero score for a given expressed emotion component (e.g., pairs where both the user and the designer had a score of zero for negative expressed emotion).

For the evaluation of the relationship between LSM and similarity in expressed emotions with eliminated user–designer pair (M = 0.883, SD = 0.077), the data distribution was not normal, and Spearman's correlation was medium and positively significant (r = 0.538, p = <.001), as shown in Fig. 4(a). The relationship between LSM and the similarity in expressed emotions without eliminating any user–designer pair (M = 0.865, SD = 0.045) was also medium and positively significant (r = 0.409, p = 0.007), as shown in Fig. 4(b). Pearson's correlation was used since they were both approximately normally distributed.

Fig. 4
(a) The scatterplot of LSM and expressed emotion similarity scores between the designer and user (elimination of pairs with missing emotion components). (b) The scatterplot of LSM and expressed emotion similarity scores between the designer and user (no elimination). The confidence interval for all figures is 0.95.
Fig. 4
(a) The scatterplot of LSM and expressed emotion similarity scores between the designer and user (elimination of pairs with missing emotion components). (b) The scatterplot of LSM and expressed emotion similarity scores between the designer and user (no elimination). The confidence interval for all figures is 0.95.
Close modal

Further investigation was done on the relationship between LSM and the similarity between the designer's and user's expressed emotion components (positive, negative, and neutral). This analysis utilized the two similarity evaluation approaches described above: with elimination and without elimination. There is a significant correlation between LSM and the similarity in neutral expressed emotion for both approaches; however, the correlation differs when it comes to the similarity in positive and negative expressed emotions, as shown in Table 6. When comparing the coefficient of variation (CV) of the expressed emotion components for both approaches, the approach without elimination has a significantly lower CV, as shown in Table 6. In contrast, for overall expressed emotion similarity, both approaches exhibit comparable CV values (CVwith-elimination = 0.087, CVwithout-elimination = 0.052). The similarity of emotion components without eliminating any user–designer pair is more consistent and reliable, even though there were user–designer pairs with emotion components with zero scores. This shows that the elimination of user–designer pairs negatively impacts the consistency of the data representation, consequently reducing the reliability of eliminating user–designer pairs.

Table 6

The correlation of LSM and similarity in emotion components

MeanSDrp valueCV
Positivee0.3430.1620.1300.4120.472
Negativee0.1250.1010.321b0.0380.808
Neutrale0.8420.1190.542c<0.0010.141
Positive0.4270.0580.275a0.0780.136
Negative0.5790.013−0.0250.8760.022
Neutral0.8190.0470.376b0.0140.057
MeanSDrp valueCV
Positivee0.3430.1620.1300.4120.472
Negativee0.1250.1010.321b0.0380.808
Neutrale0.8420.1190.542c<0.0010.141
Positive0.4270.0580.275a0.0780.136
Negative0.5790.013−0.0250.8760.022
Neutral0.8190.0470.376b0.0140.057

Note: Positivee, negativee, and neutrale—emotion components for the similarity in expressed emotions with elimination; positive, negative, and neutral—emotion components for the similarity in expressed emotions without elimination.

a

p < 0.1.

b

p < 0.05.

c

p < 0.001.

Therefore, LSM has an almost significant positive correlation with positive expressed emotion similarity but no significant correlation with negative expressed emotion similarity.

3.4 Verbal Mimicry Between the Designer and User Increases as the Interview Progresses.

For the investigation of RQ4, we split the transcripts into three major parts as explained in the method section. We investigated the LSM in each phase. We hypothesize that as the designer and user continue in the interview, the LSM should increase, indicating an increase in alignment in their psychological world resulting in empathic understanding. We found that overall, the LSM of the last phase, shown as phase 3 in Fig. 5, was statistically greater than the two earlier phases. The second phase, shown as phase 2 in Fig. 5, was also statistically greater than the first phase, shown as phase 1 in Fig. 5. This is shown in the box plot in Fig. 5.

Fig. 5
Investigation of LSM across three phases during the interview
Fig. 5
Investigation of LSM across three phases during the interview
Close modal

We further investigated the LSM of each section for the transcripts. We found that 67% of the user interviews showed the LSM increasing from Secs. 1–3. However, 2 out of the transcripts increased from Secs. 1 and 2 and remained the same for Secs. 2 and 3. Furthermore, we carried out a post hoc analysis of the interviews with increasing LSM based on the EA and ETA scores. We discovered that 71% of the interviews with the designer's EA greater than the EA's median score (0.218) had an increasing LSM through the sections. 78% of those with the designer's ETA greater than the median score (0.500) also had an increase in LSM through the sections. 73% of the interviews with both the designer's EA and ETA greater than the median had the same LSM trend.

We also investigated the transcripts in the top quartile designer's EA and ETA scores (Q3) (i.e., EA > 0.283 and ETA > 0.600). For the designer's EA score, 67% had an increasing LSM trend. Eighty-three percent of the designer's ETA scores had similar trends. For interviews with both the designer's EA and ETA scores above the third quartile, they all had an increasing LSM throughout the phases.

4 Discussion

We investigated if unconsciously produced verbal mimicry as a measure of empathic mental processes using LSM can be used to measure the empathic ability of a designer during user interviews.

4.1 Validating Language Style Matching as a True Measure of Verbal Mimicry.

Verbal mimicry was found between the designer and the user. The actual LSM score was significantly different and higher than the random data LSM score. This result supports H1, showing that the measure can be implemented even though it pays attention to only function words, which have the least number in the vocabulary but account for over half of the terms often used in conversation [52], without considering the context. This further elucidates the communication alignment taking place between the dyads, i.e., the designer and the user. In contrast, the randomly generated words, though they had function words, were not created in synch with the designer.

4.2 The Unclear Relationship Between Empathic Mental Processes and Empathic Understanding.

The study examined the relationship between verbal mimicry, an operationalization of empathic mental processes (measured as LSM), and empathic understanding, which is measured as accuracy in understanding mental contents and recognizing emotional tones. The result of no correlation between LSM and EA or ETA agrees with previous literature. These studies used a different measure of empathic mental processes to examine its relationship with the designer's empathic understanding of the user. Chang-Arana et al. [18] investigated the relationship between empathic tasks and physiological measures. They did not find a correlation between the designer's and the user's facial mimicry using EMG and the designer's accuracy in understanding mental content. Piispanen [21] also did not find a correlation between heart rate synchrony and the designer's accuracy in both understanding mental content and emotion recognition in the study on the relationship between behavioral and physiological empathy measures. This result was consistent for all the variations of empathic and emotional tone accuracy used in this study.

Our result did not support H2a and H2b. Given that verbal mimicry measured by LSM is affective and continuous like other physiological measures of empathic mental processes. It is still uncertain whether there should be a relationship between empathic mental processes measured as mimicry (verbal or nonverbal) and empathic understanding measured as empathic accuracy, which is a cognitive and discrete measure. Even though, literature from psychology, psychotherapy, and neuroscience have explained empathy (conceptualized as empathic understanding in design) to be a result of mental processes [53,54]. Studies so far do not show evidence to support the relationship between the two conceptualizations of empathy in design research.

However, in psychology, Jospe et al. [55] found an almost significant correlation between heart rate synchrony, using the heartbeat per second synchrony (a measure of the empathic mental processes) and continuous empathic accuracy (a measure of empathic understanding), which is different from Piispanen's findings [21] when the heart rate synchrony concordance index and the discrete empathic accuracy measure was used. Literature has also shown that there is a positive relationship between LSM and empathy in psychology and psychotherapy [31,32]. However, the measure of empathy used was the trait measure, which has been argued to measure a different construct of empathy from behavioral measures such as empathic accuracy [56].

4.3 Language Style Matching and Expressed Emotion Similarity, Measures of Empathic Mental Processes Showed Significant Relationship.

The investigation of the correlation between LSM and the similarity in expressed emotions provided support for H3. This result is consistent with past literature on communication and social support, even though the measures of emotions are different. Rains [42] used the medical outcome survey social support survey [57] to measure emotional support as a construct of social support between health bloggers and their readers. LSM was found to correlate with the reader's perceived emotional support. Malloch and Taylor [58] also found that LSM was a mediator for emotional self-disclosure between posters and readers in online support groups using the LIWC lexicon (affect words, cognitive processing words, and I pronouns).

Further investigation on the correlation between LSM and the similarity in components of emotions shows that LSM has a correlation with the neutral and positive emotion similarity between the user and the designer during the interview. This result is consistent with the findings of Salmi et al. [20]. They found that when there was mimicry between the designer and the user, the designer guessed the user's reported positive emotional tone more accurately compared to the negative emotional tone. Also, Bowen et al. [59], in their study on LSM and social interactions, discovered that LSM has a positive correlation with the use of positive emotional words in social support discussions but the reverse in discussions about relationship stressors. The subject of the user interview could have impacted the use of more positive emotions than negative emotions words. The correlation between LSM and expressed emotions similarity also indicates that LSM is a measure of empathic mental processes.

4.4 The Progressive Increase of Language Style Matching During User Interview Supports Interpersonal Relationships Theory.

The increase in the LSM as the interview progressed supports H4. This shows a positive adjustment in language use during the interview, aimed at attaining greater alignment in language style. This aligns with the expectation of the communication accommodation theory [43] using the convergence strategy [60]. This observation is expected since the designers aim to connect and understand the user empathically. The large percentage of transcripts with the LSM increase is also expected. Previous studies have shown that LSM increase through sessions or time is related to perceived emotional support and social interaction [42], rapport [61], and friendship status [41]. This is similar to the aim of a designer during a user study. The designer aims to build good social interaction and rapport and create an environment friendly enough for the user to share their experiences and feelings.

The convergence in the language style between the designer and the user and its relation to positive outcomes is further buttressed by the large percentage of transcripts with increased LSM contained in the interviews with high designer's EA and /or ETA. In addition, we also found that the analysis of transcripts with and without an increase in LSM is consistent with the designer's experience. The higher the designer's level of experience and expertise, the higher the percentage of the interviews they were involved in, that had increasing LSM.

5 Limitations and Future Work

There are a few limitations that should be acknowledged. This study's findings are limited to the empathic understanding measure used, empathic accuracy. It could be useful to implement other measures of empathic understanding. Also, the sample size used in this study might have impacted the results of the correlational analysis.

The findings presented in this article agree with previous findings in design research on the relationship between empathic mental processes and empathic understanding. However, it is important for future work to investigate which aspects of empathy are essential in design. Although we did not find a relationship between LSM and empathic accuracy, which is a cognitive and discrete measure of empathic understanding, future work should investigate how verbal mimicry, such as LSM, relates to other measures of empathic understanding. Additionally, the impact of verbal mimicry on design outcomes should be investigated. Previous findings have shown LSM as a predictor of mutual understanding [58] and outcomes in therapy [62,63], treatment engagement [64], and negotiation [33].

The observation on the level of expertise and the alignment in LSM needs further investigation due to the small number of designers in the study. Finally, future work could also investigate a combination of unconscious verbal mimicry, such as LSM and other measures of empathy, to evaluate the designer's empathic ability.

6 Conclusion

The objective of this study was to investigate unconscious verbal mimicry as a measure of the designer's empathic ability during user interviews. To achieve this, we examined using language style matching (LSM) to measure empathic mental processes. The result from this study shows that LSM reflects the verbal mimicry between the designer and the user. While the relationship between LSM and empathic accuracy was not found to be significant, LSM showed a correlation with expressed emotion similarity and indicated a trend of increasing harmony throughout the interview. This highlights a process of communication accommodation between the designer and the user. This insight calls for further investigation into the relationship between LSM and other measures of empathic understanding.

Overall, the findings from this study offer valuable insights for design research, particularly in the context of empathic design, presenting LSM as a promising, automatic measure of empathy. This approach has the potential to overcome issues such as bias, time consumption, and the need for specialized equipment and laboratory settings in empathic design studies. This measure of empathy provides a potential metric to evaluate the designer's empathic ability during user interviews. Furthermore, providing a criterion for making design decisions that will contribute to a better design outcome. This measure can also be a potential baseline metric to study how to improve the designer's empathic understanding and interventions for the designer's empathic ability. This work offers design practice a promising method of measuring the accuracy of a designer's user understanding.

Acknowledgment

This study would not be possible without graduate research funding from the University of Melbourne, Australia. We also thank the anonymous reviewers for their valuable feedback.

Conflict of Interest

There are no conflicts of interest.

Data Availability Statement

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.

References

1.
Surma-Aho
,
A.
,
Björklund
,
T.
, and
Hölttä-Otto
,
K.
,
2018
, “
An Analysis of Designer Empathy in the Early Phases of Design Projects
,”
DS 91: Proceedings of NordDesign 2018
,
Linköping, Sweden
,
Aug. 14–17
.
2.
Heylighen
,
A.
, and
Dong
,
A.
,
2019
, “
To Empathise or Not to Empathise? Empathy and Its Limits in Design
,”
Des. Stud.
,
65
, pp.
107
124
.
3.
Chang-Arana
,
Á. M.
,
Surma-Aho
,
A.
,
Hölttä-Otto
,
K.
, and
Sams
,
M.
,
2022
, “
Under the Umbrella: Components of Empathy in Psychology and Design
,”
Des. Sci.
,
8
, p.
e20
.
4.
Dong
,
Y.
,
Dong
,
H.
, and
Yuan
,
S.
,
2017
, “
Empathy in Design: A Historical and Cross-Disciplinary Perspective
,”
Advances in Neuroergonomics and Cognitive Engineering: Proceedings of the AHFE 2017 International Conference on Neuroergonomics and Cognitive Engineering, The Westin Bonaventure Hotel
,
Los Angeles, CA
,
July 17–21
,
Springer
, pp.
295
304
.
5.
Kouprie
,
M.
, and
Visser
,
F. S.
,
2009
, “
A Framework for Empathy in Design: Stepping Into and out of the User's Life
,”
J. Eng. Des.
,
20
(
5
), pp.
437
448
.
6.
Hess
,
J. L.
, and
Fila
,
N. D.
,
2016
, “
The Manifestation of Empathy Within Design: Findings From a Service-Learning Course
,”
CoDesign
,
12
(
1–2
), pp.
93
111
.
7.
Smeenk
,
W.
,
Sturm
,
J.
,
Terken
,
J.
, and
Eggen
,
B.
,
2019
, “
A Systematic Validation of the Empathic Handover Approach Guided by Five Factors That Foster Empathy in Design
,”
CoDesign
,
15
(
4
), pp.
308
328
.
8.
Surma-Aho
,
A.
, and
Hölttä-Otto
,
K.
,
2022
, “
Conceptualization and Operationalization of Empathy in Design Research
,”
Des. Stud.
,
78
, p.
101075
.
9.
Postma
,
C. E.
,
Zwartkruis-Pelgrim
,
E.
,
Daemen
,
E.
, and
Du
,
J.
,
2012
, “
Challenges of Doing Empathic Design: Experiences From Industry
,”
Int. J. Des.
,
6
(
1
), pp.
59
70
. https://www.ijdesign.org/index.php/IJDesign/article/view/1008/403
10.
Raviselvam
,
S.
,
Hwang
,
D.
,
Camburn
,
B.
,
Sng
,
K.
,
Hölttä-Otto
,
K.
, and
Wood
,
K. L.
,
2022
, “
Extreme-User Conditions to Enhance Design Creativity and Empathy-Application Using Visual Impairment
,”
Int. J. Des. Creat. and Innov.
,
10
(
2
), pp.
75
100
.
11.
Alzayed
,
M. A.
,
McComb
,
C.
,
Menold
,
J.
,
Huff
,
J.
, and
Miller
,
S. R.
,
2021
, “
Are you Feeling Me? An Exploration of Empathy Development in Engineering Design Education
,”
ASME J. Mech. Des.
,
143
(
11
), p.
112301
.
12.
Davis
,
M. H.
,
1980
, “
A Multidimensional Approach to Individual Differences in Empathy
,”
JSAS Cat. Sel. Doc. In Psychol.
,
10
(
85
), https://www.uv.es/~friasnav/Davis_1980.pdf, Accessed November 10, 2023.
13.
Surma-Aho
,
A.
,
BjörklundKatja
,
T.
, and
Hölttä-Otto
,
K.
,
2018
, “
Assessing the Development of Empathy and Innovation Attitudes in a Project-based Engineering Design Course
,”
2018 ASEE Annual Conference & Exposition
,
Salt Lake City, UT
,
June 23–27
.
14.
Baron-Cohen
,
S.
, and
Wheelwright
,
S.
,
2004
, “
The Empathy Quotient: an Investigation of Adults With Asperger Syndrome or High Functioning Autism, and Normal Sex Differences
,”
J. Autism Dev. Disord.
,
34
(
2
), pp.
163
175
.
15.
Sleeswijk Visser
,
F.
, and
Kouprie
,
M.
,
2008
, “
Stimulating Empathy in Ideation Workshops
,”
10th Anniversary Conference on Participatory Design
,
Bloomington, IN
,
Oct. 1–4
, pp.
174
177
.
16.
Zhou
,
Q.
,
Valiente
,
C.
, and
Eisenberg
,
N.
,
2003
, “Empathy and its Measurement,”
Positive Psychological Assessment: A Handbook of Models and Measures
,
S. J.
Lopez
, and
C. R.
Snyder
, eds.,
American Psychological Association
, pp.
269
284
.
17.
Ickes
,
W.
,
1993
, “
Empathic Accuracy
,”
J. Pers.
,
61
(
4
), pp.
587
610
.
18.
Chang-Arana
,
ÁM
,
Piispanen
,
M.
,
Himberg
,
T.
,
Surma-Aho
,
A.
,
Alho
,
J.
,
Sams
,
M.
, and
Hölttä-Otto
,
K.
,
2020
, “
Empathic Accuracy in Design: Exploring Design Outcomes Through Empathic Performance and Physiology
,”
Des. Sci.
,
6
, p.
e16
.
19.
Li
,
J.
,
Surma-Aho
,
A.
, and
Hölttä-Otto
,
K.
,
2021
, “
Measuring Designers' Empathic Understanding of Users by a Quick Empathic Accuracy (QEA)
,”
International Design Engineering Technical Conferences and Computers and Information in Engineering Conference
,
Virtual, Online
,
Aug. 17–19
, Vol. 85420, American Society of Mechanical Engineers, p. V006T06A027.
20.
Salmi
,
A.
,
Li
,
J.
, and
Holtta-Otto
,
K.
,
2023
, “
Automatic Facial Expression Analysis as a Measure of User-Designer Empathy
,”
ASME J. Mech. Des.
,
145
(
3
), p.
031403
.
21.
Piispanen
,
M.
,
2020
, “
Behavioral and Physiological Correlates of Empathy and Empathic Accuracy in Human-Centered Design
,” MSc Thesis, Aalto University, Finland
22.
Hess
,
J. L.
,
Fila
,
N. D.
,
Kim
,
E.
, and
Purzer
,
S.
,
2021
, “
Measuring Empathy for Users in Engineering Design
,”
Int. J. Eng. Educ.
,
37
(
3
), pp.
733
743
.
23.
Drouet
,
L.
,
Bongard-Blanchy
,
K.
,
Koenig
,
V.
, and
Lallemand
,
C.
,
2022
, “
Empathy in Design Scale: Development and Initial Insights
,”
CHI Conference on Human Factors in Computing Systems Extended Abstracts
,
New Orleans LA
,
Apr. 29–May 5
, pp.
1
7
.
24.
Nguyen
,
S.
,
Beck
,
D.
, and
Holtta-Otto
,
K.
,
2023
, “
Predicting Empathic Accuracy From User-Designer Interviews
,”
21st Annual Workshop of the Australasian Language Technology Association
,
Melbourne, Australia
,
Nov. 29–Dec. 1
,
Association for Computational Linguistics
, pp.
125
129
.
25.
Fabunmi
,
O.
,
Halgamuge
,
S.
,
Beck
,
D.
, and
Holtta-Otto
,
K.
,
2025
, “
Large Language Models for Predicting Empathic Accuracy Between a Designer and a User
,”
ASME J. Mech. Des.
,
147
(
4
), p.
041401
.
26.
Ghosh
,
S.
,
Maurya
,
D.
,
Ekbal
,
A.
, and
Bhattacharyya
,
P.
,
2022
, “
Team IITP-AINLPML at WASSA 2022: Empathy Detection, Emotion Classification and Personality Detection
,”
12th Workshop on Computational Approaches to Subjectivity, Sentiment & Social Media Analysis
,
Dublin, Ireland
,
May 26
,
Association for Computational Linguistics
, pp.
255
260
.
27.
Lee
,
A.
,
Kummerfeld
,
J. K.
,
An
,
L.
, and
Mihalcea
,
R.
,
2023
, “
Empathy Identification Systems Are Not Accurately Accounting for Context
,”
17th Conference of the European Chapter of the Association for Computational Linguistics
,
Dubrovnik, Croatia
,
May 2–6
,
Association for Computational Linguistics
, pp.
1686
1695
.
28.
Sharma
,
A.
,
Miner
,
A. S.
,
Atkins
,
D. C.
, and
Althoff
,
T.
,
2020
, “
A Computational Approach to Understanding Empathy Expressed in Text-Based Mental Health Support
,” arXiv preprint arXiv:2009.08441.
29.
Gibson
,
J.
,
Malandrakis
,
N.
,
Romero
,
F.
,
Atkins
,
D. C.
, and
Narayanan
,
S. S.
,
2015
, “
Predicting Therapist Empathy in Motivational Interviews Using Language Features Inspired by Psycholinguistic Norms
,”
Interspeech 2015
,
Dresden, Germany
,
Sept. 6–10
, pp.
1947
1951
.
30.
Xiao
,
B.
,
Can
,
D.
,
Georgiou
,
P. G.
,
Atkins
,
D.
, and
Narayanan
,
S. S.
,
2012
, “
Analyzing the Language of Therapist Empathy in Motivational Interview Based Psychotherapy
,”
2012 Asia Pacific Signal and Information Processing Association Annual Summit and Conference
,
Hollywood, CA
,
Dec. 3–6
, IEEE, pp.
1
4
.
31.
Lord
,
S. P.
,
Sheng
,
E.
,
Imel
,
Z. E.
,
Baer
,
J.
, and
Atkins
,
D. C.
,
2015
, “
More Than Reflections: Empathy in Motivational Interviewing Includes Language Style Synchrony Between Therapist and Client
,”
Behav. Ther.
,
46
(
3
), pp.
296
303
.
32.
Pérez-Rosas
,
V.
,
Mihalcea
,
R.
,
Resnicow
,
K.
,
Singh
,
S.
, and
An
,
L.
,
2017
, “
Understanding and Predicting Empathic Behavior in Counseling Therapy
,”
55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
,
Vancouver, Canada
,
July 30–Aug. 4
,
Association for Computational Linguistics
, pp.
1426
1435
.
33.
Ireland
,
M. E.
, and
Pennebaker
,
J. W.
,
2010
, “
Language Style Matching in Writing: Synchrony in Essays, Correspondence, and Poetry
,”
J. Pers. Soc. Psychol.
,
99
(
3
), p.
549
.
34.
Niederhoffer
,
K. G.
, and
Pennebaker
,
J. W.
,
2002
, “
Linguistic Style Matching in Social Interaction
,”
J. Lang. Soc. Psychol.
,
21
(
4
), pp.
337
360
.
35.
Müller-Frommeyer
,
L. C.
, and
Kauffeld
,
S.
,
2022
, “
Capturing the Temporal Dynamics of Language Style Matching in Groups and Teams
,”
Small Group Res.
,
53
(
4
), pp.
503
531
.
36.
Ireland
,
M. E.
,
Slatcher
,
R. B.
,
Eastwick
,
P. W.
,
Scissors
,
L. E.
,
Finkel
,
E. J.
, and
Pennebaker
,
J. W.
,
2011
, “
Language Style Matching Predicts Relationship Initiation and Stability
,”
Psychol. Sci.
,
22
(
1
), pp.
39
44
.
37.
Ferguson
,
S. A.
, and
Olechowski
,
A.
,
2023
, “
Are We Equal Online?: An Investigation of Gendered Language Patterns and Message Engagement on Enterprise Communication Platforms
,”
Proc. ACM Human-Comput. Interact.
,
7
(
CSCW2
), pp.
1
29
.
38.
Agee
,
A.
,
2019
, “
Language Style Matching as a Measure of Librarian/Patron Engagement in Email Reference Transactions
,”
J. Acad. Librariansh.
,
45
(
6
), p.
102069
.
39.
Gonzales
,
A. L.
,
Hancock
,
J. T.
, and
Pennebaker
,
J. W.
,
2010
, “
Language Style Matching as a Predictor of Social Dynamics in Small Groups
,”
Commun. Res.
,
37
(
1
), pp.
3
19
.
40.
Ireland
,
M. E.
, and
Henderson
,
M. D.
,
2014
, “
Language Style Matching, Engagement, and Impasse in Negotiations
,”
Negot. Confl. Manag. Res.
,
7
(
1
), pp.
1
16
.
41.
Kovacs
,
B.
, and
Kleinbaum
,
A. M.
,
2020
, “
Language-Style Similarity and Social Networks
,”
Psychol. Sci.
,
31
(
2
), pp.
202
213
.
42.
Rains
,
S. A.
,
2016
, “
Language Style Matching as a Predictor of Perceived Social Support in Computer-Mediated Interaction Among Individuals Coping With Illness
,”
Commun. Res.
,
43
(
5
), pp.
694
712
.
43.
Giles
,
H.
, and
Ogay
,
T.
,
2007
, “Communication Accommodation Theory,”
Explaining Communication: Contemporary Theories and Exemplars
, 1st ed.,
B. B.
Whaley
, and
W.
Samter
, eds.,
Routledge
,
New York
, pp.
293
310
.
44.
Smith
,
R. E.
, and
Smoll
,
F. L.
,
2007
, “Social-Cognitive Approach to Coaching Behaviors,”
Social Psychology in Sport
,
S.
Jowett
, and
D.
Lavallee
, eds.,
Human Kinetics
,
Champaign, IL
, pp.
75
90
.
45.
Lorimer
,
R.
, and
Jowett
,
S.
,
2013
, “Empathic Understanding and Accuracy in the Coach–Athlete Relationship,”
Routledge Handbook of Sports Coaching
, 1st ed.,
P.
Potrac
,
W.
Gilbert
, and
J.
Denison
, eds.,
Routledge
,
London
, pp.
321
332
.
46.
Chui
,
H.
,
Li
,
X.
, and
Luk
,
S.
,
2022
, “
Therapist Emotion and Emotional Change With Clients: Effects on Perceived Empathy and Session Quality
,”
Psychotherapy
,
59
(
4
), p.
594
.
47.
Pennebaker Conglomerates Inc.
, “
LIWC-22 website
,” https://www.liwc.app/help/lsm
48.
Li
,
J.
,
Surma-Aho
,
A.
,
Chang-Arana
,
ÁM
, and
Hölttä-Otto
,
K.
,
2021
, “
Understanding Customers Across National Cultures: The Influence of National Cultural Differences on Designers' Empathic Accuracy
,”
J. Eng. Des.
,
32
(
10
), pp.
538
558
.
49.
Hutto
,
C.
, and
Gilbert
,
E.
,
2014
, “
Vader: A Parsimonious Rule-Based Model for Sentiment Analysis of Social Media Text
,”
Eighth International AAAI Conference on Web and Social Media
,
Ann Arbor, MI
,
June 1–4
, pp.
216
225
.
50.
N/A
,
2013
, “
Yelp 2013 (Version v1) [Data set] Zenodo
.”
51.
Chang-Arana
,
ÁM
,
Surma-Aho
,
A.
,
Li
,
J.
,
Yang
,
M. C.
, and
Hölttä-Otto
,
K.
,
2020
, “
Reading the User's Mind: Designers Show High Accuracy in Inferring Design-Related Thoughts and Feelings
,”
International Design Engineering Technical Conferences and Computers and Information in Engineering Conference
,
Virtual, Online
,
Aug. 17–19
,
Vol. 83976. American Society of Mechanical Engineers, p. V008T08A029
.
52.
Chung
,
C.
, and
Pennebaker
,
J.
,
2011
, “The Psychological Functions of Function Words,”
Social Communication
,
K.
Fiedler
, ed.,
Psychology Press
,
New York
, pp.
343
359
.
53.
Preston
,
S. D.
, and
de Waal
,
F B. M.
,
2002
, “
Empathy: Its Ultimate and Proximate Bases
,”
Behav. Brain Sci.
,
25
(
1
), pp.
1
20
.
54.
Singer
,
T.
,
Seymour
,
B.
,
O'doherty
,
J.
,
Kaube
,
H.
,
Dolan
,
R. J.
, and
Frith
,
C. D.
,
2004
, “
Empathy for Pain Involves the Affective But Not Sensory Components of Pain
,”
Science
,
303
(
5661
), pp.
1157
1162
.
55.
Jospe
,
K.
,
Genzer
,
S.
,
Klein Selle
,
N.
,
Ong
,
D.
,
Zaki
,
J.
, and
Perry
,
A.
,
2020
, “
The Contribution of Linguistic and Visual Cues to Physiological Synchrony and Empathic Accuracy
,”
Cortex
,
132
, pp.
296
308
.
56.
Zaki
,
J.
,
Bolger
,
N.
, and
Ochsner
,
K.
,
2008
, “
It Takes Two: The Interpersonal Nature of Empathic Accuracy
,”
Psychol. Sci.
,
19
(
4
), pp.
399
404
.
57.
Sherbourne
,
C. D.
, and
Stewart
,
A. L.
,
1991
, “
The MOS Social Support Survey
,”
Soc. Sci. Med.
,
32
(
6
), pp.
705
714
.
58.
Malloch
,
Y. Z.
, and
Taylor
,
L. D.
,
2019
, “
Emotional Self-Disclosure in Online Breast Cancer Support Groups: Examining Theme, Reciprocity, and Linguistic Style Matching
,”
Health Commun.
,
34
(
7
), pp.
764
773
.
59.
Bowen
,
J. D.
,
Winczewski
,
L. A.
, and
Collins
,
N. L.
,
2017
, “
Language Style Matching in Romantic Partners' Conflict and Support Interactions
,”
J. Lang. Soc. Psychol.
,
36
(
3
), pp.
263
286
.
60.
Dragojevic
,
M.
,
Gasiorek
,
J.
, and
Giles
,
H.
,
2016
, “Accommodative Strategies as Core of the Theory,”
Communication Accommodation Theory: Negotiating Personal Relationships and Social Identities Across Contexts
, Vol.
1
,
H.
Giles
, ed.,
Cambridge University Press
,
Cambridge, UK
, pp.
36
59
.
61.
Kory-Westlund
,
J. M.
, and
Breazeal
,
C.
,
2019
, “
A Long-Term Study of Young Children's Rapport, Social Emulation, and Language Learning With a Peer-Like Robot Playmate in Preschool
,”
Front. Rob. AI
,
6
, p.
81
.
62.
Borelli
,
J. L.
,
Sohn
,
L.
,
Wang
,
B. A.
,
Hong
,
K.
,
DeCoste
,
C.
, and
Suchman
,
N. E.
,
2019
, “
Therapist–Client Language Matching: Initial Promise as a Measure of Therapist–Client Relationship Quality
,”
Psychoanal. Psychol.
,
36
(
1
), p.
9
.
63.
Stamatis
,
C. A.
,
Meyerhoff
,
J.
,
Liu
,
T.
,
Hou
,
Z.
,
Sherman
,
G.
,
Curtis
,
B. L.
,
Ungar
,
L. H.
, and
Mohr
,
D. C.
,
2022
, “
The Association of Language Style Matching in Text Messages With Mood and Anxiety Symptoms
,”
Procedia Comput. Sci.
,
206
, pp.
151
161
.
64.
Albano
,
G.
,
Salerno
,
L.
,
Cardi
,
V.
,
Brockmeyer
,
T.
,
Ambwani
,
S.
,
Treasure
,
J.
, and
Lo Coco
,
G.
,
2023
, “
Patient and Mentor Language Style Matching as a Predictor of Working Alliance, Engagement With Treatment as Usual, and Eating Disorders Symptoms Over the Course of an Online Guided Self-Help Intervention for Anorexia Nervosa
,”
Eur. Eat. Disord. Rev.
,
31
(
1
), pp.
135
146
.