Adult attachment representation and computer-assisted language measures: what can we learn from the...

18
Psychotherapy Research 10(4) 390–407, 2000 ©2000 Society for Psychotherapy Research THE RELATIONSHIP AMONG ATTACHMENT REPRESENTATION, EMOTION-ABSTRACTION PATTERNS, AND NARRATIVE STYLE: A COMPUTER-BASED TEXT ANALYSIS OF THE ADULT ATTACHMENT INTERVIEW Anna Buchheim Erhard Mergenthaler Section of Informatics in Psychotherapy, University of Ulm, Germany This article is based on the first author’s dissertation at the University of Ulm. We would like to thank Dr. Dan Pokorny of the University of Ulm, for his statistical advice and valuable discussions on the content. Many thanks also to Dr. Fabienne Becker-Stoll and Dr. Peter Zimmermann, University of Regensburg, for their corating in the Kobak-Attachment Q-Sort-method and valuable support. Correspondence concerning this article should be addressed to Dr. Dipl.- Psych. Anna Buchheim, Department of Psychotherapy and Psychosomatic Medicine, University of Ulm, Am Hochsträß 8, 89081 Ulm, Germany. E-mail address: [email protected]. 390 The main aim of this research was to test the discriminant capacity of computer-based, linguistic text measures which are economically com- pilable (Emotion-Abstraction Patterns, Mergenthaler, 1996; Computerized Referential Activity, Mergenthaler & Bucci, 1999) in differentiating between complex attachment representations in the Adult Attachment Interview (Main & Goldwyn, 1994). The study of N = 40 healthy controls produced the consistent result that, of the two insecure attachment categories, the group “dismissing” (n = 10) showed the lowest means on all text mea- sures, whereas the group “preoccupied” (n = 10) showed the highest means. The mean of the attachment group “secure” (n = 20) lay between these groups. This ranking is consistent with the results of studies which had another research focus, showing deactivation or hyperactivation of attachment-relevant information in both insecure attachment groups, as well as flexibility in the “secure” group. We also discuss whether the coincidence of language markers for emotion and abstraction, as well as Computerized Referential Activity, is adequate to operationalize the con- struct of coherence in narrative style. The present study is based on the theory of attachment developed by John Bowlby as a model of emotional development. According to Bowlby (1969, 1973, 1980), the degree of integration or connection between disparate memory systems affects the adaptivity of an inner working model, which is understood as regulating the emo- tional and cognitive access to attachment-relevant information. Methodologically, the study makes reference to theories and approaches that have proven themselves to be sensitive in the context of process-oriented psycho- therapy research. Using economical computer-assisted measures, these approaches

Transcript of Adult attachment representation and computer-assisted language measures: what can we learn from the...

390 BUCHHEIM AND MERGENTHALERPsychotherapy Research 10(4) 390–407, 2000©2000 Society for Psychotherapy Research

THE RELATIONSHIP AMONG ATTACHMENTREPRESENTATION, EMOTION-ABSTRACTIONPATTERNS, AND NARRATIVE STYLE:A COMPUTER-BASED TEXT ANALYSISOF THE ADULT ATTACHMENT INTERVIEW

Anna BuchheimErhard MergenthalerSection of Informatics in Psychotherapy, University of Ulm, Germany

This article is based on the first author’s dissertation at the University of Ulm. We would like to thankDr. Dan Pokorny of the University of Ulm, for his statistical advice and valuable discussions on thecontent. Many thanks also to Dr. Fabienne Becker-Stoll and Dr. Peter Zimmermann, University ofRegensburg, for their corating in the Kobak-Attachment Q-Sort-method and valuable support.

Correspondence concerning this article should be addressed to Dr. Dipl.- Psych. Anna Buchheim,Department of Psychotherapy and Psychosomatic Medicine, University of Ulm, Am Hochsträß 8, 89081Ulm, Germany. E-mail address: [email protected].

390

The main aim of this research was to test the discriminant capacity ofcomputer-based, linguistic text measures which are economically com-pilable (Emotion-Abstraction Patterns, Mergenthaler, 1996; ComputerizedReferential Activity, Mergenthaler & Bucci, 1999) in differentiating betweencomplex attachment representations in the Adult Attachment Interview(Main & Goldwyn, 1994). The study of N = 40 healthy controls producedthe consistent result that, of the two insecure attachment categories, thegroup “dismissing” (n = 10) showed the lowest means on all text mea-sures, whereas the group “preoccupied” (n = 10) showed the highestmeans. The mean of the attachment group “secure” (n = 20) lay betweenthese groups. This ranking is consistent with the results of studies whichhad another research focus, showing deactivation or hyperactivation ofattachment-relevant information in both insecure attachment groups, aswell as flexibility in the “secure” group. We also discuss whether thecoincidence of language markers for emotion and abstraction, as well asComputerized Referential Activity, is adequate to operationalize the con-struct of coherence in narrative style.

The present study is based on the theory of attachment developed by John Bowlbyas a model of emotional development. According to Bowlby (1969, 1973, 1980), thedegree of integration or connection between disparate memory systems affects theadaptivity of an inner working model, which is understood as regulating the emo-tional and cognitive access to attachment-relevant information.

Methodologically, the study makes reference to theories and approaches thathave proven themselves to be sensitive in the context of process-oriented psycho-therapy research. Using economical computer-assisted measures, these approaches

ATTACHMENT REPRESENTATION AND LINGUISTIC MEASURES 391

operationalize the verbalization of emotions and cognitions, and the interaction be-tween them (Mergenthaler, 1996, 1997), as well as the liveliness of narrative style(Bucci, 1997; Mergenthaler & Bucci, 1999). These three theories and methods, whichare described in the following paragraphs, are intended, in various ways, to mea-sure the narrative competence of a speaker, where competence is described, forexample, as coherent discourse or successful communicative tone. The differing skillsof people in verbalizing, integrating, and reflecting upon feelings provide a com-mon focus for these approaches.

ATTACHMENT AND THE ADULT ATTACHMENT INTERVIEW

The concept of the inner working model (Bowlby, 1969, 1980), using cognitive andschema-theory models, offers an explanatory basis for the influence of early attach-ment-relevant patterns of organization on later development and resultant behavior.Inner working models are described as organizational structures that influence at-tention, memory, action, and, later, language (Main, Kaplan, & Cassidy, 1985; Main,1991; Bretherton, 1999). Due to experiences of interaction, they first occur with at-tachment figures. In everyday exchange, increasingly fixed patterns are established,which function largely automatically, hence unconsciously and procedurally. Theexperiences relating to how the environment and attachment figures function areintegrated into a unitary whole by way of active construction, and eventually, withincreasing linguistic skills, are represented in the form of language. According to thistheory, the relationship quality in the interacting dyad affects developing self-esteemand the ability to regulate affect in a balanced and flexible way.

To identify and operationalize the linguistic organization of attachment-relevantstatements, George and coworkers developed the Adult Attachment Interview (AAI;George, Kaplan, & Main, 1985). It measures the current representations of attach-ment experiences in terms of the past and the present on the basis of narratives. Theinterview focuses essentially on the memory of past attachment relationships andthe access to attachment-relevant thoughts and feelings, as well as the estimatedinfluence of attachment experiences on further development. The questioning tech-nique aims at surprising the unconscious, that is, evaluating the extent to which aspeaker is capable of spontaneously recounting his or her childhood history in acooperative, coherent, and plausible way. The evaluation focuses, through analysisof the literal transcription, on the coherence of the discourse (Grice, 1975), as wellas on the emotional and cognitive integration skills in the narrated attachment expe-riences. Defence-related incongruences in the verbal flow of the interview are con-sidered and the cognitive processing of attachment experiences are classified as at-tachment representations. The inter-individual differences in the assessed attachmentrepresentations form three main categories: “secure” (F); “dismissing” (Ds); “preoc-cupied” (E) (Main & Goldwyn, 1994).1

• Adults with the classification “secure” (F) give open, coherent, and consistentaccounts of their childhood memories, regardless of whether they were posi-tive or negative. They are able to integrate their various experiences into a

1The abbreviations given are commonly used in the attachment literature: F for “free to evaluate”; Dsfor “dismissing”; and E for “enmeshed-preoccupied.” (For transcript examples on this see, e.g., Buchheim,Brisch, & Kächele, 1998.)

392 BUCHHEIM AND MERGENTHALER

self-valuing unitary whole and reflect upon their accounts during their inter-views. These persons can easily access the topics asked about and show afeeling for balance.

• Adults with the classification “dismissing” (Ds) give incoherent, incompleteaccounts of the experiences and often show gaps in memory. As a defenseagainst the surfacing of painful memories, they minimize the importance ofattachment. These subjects insist on normality and inner independence fromothers. Attachment figures are mostly presented positively, without concreteexamples given for this. Possible negative influences are denied.

• Adults with the classification “preoccupied” (E) recount in an excessive, oftennonobjective, and angry way the conflicts experienced with their attachmentfigures. They appear enmeshed and give the impression of having just hadthese experiences yesterday. In doing so, they generalize and analyze theirconflict-ridden comments using apparently exaggerated pseudopsychologicalanalyses without really being able to distance themselves from them. Preoc-cupied subjects characteristically oscillate between positive and negative evalu-ations, without being conscious of this contradiction. In general, their lan-guage seems confused, unclear, and vague.

A reliable and valid method for evaluating the AAI using the classification of Mainand Goldwyn is found in the Q-sort method by Kobak (1993). Kobak’s prototype forthe dimension “secure” is distinguished by attachment-relevant memories and feelingsbeing able to be recalled from memory in a flexible way. The prototype of the dimen-sion “dismissing” is characterized by a deactivation of attachment-relevant information(i.e., attention is diverted from these experiences). The prototype for the dimension“preoccupied” is characterized by a hyperactivation of attachment-relevant informa-tion. In a recent review of the relationship between attachment and affect regulation(Fuendeling, 1998; see also Magai, 1999), discussion of the results concluded that peoplewith the classification “secure” consistently show balanced attention to affective topics,while the “preoccupied” show balanced attention to negative affect and the “dismiss-ing” use defensive strategies against positive as well as negative affect.

In her contribution on metacognitive monitoring, Main (1991) convincingly arguesthat the quality of maternal metacognitive skills is the critical cause of childhood at-tachment security or insecurity. This means that the way one thinks, the representa-tions of one’s own childhood experiences, and ultimately the degree of coherent orga-nization of these thoughts, determines the felt security for one’s child. This has beenshown through the effect of intergenerational transmission of attachment experiences(see, e.g., Main et al., 1985; Fonagy, Steele, & Steele, 1991; Benoit & Parker, 1994).

In sum, individual differences exist in the way people cognitively structure at-tachment relevant information and regulate related feelings. The AAI and the dis-course analytical approach to its evaluation offer a reliable and valid heuristic formeasuring both aspects (see review by Hesse, 1999).

THE THERAPEUTIC CYCLE AND EMOTION-ABSTRACTION PATTERNS

The theory of the Therapeutic Cycle was developed by Mergenthaler in the early 1990sfor identifying key moments in verbatim transcripts of psychotherapeutic treatment dia-logues. This methodological approach was designed “to arrive at a feasible theory andto support it empirically using a hypothesis-generating method with a minimum of

ATTACHMENT REPRESENTATION AND LINGUISTIC MEASURES 393

variables” (Mergenthaler, 1997, p. 7). He based the method on computerized proce-dures, which identified verbal markers of emotions and thought processes in transcripts.

In contrast to computer assisted content analysis (Stone, Dunphy, Smith, & Ogilvie,1966), in which the content of a text itself is the object of measurement, the markerapproach determines operational indicators, that allow us to map the underlyingcognitive and emotional states from the linguistic appearance of a text. “This reduc-tionistic approach intentionally contrasts itself to those methods which cognitivescientists in the area of computer linguistics or artificial intelligence use to modelhuman cognitive skills with a computer” (Mergenthaler & Bucci, 1999, p. 343).

Mergenthaler (1996, 1997) chose the emotional event in a therapy session as acentral phenomenon in psychotherapy of any school (see, e.g., Greenberg & Safran,1987). He assumes that the emotion experienced occurs in varying intensities duringthe psychotherapeutic process, and can bring about change. Thinking about feelingsor things—a skill which is psychotherapeutically relevant—occurs cognitively by wayof reflective abstraction‚ and is directly represented linguistically in abstract wordforms. When we experience something, we might, for example, speak of being ten-der. When we think about the experience, we speak about tenderness. Linguisticabstraction is a process by which new word forms are created by adding a limitednumber of endings like -ness or -ment, thereby expressing thought processes.

The Therapeutic Cycle, described by Mergenthaler, brings the construct of Emo-tional Tone (ET) as linguistic manifestation of the emotional event, and the constructof Abstraction (AB) as linguistic manifestation of cognitive-reflective processes, intoa chronological and quantitatively prototypological order. It is assumed that within acycle, first emotion tone and, chronologically later, abstraction increases and thendecreases again in a sinusoidal pattern. These individual phases are operationalizedas Emotion-Abstraction Patterns (Mergenthaler, 1997, p. 12):

• A. Relaxing: low emotion and low abstraction. The person recounts episodesfrom his or her life which bear no direct relation to the problem. The natureof the account is descriptive. This state could be described either as relax-ation or as disorientation (the person doesn’t know what to talk about).

• B. Reflecting: low emotion and high abstraction. Typically, topics are broughtup which show a high level of reflection without feelings intervening. Psy-chodynamically, this could be interpreted as a kind of defense mechanism(rationalization/intellectualization), or as the expression of successful copingwith a psychological impact (as in moments after insight).

• C. Experiencing: high emotion and low abstraction. Here the person is in astate of emotional experience. Conflict-related topics or topics with intensivelyexperienced positive or negative meaning are brought up and emotionallyverbalized without the person expressing him- or herself in an abstract, in-sightful way.

• D. Connecting: high emotion and high abstraction. The person is in a statewhich can be associated with insight, as emotionally positive as well as nega-tive experiences are verbalized and at the same time reflected on. From apsychoanalytical perspective, it could reflect a state of working through.

The theory of the Therapeutic Cycle with its components of Emotion-Abstrac-tion Patterns was first validated on two psychotherapeutic text samples. In a cross-sectional study, it was shown that successful patients (n = 10) were more frequentlyidentified by the Connecting pattern than unsuccessful patients. In a longitudinal study,

394 BUCHHEIM AND MERGENTHALER

the prototypical course of the Therapeutic Cycle was confirmed with clearly highfrequencies of the Connecting pattern in the middle of the therapy (Mergenthaler,1996). This result is in accordance with the therapeutic goal that patients who ben-efit from psychotherapy should develop the skill of verbalizing and at the same timereflecting on emotional experiences. The meaning of the Connecting pattern as anactive ingredient for therapeutic success has since been shown in other clinical stud-ies (Mergenthaler, 1998; Böhmer, 2000).

REFERENTIAL ACTIVITY

Bucci’s Multiple Code Theory (1997) is based on the cognitive psychological Dual CodeTheory put forward by Paivio (1971, 1978; Bucci, 1985), which postulates that our knowl-edge is represented in long-term memory in two independent, partially connected cod-ing systems (verbal and nonverbal). The verbal system, following Bucci, is the sym-bolic code of language, which allows us to communicate with other people and to applyprinciples of logic. Its contents are discrete entities (words) that refer to other entitiesand that are processed sequentially. The nonverbal system, in contrast to Paivio’s ap-proach, is multichanneled and contains representations of sensory experience, bodilystates, and motor activity. This system works according to analogue principles. Connec-tions between the two systems allow us to name what we experience and to recognizewhat was named. The ability of an individual to express nonverbal experience in wordsis referred to by Bucci as Referential Activity (RA). Bucci assumes that RA varies be-tween individuals and is reflected in a characteristic style of speech which is measur-able using a rating scale (Bucci & Kabasakalian-McKay, 1992) or the ComputerizedReferential Activity (CRA) (Mergenthaler & Bucci, 1999). According to Strunk and White(1972), the typical speech characteristics of RA are the concreteness, the specificity, theclarity, and the imagery of verbal utterances. In general, people with high RA use arichly pictorial narrative style; they describe events elaborately, clearly, and specifically,catching the interest of the listener and producing a lively, vivid picture in their narra-tive. People with low RA, on the other hand, stay more abstract and general or makevague, diffuse statements. Emotional experience is not directly communicated.

In a meta-analysis (n = 23), Samstag (1996) showed that there is a significantrelationship between high RA and the ability of a person to linguistically and emo-tionally integrate experiences. In clinical studies, high RA was related to an increasedoccurrence of transference topics and to better treatment success (Bucci, 1997).

METHOD

PARTICIPANTS

This study is based on verbatim transcripts from AAIs with a nonclinical sample2

(for content-related guidelines, see Main, 1994; for transcription matters see Mergen-

2These AAIs are drawn from a larger sample of an attachment project with parents of very low birthweight infants (principal investigators: Dr. Karl Heinz Brisch, Prof. Dr. H. Kächele, and Prof. Dr. F.Pohlandt, University of Ulm), supported by the German Research Foundation DFG under Br/1574/1–1,1–2,1–3. The AAI assessment was done six months past delivery and the subjects are considered nolonger to be influenced by the traumatic aspects of premature birth.

ATTACHMENT REPRESENTATION AND LINGUISTIC MEASURES 395

thaler & Stinson, 1992). All proper names and other identifying information werereplaced by pseudonyms before these texts became part of the study presented here.The sample, therefore, can be considered anonymous and no longer subject to pri-vacy issues. The categorization into one of the three attachment representations wasdone according to Kobak’s (1993) Q-sort-method. The drawing of the sample wascompleted when 20 subjects were classified as “secure,” 10 as “dismissing,” and 10as “preoccupied.” Of those subjects, 25 were female and 15 male. On average, thesubjects were 33.2 years old (ranging between 23 and 45) at the time of the inter-view. Thirteen subjects had a secondary school education up to year 9 (a German“Hauptschulabschluß”), 13 up to year 10 (a German “Realschulabschluß”), and 14 weregrammar school graduates. Four had no secondary school certificate, 28 had com-pleted apprenticeships, 7 had higher education, and for one subject no informationwas given regarding education level. No significant differences were found in theattachment categories for sex or any other sociodemographic variable (sex: Fisher’sExact Test for 2 × 3 table, p = .69, n. s.; age: ANOVA p = .58, n. s.; education: Fisher’sExact Test p = .62, n. s.). It may be assumed that the expected main results, namelythe relationship between the attachment categories and the linguistic measures, willnot be influenced by sociodemographic variables.

PROCEDURE

For the computer-based text analyses, the interviews were additionally segmentedby computer into word blocks of 150 words each and compiled into a main textbody. The analysis was carried out with the text analysis program TAS/C (Mergenthaler,1993). For each interview, the absolute and relative frequency for the text length(token) were calculated, as were the ET (overall and specifically as positive andnegative), the AB, the proportions of the Emotion-Abstraction Patterns (A, B, C, D)and the Computerized Referential Activity (CRA). Regarding verbal activity‚ the group“preoccupied” showed the highest proportion of all tokens (text length) with 7,324words; “dismissing” with 4,775, the lowest; and “secure,” with 6,690, lay in the middle.

The text analysis data and the categorizations into attachment representationswere summarized in a data matrix (N = 40) and then evaluated with the softwarepackage SYSTAT for Macintosh. Nonparametric procedures were used, since we didnot assume a normal distribution in the data from the texts. Comparisons over allthree groups were done using Kruskal-Wallis tests. Paired comparisons were donewith the Mann-Whitney U-Test. The effect size (d) was defined by Cohen (1977) asthe difference between two group averages divided by an estimation of the standarddeviation. An effect size >0.2 is considered weak, >0.4 moderate, and >0.8 strong.

MEASURES

Attachment Q-Sort by Kobak (1993). The Attachment Q-Sort is considered areliable alternative method to the coding system of Main & Goldwyn (1994). It isbased on the summary of representative items with which the coherence of the com-pleted AAI, as well as the experiences of the interviewee with his or her parents andtheir evaluation or processing may be rated. Kobak (1993) created a total of 100 typicalitems which, using an expert’s ratings, can be assigned to the prototypical idealQ-sorts “secure,” “dismissing,” and “preoccupied.”

First, two trained raters formed an overall impression by reading the entire in-terview. Then they assigned the 100 items of one scale with the descriptions “very

396 BUCHHEIM AND MERGENTHALER

characteristic” to “very uncharacteristic,” according to a given distribution with setfrequencies. Then, from the Q-Sorts of both raters, a combined Q-sort score was formedand the reliability was calculated. If the reliability (Spearman-Brown) was belowrs = 0.58, a third rater was included in the procedure (normally the values were be-tween rs = 0.64 and rs = 0.93; see Zimmermann, 1994; Becker-Stoll, 1997). Correla-tions were then calculated between the average score and the ideal prototypes. Thehighest correlation then determined the classification of a subject into the correspond-ing attachment style. The average interrater reliability between the two independentraters was with three classes κ = 0.88.3

The Q-Sort procedure by Kobak (1993) also enabled the summary of individualitems into a sum-score in terms of their contents and the formation of so-called mega-items. As in the work of Winter (1997), the mega-items “integration skill” and “co-herence” were used in order to relate them to the linguistic variables in the presentstudy.4 The mega-item “integration skill” summarized the extent to which a personis able to connect concrete memories with general abstract relations, to integratepositive and negative events, and to accommodate those into a well developed un-derstanding of one’s own personality. The mega-item “coherence” summarized theextent to which people appeared believable in the interview and whether they an-swered the questions asked and did so in a structured and clear way. It also deter-mined whether the information given was free of contradiction, lively, detailed enough,and simple. To calculate the internal consistency of the mega-items, an item analysiswas carried out, which for the sample of the present study resulted in a high internalconsistency for both mega-items (α = 0.96).

Among other reasons, the Kobak method was chosen because the evaluation ofthe interviews required two independent raters and so showed a calculable reliability.

Computer-based text analysis. Text analyses are based on dictionaries for the Ger-man language, which contain word forms as markers for the constructs to be measured(Mergenthaler & Kächele, 1996; Mergenthaler, 1996, 1997, 1998). In order to cover thebiggest possible amount of text, a remainder word list was compiled from the entiretext and scanned for words as markers of ET or AB and the dictionaries were updated.

The ET dictionary consisted of word forms which are suitable to express affect(but not always to represent an affect). The German dictionary now comprises over10,000 entries and covers in this sample 4.3% of the text, 2.2% assigned to positiveand 2.1% to negative ET.

The AB dictionary was compiled using an automatic suffix analysis (Günther &Groeben, 1978) and expanded. It contained words with endings which are typical forabstract word forms (e.g. -keit, – heit, and -ung, which are comparable to -ness, -ment,or -tion in the English language). From the list of remainder words, additional abstractnouns were taken, which were not recognized by the suffix analysis. The Germandictionary now comprises almost 7,000 entries and covers 2% of the text in our sample.

Emotion-Abstraction Patterns. To capture the quantitative interaction of the mag-nitude of each of the two variables ET and AB, the four Emotion-Abstraction Pat-

3The first author is certified by Mary Main and Eric Hesse for coding the AAI. Thanks to Dr. FabienneBecker-Stoll, the second rater (also reliable in the Main-system), and Dr. Peter Zimmermann, the thirdrater, for reliable corating of the AAIs.4Thanks to Dipl.- Psych. M. Winter and Dr. P. Zimmermann for their cooperation and help in respect ofthe mega-items in this study.

ATTACHMENT REPRESENTATION AND LINGUISTIC MEASURES 397

terns were determined. The relative frequencies of the ET and AB words were stan-dardized using z-transformations over all subjects. With these sample-specific z-scores,the Emotion-Abstraction Pattern was determined for each word block as follows:Relaxing: ET and AB < 0; Reflecting: ET < 0 and AB > 0; Experiencing: ET > 0 andAB < 0; Connecting: ET and AB > 0. As the final step, the relative frequencies of theword blocks assigned to the patterns A, B, C, and D were calculated casewise.

The CRA dictionary. This dictionary allowed the automatic assessment of a nar-rative language style in text samples (Mergenthaler & Bucci, 1999) based on theconcept of Referential Activity. The German version was derived as a translation ofthe original English dictionary. These words represented the concreteness, specific-ity, clarity, and imagery of a spoken text. These included function and action wordsas well as pronouns, which are typically used when a person recounts narratives orepisodes. The calculation of the CRA was done using the percentage of all high RAwords5 in the given text sample, multiplied by the natural logarithm of the numberof words in the text (Mergenthaler & Bucci, 1999).

Due to the diversity of verbal practices with attachment-relevant information, aswell as of the skill of verbalizing emotions and cognitions within the attachmentgroups, it was assumed that differences between the attachment groups would bediscernable with the text measures.

Hypotheses. The explorative character of the present study is expressed in thefollowing hypotheses: There are differences between the three attachment groups interms of the variables: (1) ET with positive and negative valence; (2) AB; (3) Emo-tion-Abstraction Patterns A, B, C, D; and (4) CRA.

RESULTS

RELATIONSHIPS BETWEEN ATTACHMENT REPRESENTATIONAND THE LINGUISTIC MEASURES

According to our first hypothesis for ET (see Table 1), there were significantdifferences between all attachment groups (p > .05). The “preoccupied” group ver-balized emotions most frequently. Compared to the “dismissing” group, the differ-ence showed a large effect (d = 1.61). Compared to the “secure” group, it was smaller(d = 0.85). The “dismissing” group verbalized the least emotion. However, the dif-ference between this and the “secure” group was not significant. The effect size wasmoderate (d = 0.36). With the differentiation of positive and negative ET, it was re-vealed that the differences found were largely due to differential use of negativeemotion words. The “preoccupied” group verbalized the most negative emotions,and the differences with the “dismissing” group and the “secure” group showed largeeffects (d = 1.21 and 1.09). For positive ET, there were smaller effects in the overallcomparisons.

In AB (see Table 1), the difference between all groups revealed a statistical trendbut was not significant (p > .10). Subjects in the “dismissing” group used AB the least.

5The German version has 167 high RA words; Spearman correlations for various literary texts compar-ing the original with the translation were around .7.

398 BUCHHEIM AND MERGENTHALER

The “preoccupied” group showed the most AB. Here also, the difference betweenthe extreme groups was most evident. The difference between the “secure” and“preoccupied” groups revealed a statistical trend with a strong effect size (d = 1.16).There was a small effect between the “secure” and “dismissing” groups (d = .35).The second hypothesis was partly confirmed.

According to our hypothesis, the narrative style, measured as CRA (see Table 1),was highly significantly different over all three attachment groups (p > .01). The lowestvalues were shown by the “dismissing” group, and the highest by the “preoccupied”group. Again, the subjects of the “secure” group lay in the middle. The differencesbetween the extreme groups showed large effects (d = 1.62), as did the differencesbetween the “secure” and “dismissing” groups (d = 1.19). Between the secure and“preoccupied” group there was a small effect (d = 0.35). Hypothesis four can there-fore also be confirmed.

A stepwise discriminant analysis with the linguistic variables allowed the twovariables negative ET and CRA to be chosen as predictors of attachment groups. Usingthese two variables, the three attachment groups could be predicted with a successrate of 60% (Jackknife method), which lies clearly above the 33.3% correspondenceto chance. Prediction was relatively good for both extreme groups, the “dismissing”and the “preoccupied,” with 70% success each, and the “secure” group with 50%.The “wrong” predictions here fell symmetrically into both of the insecure categories,which is confirmed by the structure of the stated results.

For the Emotion-Abstraction Patterns (see Table 2), the most distinct differencesbetween all attachment groups were found for the patterns Relaxing (p > .05) andConnecting (p > .05). While the “secure” group again lay in the middle, the “dis-missing” and “preoccupied” groups showed contrary results. The “dismissing” grouphad the highest values in pattern A, and the “preoccupied” group had the highestvalues for pattern D. Both groups were different for pattern A (d = –1.51) and forpattern D (d = 1.23) with large effects. The differences for the “dismissing” and“preoccupied” groups showed statistical trends toward significance. In pattern B,the “secure” group was least represented. The difference in the “preoccupied” groupshowed a statistical trend. For the pattern C there were no differences between thegroups.

TABLE 1. Differences in the Linguistic Measures Withinthe Attachment Categories

Attachment categories Statistics

Linguistic Ds F Emeasures n = 10 n = 20 n = 10 Ds/F Ds/E F/E Ds/F/E

M SD M SD M SD d d d H

ET 3.95 .55 4.23 .83 4.86 .58 .36 1.61 .85 7.59**

pos. ET 2.02 .27 2.20 .55 2.40 .54 .39 .90 .37 2.18neg. ET 1.94 .52 2.02 .43 2.45 .30 .19 1.21 1.09 9.09**

AB 1.66 .51 1.92 .81 2.50 .89 .35 1.16 .69 5.18*

CRA 1.92 .15 2.15 .20 2.22 .21 1.19 1.62 .35 10.09***

Note. Ds = dismissing, F = secure, E = preoccupied, M = mean, SD = standard deviation, d = effect size;H: p Kruskal-Wallis-Test; ET = Emotional Tone, pos ET = positive Emotional Tone, neg ET = negativeEmotional Tone, AB = Abstraction, CRA = Computerized Referential Activity.*p < .10; **p < .05; ***p < .01.

ATTACHMENT REPRESENTATION AND LINGUISTIC MEASURES 399

RELATIONSHIPS BETWEEN MEGA-ITEMSOF THE ATTACHMENT Q-SORT AND LINGUISTIC MEASURES

WITHIN THE ATTACHMENT CATEGORIES

The correlations (Spearman) between the two mega-items “emotional integrationskill” and “coherence” with the linguistic variables are shown below in Tables 3 and 4.

The results in Table 3 describe the subjects in the “secure” group in this way: Themore these people emotionally integrated attachment-relevant events, the less theywere found in a Relaxing state (low levels of ET and AB; p > .10) and the more theyshowed the Connecting pattern (state of simultaneous reflection on and verbalizationof emotional contents, ET and AB were high; p > .05). For the “preoccupied” group,exactly the opposite was true: There was a positive relationship between the mega-item “integration skill” and the Relaxing pattern (p > .10) as well as a negative relation-ship between the mega-item “integration skill” and the Connecting pattern. This corre-lation was, however, nonsignificant. A highly significant negative correlation betweenthat mega-item and pattern B was also discovered: The more the interviews of this

TABLE 2. Differences in the Emotion-Abstraction-Patterns (EAP) MeasuresWithin the Attachment Categories

Attachment categories Statistics

EAP Ds F En = 10 n = 20 n = 10 Ds/F Ds/E F/E p-Ds/F/E

M SD M SD M SD d d d H

Pattern A 42.48 17.03 36.39 22.49 10.89 12.40 –.29 –1.51 –.83 6.17*

Pattern B 20.22 10.69 18.98 9.65 21.09 9.57 –.12 .08 .22 0.33Pattern C 21.29 7.72 20.86 8.27 22.81 13.59 –.05 .14 .19 0.52Pattern D 16.00 9.15 23.77 20.30 36.20 21.27 .44 1.23 .60 5.64*

Note. Ds = dismissing, F = secure, E = preoccupied, M = mean, SD = standard deviation, d = effect size;H: Kruskal-Wallis-Test; Pattern A = Relaxing, Pattern B = Reflecting, Pattern C = Experiencing, PatternD = Connecting.*p < .05.

TABLE 3. The Correlative Relationships (Spearman) Between the Mega-Item“Emotional Integration Skill” and the Linguistic Variables

Mega-item “emotional integration skill”

Linguistic measures Dismissing (n = 10) Secure (n = 20) Preoccupied (n = 10)

ET .17 .30 .34AB .04 .22 –.46A .05 –.43+ .59+

B .16 .29 –.79**

C –.29 .003 .49D .04 .49* –.44CRA .44 –.06 .36

Note. ET = Emotional Tone, AB = Abstraction, Pattern A = Relaxing, Pattern B = Reflecting, Pattern C =Experiencing, Pattern D = Connecting, CRA = Computerized Referential Activity.*p < .05; **p < .01.

400 BUCHHEIM AND MERGENTHALER

group were characterized by “integration skill,” the less they contained signs of reflec-tion (p > .01). For the group “dismissing,” no significant relationships were found.

An interesting result (see Table 4) was a trend in the “preoccupied” group indicat-ing a negative relationship between the mega-items “coherence” and Connecting. Thissuggests that when these people showed signs of the Connecting pattern, they werenot coherent in the interview. This finding is relevant because this group showed thehighest proportion of the Connecting pattern, compared to the “secure” and “dismiss-ing” groups (see previous results). The test of the equality of the three correlationsshowed on a 10% level of significance that the attachment groups might be different interms of Emotion-Abstraction Pattern Connecting (p > .10).

DISCUSSION

The aim of this study was to bring together two different approaches—a discourseanalytic and a computer-aided method—to the textual material of the AAI. Themethodological challenge was to perform the linguistic analysis of the AAI usingeconomical text analysis methods and then to evaluate this approach in terms of itscompatibility and usefulness.

The most important result of the present study is that significant differences amongthe attachment categories in the linguistic measures were found, supporting all ofthe hypotheses we have postulated. A consistent rank-order was revealed across thethree attachment categories (“dismissing,” “secure,” “preoccupied”) in the means ofthe linguistic measures (see Table 1): The “dismissing” group consistently showedthe lowest values, the “secure” group the intermediate ones, and the “preoccupied”group the highest means. We conclude from this that a rough differentiation usingonly the categories “secure” and “insecure” is not suitable. In our study, the twoinsecure categories “dismissing” and “preoccupied” form divergent groups on a lin-guistic level. Putting them into a single group would hide these differences. Theopposing strategies of blocking attachment-relevant feelings described in Kobak’sQ-Sort method (Kobak, 1993; Kobak, Cole, Ferrez-Gilles, Fleming, & Gamble, 1993)—the deactivating strategy of the “dismissing” versus the hyperactivating strategy ofthe “preoccupied”—is clearly confirmed by our findings.

TABLE 4. The Correlative Relationships (Spearman) Between the Mega-Item“Coherence” and the Linguistic Variables

Mega-item “coherence”

Linguistic measures Dismissing (n = 10) Secure (n = 20) Preoccupied (n = 10)

ET .10 .24 –.29AB –.16 .12 –.47A .45 –.23 .25B –.35 .33 –.36C –.46 –.18 .51D .19 .26 –.54*

CRA .32 –.08 –.08

Note. ET = Emotional Tone, AB = Abstraction, Pattern A = Relaxing, Pattern B = Reflecting, Pattern C =Experiencing, Pattern D = Connecting, CRA = Computerized Referential Activity.*p < .10.

ATTACHMENT REPRESENTATION AND LINGUISTIC MEASURES 401

• Subjects from the “dismissing” group use the fewest words with ET in the three-group comparison, show the lowest proportion of the Emotion-AbstractionConnecting pattern, but most frequently use the complementary Relaxingpattern, a state of disorientation in which they hardly mention or considerfeelings, or do not even know what to talk about. They also show little CRA,i.e., they rarely express themselves clearly, specifically, concretely, or viv-idly. The results of the text analysis indicate that emotionally distanced lan-guage and verbal paucity “is associated with a notable absence of expres-sions of emotional vulnerability” (Hesse, 1999, S. 424).

• On the contrary, subjects in the “preoccupied” group most frequently usedemotionally negative words in the three-group comparison. They most fre-quently showed the Connecting pattern, that is, they used both concepts ofemotion and abstraction more often at the same time. Their style of speechwas overly characterized by clear, specific, concrete, and vivid expression asshown by high CRA proportions. Obviously, the verbally excessive, enmeshed,and conflict-laden character of this group reflected a comparatively high ver-bal productivity in most linguistic measures. According to the definition ofthis category, preoccupied subjects in this sample verbalized negative emo-tions in a heightened manner: “[P]reoccupied (sub)categories are associatedwith the expression of anger and of fear, respectively” (Hesse 1999, p. 424).Interesting, and at the same time astonishing, is the high concurrent use of ETand AB. This result will be further discussed below, comparing two transcriptexamples.

• The subjects classified as “secure” showed moderate values in all linguisticcomputer measures. This seemed to indicate a balanced approach to the ver-balization of feelings and thought processes, and the contribution of narra-tive material. The moderateness of this group is consistent with other resultsfrom attachment research which show flexibility and balance in the regula-tion of emotion and cognition in securely attached persons (Zimmermann,1994, 1999; Hesse, 1999).

The confirmation of our hypotheses, which we expressed nondirectionally, is inour view no trivial result. First, all comparisons were carried out using relative fre-quencies. Already-known differences in the speech activity of the three attachmentgroups do not therefore imply a differentially frequent use of cognitive concepts.Furthermore, the definitions of the attachment types on the one hand and the Emo-tion-Abstraction Patterns and CRA on the other hand are conceptually different fromeach other. We, therefore, considered these results to be a mutual validation of thesedifferent concepts.

It should, however, be critically noted that the AAI (George et al., 1985) wasdeveloped in a nonclinical context and therefore lent itself to an explorative approach,in combination with the computer-based methods which were validated in a psy-chotherapeutic context. We therefore did not make any directional hypotheses. Also,the semistructured AAI, with its standardized questions, often interrupts the flow ofspeech and steers it in a new direction, or induces positive or negative emotions inthe interviewee. In contrast to transcribed psychotherapy sessions, the AAI providesnarrative material which is primarily unmarked by free association, but rather by animposed structure due to the specific interventions of the interviewer. The theory ofthe Therapeutic Cycle by Mergenthaler, in its definition, relates to texts which pro-vide freely developing dialogues between patient and therapist.

402 BUCHHEIM AND MERGENTHALER

Typically, it is only in the course of their therapy that patients develop the kindof expression, including in their style of speech, which resembles that of a normalpopulation. For psychodynamically oriented therapy with patients with neurotic dis-orders, these changes can be described using the concept of the Therapeutic Cycle.ET and AB, in the course of a therapy and with a time lag, increase and then declineagain, with a higher baseline for the Connecting pattern at the end of the therapy(Mergenthaler, 1996).

The Connecting pattern is understood by Mergenthaler in the sense of a drugwhich, similar to an antibiotic, is especially needed during an acute illness, but which,after recovery, is needed only in smaller doses (when it would be comparable withthe auto-immune function after stopping antibiotics). These considerations may alsobe valid for CRA when measuring RA as part of the Referential Cycle. The notionthat “high RA is associated with . . . better therapeutic outcome” (Mergenthaler &Bucci, 1999, p. 342) may have to be reconsidered as well, taking into account theprocesses of symbolizing (Bucci, 1997), as well as psychodiagnostic criteria. Pilotstudies with other therapeutic orientations and spectrums of illness show partly dif-ferent courses (Nicoló et al., 2000; Mergenthaler, 2000).

We would expect that a successful patient is comparable with a “secure” subjectfrom a normal population. These people characteristically are able to verbalize feel-ings and at the same time reflect on them, as well as use concrete, clear, specific,and vivid language. The results of this study show that the “secure” subjects are lo-cated between the extremes, which is where successful patients should also be real-istically located. In future studies with the Therapeutic Cycle Model and CRA, thisshould certainly be considered. The “preoccupied” group had the highest propor-tion of Connecting, which could give the impression that they were verbalizing andreflecting on conflicts at the same time. However, these subjects do not have theability to integrate positive as well as negative emotions, nor to adopt a self-reflectivecoherent metacognitive perspective. According to Main & Goldwyn (1994), a personwith the attachment classification “preoccupied” may speak with an “air of author-ity” regarding past experiences, and may well be “pseudoprofessional,” “pseudo-analytical,” or utilize the expressions of “pop psychology.” In order to illustrate this,we have included two excerpts from the transcripts.

The excerpt shown in Table 5 is from a female subject classified as “preoccu-pied.” She is asked about how she sees her childhood experiences from today’sperspective. The subject presents herself as clearly enmeshed in relation to this ques-tion. Characteristically, she is verbalizing negative, angry feelings regarding her rela-tionship with her mother in the present tense (“that I’m so aggressive”). She usesunclear quotes, so that it is not clear who is actually talking. Her attempt to distanceherself abstractly, by referring to the “talents” of each of her siblings, is not success-ful. In the last part of the sequence it becomes clear that the subject, despite pseudo-psychological analyses (“I got my self-image at University, where I kept my distance”)is still entangled with her story (“but it was too directive, that still gets to me”). Thispassage shows that emotion and abstract words also occur frequently, but are juxta-posed in a nonintegrated manner.

The transcript of a person classified as “secure” (see Table 6) gives, in contrastto the last example, a more integrated impression. This subject is also asked howshe sees her earlier experiences from today’s perspective. She gives an open ac-count of her difficulties with her father during puberty (“as a kid I was like always abit scared of him”), and indicates that she also sometimes distanced herself from himin order to show her limits. Her attempts at solving the problem seem more convinc-

ATTACHMENT REPRESENTATION AND LINGUISTIC MEASURES 403

ing. She is able to express her development verbally, without anger (accusations inthe present), and to illustrate from a metaperspective the things which indicate hernewly gained autonomy (“now we’re kind of partners, so now what I say is just asgood as what he says”). The different problematic feelings seem to be more inte-grated and end in this passage in an overall self-reflective judgement (“but it’s now,kind of, it’s like a partnership”).

Both these examples should be considered prototypical, as they reflect the maindifferences between the two attachment categories “secure” and “preoccupied.” Theyare intended to demonstrate that the coincidence of verbalized emotions and ab-stractions per se cannot automatically be considered insight. It seems advisable withthis text material to analyze the passages in terms of their coherence. This view couldalso be significant for the identification of “pseudoinsightful” passages in transcribedpsychotherapy sessions. They may be recognized as a turning point through the text-analysis, but from a therapist’s subjective point of view, they may appear to be ratherfruitless. This also confirms the independence of the construct coherence. Neitherthe Emotion-Abstraction Patterns nor the CRA can measure this aspect. Since, how-ever, coherence is one of the most important characteristics of the “secure” group, itis not reasonable to expect that the computer-assisted text measures presented herewould be predictive or discriminant of the other attachment groups. Here, close-updiscourse analyses are currently the choice method. The identification of linguisticmarkers for the construct of coherence, therefore, will be part of our future research.

The differences between the “secure” and “preoccupied” groups with respectto CRA do not reach significance in this study, although the “preoccupied” groupshowed a slightly higher mean (see Table 1). This indicates that both groups arecharacterized in this study by concrete, clear, specific, and vivid language. Thesimilarity of these groups in terms of CRA would appear to fit with the coding

TABLE 5. Transcript Example: Excerpt From a Passage of the EmotionAbstraction Pattern “Connecting” (High Emotion, High Abstraction) of anInsecure-Preoccupied Person

I: Hmhm. Do you have the feeling that these experiences still trouble you? How do yousee them from today’s perspective?

S: Ah, like I said, well, these dreams get to me.I: Hmhm.S: That I’m so aggressive towards to my mother. So aggressive that I really “You’re not

going to start bossing” and “I’m going to do this now,” or “I know that better thanyou too.”

I: Hmhm.S: Or so that and that I was really like I think there was really a too unnatural tone

there. I wouldn’t even say that she did it badly or meanly or anything, but it wasunnaturally strong and I hope that I don’t like the same sometimes it seems to me sodominant I don’t know intensive that I don’t use so much pressure. Like on that wewere really on us all we all have to look at how independent—each one of us hasour talents, for Reni theology, for my sister art and crafts, my brother’s a doctor, forme—all found directions but, and in that direction we can really

I: So what do do you think stopped you?S: (takes a breath) We like had a hard decision, and I know so little who who you are,

yeah self-esteem I got my self-image at University where I kept my distance, but itwas too directive, that still gets to me (takes a breath) really gets to me.

Note. I = interviewer, S = subject; translated from literal German transcription.

404 BUCHHEIM AND MERGENTHALER

difficulties of clinical raters in discriminating between the groups “secure” and“preoccupied.” This may be due to the fact that the latter group often misleadswith its narrative richness, imagery, and specificity, and that the CRA measure doesnot take into account the aspect of incoherence (e.g., going off the track, anger,passivity). On the other hand, people classified as secure sometimes give the im-pression of being verbally excessive, although they do come back to the topicaccording to Grice’s (1975) coherence criteria. These boundary-zones are barelyidentifiable, since both groups presumably show a similar narrative surface-structure. It would be well worth discussing whether there may be a high amountof Referential Activity which, despite the richness of concreteness, detail, imagery,and specificity of the narrative style, represents an underlying irresolution andconflict-stricken state. Again, the work on this question will be the object of fur-ther studies (see also Buchheim, 2000).

In sum, we may note that this computer-aided text analysis of the AAI has re-sulted in a consistent differentiation of the three classical attachment types, andcorroborates the findings and theory of previous attachment studies. The fact thatthe “secure” group is located in the middle value-range between the two extremegroups (“dismissing” and “preoccupied”) is a substantial finding for clinical research.The next step should be to test, using a larger sample as well as with clinical andnonclinical subjects, whether the results reported here can be confirmed. The dis-cussion of the results in relation to the concept of coherence shows that the mereconsideration of the score on the language measures within the attachment groupsis not suitable as a direct substitution of a complex discourse analysis of the AAI.The discriminant analyses suggest that the computer-assisted language measures

TABLE 6. Transcript Example: Excerpt From a Passage of the EmotionAbstraction Pattern “Connecting” (High Emotion, High Abstraction) of aSecure-Autonomous Person

I: Now, if we could talk about the change in the relationship to your parents. So like,this process from the past to now, what kind of a feeling have you got, did therelationship change or has the relationship stayed more the same?

S: Hm. Well, it’s somehow (laughs) totally changed. Well, like with my dad, I mean as akid was like always a bit scared of him, really of his authority or also that he getsdrunk, going to such an excess or something, and then like as a teenager I wa- wasmore and more distanced, so when I was like, 14 or something, from that age I didn’tkiss him anymore, didn’t want to anymore, so just before that, set my limits, anddidn’t really didn’t, eh, ask him much or anything, so I like distanced myself. And thatthen went on with me for a really long time, right up to my studies, and then actually,in my firs- quite late in my studies, I somehow for me, I think I on the one hand Ifreed myself, and I could find my own opinions, at home, too. And also, yeah, I’ve—through various relationships with men—and by self or in general I’ve found adifferent relationship to men, so practically at some stage I a security to be able toface him again now, and actually I’ve now actually sought the contact myself and isnow somehow like, different now, we’re kind of partners, so what I say is just asgood as what he says, and it somehow.

I: Hmhm.S: Yeah, so now it’s like, can now suddenly get along with him or if he like, yeah,

expresses opinions, which like before then I can rebut them or I can just let it be, butit’s now kind of, it’s become like pretty much like a partnership.

Note. I = interviewer, S = subject; translated from literal German transcription.

ATTACHMENT REPRESENTATION AND LINGUISTIC MEASURES 405

for negative ET and CRA may differentiate among all three attachment groups, witheven better success if only classifying the insecurely attached subjects as either“dismissing” or “preoccupied.” Furthermore, using computer-assisted languagemeasures as a means of transcript analysis, this study provides an independentaccount for the criteria postulated by Main and Goldwyn (1994), as well as Kobak(1993), for differentiating between attachment typologies. We may conclude thatthe cognitive concepts of Emotion-Abstraction Patterns and narrative style as mea-sured through CRA not only relate to clinical aspects of therapeutic change but arealso sensitive to attachment representation.

REFERENCES

Becker-Stoll, F. (1997). Interaktionsverhaltenzwischen Jugendlichen und Müttern im Kontextlängsschnittlicher Bindungsentwicklung. Doc-toral dissertation, University Regensburg.

Benoit, D., & Parker, K. H. C. (1994). Stability andtransmission of attachment across three genera-tions. Child Development, 65, 1444–1456.

Böhmer, M. (2000). Forensiche Psychotherapie-prozeßforschung: Eine Einzelfallstudie. Doctoraldissertation, University of Ulm.

Bowlby, J. (1969). Attachment and loss. Vol. 1:Attachment. (2nd ed.). (Vol. Dt. [1975]: Bindung.München: Kindler). New York: Basic Books.

Bowlby, J. (1973). Attachment and loss. Vol. 2:Separation. Anxiety and anger. (Vol. Dt. [1976]:Trennung. München: Kindler). New York: Ba-sic Books.

Bowlby, J. (Ed.). (1980). Attachment and loss. Vol.3: Loss, sadness and depression. (Vol. III ed.).(Vol. Dt. [1983]: Verlust. München: Kindler).London: Hogarth Press.

Bretherton, I. (1999). Internal working model inattachment relationships: A construct revisited.In J. Cassidy & P. Shaver (Eds.), Handbook ofattachment (pp. 89–111). New York: GuilfordPress.

Bucci, W. (1985). Dual coding: A cognitive modelfor psychoanalytic research. Journal of theAmerican Psychoanalytic Association, 33, 571–607.

Bucci, W. (1997). Psychoanalysis and cognitivescience. New York: Guilford Press.

Bucci, W., & Kabasakalian-McKay, R., and the RAresearch group. (1992). Instructions for scoringreferential activity (RA) in transcripts of spokennarrative texts. Ulm: Ulmer Textbank.

Buchheim, A. (2000). Bindungsrepräsentation,Emotions-Abstraktionsmuster und NarrativerStil: Eine computerunterstützte Textanalyse desAdult Attachment Interviews. Doctoral disserta-tion, University of Ulm.

Buchheim, A., Brisch, K. H., & Kächele, H. (1998).Einführung in die Bindungstheorie und ihreBedeutung für die Psychotherapie. Psychother-

apie, Psychosomatik, medizinische Psychologie,48, 128–138.

Cohen, J. A. (1977). Statistical power analysis forthe behavioral sciences. (2nd ed.). New York:Academic Press.

Fonagy, P., Steele, H., & Steele, M. (1991). Mater-nal representations of attachment during preg-nancy predict the organization of infant-motherattachment at one year of age. Child Develop-ment, 62, 891–905.

Fuendeling, J.-M. (1998). Affect regulation as astylistic process within adult attachment. Jour-nal of Social and Personal Relationships, 15,291–322.

George, C., Kaplan, N., & Main, M. (1985). TheAdult Attachment Interview. Unpublished manu-script. University of California, Berkeley.

Greenberg, L., & Safran, J. (1987). Emotion inpsychotherapy: Affect, cognition, and the pro-cess of change. New York: Guilford Press.

Grice, H. P. (1975). Logic and conversation. InP. Cole & J. L. Moran (Eds.), Syntax and seman-tics (pp. 41–58). New York: Academic Press.

Günther, U., & Groeben, N. (1978). Abstraktheits-suffix-Verfahren: Vorschlag einer objektiven öko-nomischen Messung der Abstraktheit/Konkretheitvon Texten. Zeitschrift für experimentelle undangewandte Psychologie, 25, 55–74.

Hesse, E. (1999). The Adult Attachment Interview:Historical and current perspectives. In J. Cassidy& P. Shaver (Eds.), Handbook of attachment(pp. 395–433). New York: Guilford Press.

Kobak, R. R. (1993). The Attachment-Q-Sort. Un-published manuscript, University of Delaware.

Kobak, R. R., Cole, H. E., Ferrez-Gilles, R., Fleming,W. S., & Gamble, W. (1993). Attachment andemotion regulation during mother-teen-problemsolving: A control theory analysis. Child Devel-opment, 64, 231–245.

Magai, C. (1999). Affect, imagery, and attachment.In J. Cassidy & P. R. Shaver (Eds.), Handbook ofattachment (pp. 787–802). New York: GuilfordPress.

Main, M. (1991). Metacognitive knowledge, meta-

406 BUCHHEIM AND MERGENTHALER

cognitive monitoring and singular (coherent) vs.multiple models (incoherent) of attachment. InC. M. Parkes, J. Stevenson-Hinde, and P. Marris(Eds.), Attachment across life cycle (pp. 127–159). London, NY: Tavistock.

Main, M. (1994). Recording and transcribing theAdult Attachment Interview. Unpublished manu-script, University of California, Berkeley.

Main, M., & Goldwyn, R. (1994). Adult AttachmentScoring and Classification Systems. Unpublishedmanuscript, University of California, Berkeley(1st ed., 1985).

Main, M., Kaplan, N., & Cassidy, J. (1985). Securityin infancy, childhood and adulthood: A moveto the level of representation. In I. Bretherton &E. Waters (Eds.), Growing points of attachmenttheory and research (Serial No. 209, Vol. 50,pp. 66–106). Chicago: University of ChicagoPress.

Mergenthaler, E. (1993). TASIC User Manual. UlmerTextbank, Ulm.

Mergenthaler, E. (1996). Emotion-abstraction pat-terns in verbatim protocols: A new way of de-scribing psychotherapeutic processes. Journalof Consulting and Clinical Psycholology, 64,1306–1315.

Mergenthaler, E. (1997). Emotions/Abstraktions-muster in Verbatimprotokollen. Frankfurt: VAS-Verlag für akademische Schriften.

Mergenthaler, E. (1998). Cycles of Emotion-Abstrac-tion Patterns: A way of practice oriented processresearch? The British Psychological Society, Psy-chotherapy Section Newsletter, 24, 16–29.

Mergenthaler, E. (June 2000). Patterns of emo-tional tone and abstraction related to affectiveproblem solving and outcome. Paper presentedat the 31st Annual Meeting of the Society forPsychotherapy Research, Chicago, IL.

Mergenthaler, E., & Bucci, W. (1999). Linking ver-bal and non-verbal representations: Computer-analysis of referential activity. British Journal ofMedical Psychology, 72, 339–354.

Mergenthaler, E., & Kächele, H. (1996). Apply-ing multiple computerized text analytic mea-sures to single psychotherpay cases. Journalof Psychotherapy Practice and Research, 5, 307–317.

Mergenthaler, E., & Stinson, C. (1992). Psycho-therapy transcription standards. PsychotherapyResearch, 2, 125–142.

Nicoló, G., Semerari, A., Pontalti, I., Catania, D.,Carcione, A., & Mergenthaler, E. (June 2000).The therapeutic cycle model and cognitive psy-chotherapy. Paper presented at the 31st AnnualMeeting of the Society for Psychotherapy Re-search, Chicago, IL.

Paivio, A. (1971). Imagery and verbal processes.New York: Holt, Rinehardt & Winston.

Paivio, A. (1978). The relationship between verbaland perceptual codes. In E. C. F. Carterette (Ed.),Handbook of perception: perceptual coding 8(pp. 375–397). New York: Academic Press.

Samstag, N. (1996). A meta-analysis of Referen-tial Activity. Doctoral dissertation, Universityof Adelphi.

Stone, P. J., Dunphy, D. C., Smith, M. S., & Ogilvie,D. M. (1966).The general inquirer: A computerapproach to content analysis. Cambridge, MA:M.I.T. Press.

Strunck, W. J., & White, E. B. (1972). The elementsof style. New York: Macmillan.

Winter, M. (1997). Bindungsrepräsentation undVerhalten in einer komplexen Problemlösesitu-ation bei Jugendlichen: Strategien beim Umgangmit der Aufgabe und emotionale Selbsteinschät-zung. Diplomarbeit, University of Regensburg.

Zimmermann, P. (1994). Bindung im Jugendalter:Entwicklung und Umgang mit aktuellen An-forderungen. Doctoral dissertation, Universityof Regensburg.

Zimmermann, P. (1999). Emotionsregulierung imJugendalter. In W. Friedlmeier & M. Holodynski(Eds.), Emotionale Entwicklung (pp. 219–240).Heidelberg: Spektrum der Wissenschaft.

ZusammenfassungZiel der Studie war es zu prüfen, ob computergestütze, ökonomisch zugängliche linguistische Textmaße(Emotionsabstraktionsmuster, Mergenthaler 1996; computerisierte referentielle Aktivität, Mergenthaler & Bucci1999) sich dafür eignen, die schwieriger zugänglichen Bindungsrepräsentationen zu differenzieren, wie siedas Erwachsenenbindungsinterview erfaßt (Main & Goldwyn, 1994). Eine Untersuchung bei 40 gesundenPersonen ergab das konsistente Ergebnis, daß von den beiden unsicheren Bindungskategorien die Gruppeder „Vermeidenden“ (n = 10) die niedrigsten Werte in allen Textmaßen zeigten, während die Gruppe der„Verwickelten“ (n = 10) die höchsten Werte zeigten. Die Mittelwerte der „Sicheren“ Bindungsgruppe (n =20) lagen dazwischen. Diese Rangreihe stimmt mit den Ergebnissen von Studien überein, die einen anderenFokus verfolgten. Diese Studien zeigten eine Deaktivierung bzw. Überaktivierung von bindungsrelevantenInformationen in den unsicheren Bindungsgruppen und eine Flexibilität in der sicheren Gruppe. Aufbauendauf den Ergebnissen wird diskutiert, ob die Koinzidenz von sprachlichen Markern für Emotion und Abstraktionund die computerisierte referentielle Aktivität sich dazu eignen könnten, das Konstrukt narrativer Kohärenzzu operationalisieren.

ATTACHMENT REPRESENTATION AND LINGUISTIC MEASURES 407

RésuméLe but principal de cette recherche a été de tester le potentiel de mesures de texte linguistiques basées surordinateur et à compiler de façon économique (Patterns d’Emotion - Abstraction, Mergenthaler 1996 ; ActivitéRéférentielle Computerisée, Mergenthaler & Bucci 1999) comparées à des représentations d’attachementcomplexes, moins faciles à compiler, par l’Interview d’Attachement de l’Adulte (Main & Goldwyn 1994).L’étude de n = 40 sujets de contrôle en bonne santé a abouti au résultat consistant qu’entre les deux catégoriesd’attachement insécures, le groupe « détaché » (n = 10) a montré les moyennes les plus basses pour toutesles mesures de texte, et le groupe « préoccupé » les plus hautes. La moyenne du groupe d’attachement « sécure »(n = 20) s’est situé entre les deux. Ce classement est consistant avec les résultats d’études ayant un autrefocus de recherche et démontrant soit une désactivation, soit une hyperactivation d’information relative àl’attachement dans les deux groupes « insécures », ainsi que de la flexibilité dans le groupe « sécure ». Nousdiscutons également si la coïncidence de marqueurs linguistiques pour l’émotion et l’abstraction avec l’ActivitéRéférentielle Computerisée est adéquate pour opérationaliser le concept de cohérence du style narratif.

ResumenEl objetivo principal de esta investigación fue probar el grado en que las mediciones del texto lingüísticocomputarizado (Patrones de Emoción-Abstracción, Mergenthaler, 1996; Actividad Referencial Computarizada,Mergenthaler & Bucci, 1999) son capaces de diferenciar entre las representaciones complejas del apego enla Entrevista de Apego Adulto (Main & Goldwyn, 1994). El estudio de n = 40 controles sanos dio comoresultado consistente el que de las dos categorías de apego inseguro, el grupo “rechazante” (“dismissing”)(n = 10) mostró las menores medias en todas las mediciones de textos, mientras que el grupo “dependiente”(“preoccupied”) (n = 10) mostró las mayores medias. La media del grupo de apego “confiado” (“secure”) (n= 20) se ubicó entre ambos grupos. Estos valores son consistentes con los resultados de estudios con otrofoco de investigación, los que muestran tanto una desactivación como una hiperactivación de la informaciónrelevante al apego en ambos grupos de apego inseguro, así también como una flexibilidad en el grupo“confiado”. También se discute si la coincidencia de marcas lingüísticas para la emoción y la abstracción,son adecuadas para operacionalizar el constructo de coherencia en el estudio del estilo narrativo de laActividad Referencial Computerizada.

Received January 10, 2000Revision received April 24, 2000

Accepted May 24, 2000