an analysis of narrative and figurative language within online alcoholism discussion forums

16
An Analysis of Narrative and Figurative Language within Online Alcoholism Discussion Forums Emily Polander & Valerie L. Shalin Published online: 16 February 2013 # Springer Science+Business Media New York 2013 Abstract This study examined lay individualsuse of narrative and figurative language for communicating biomedical and psychosocial concepts in online alcoholism discussion forums. We performed quantitative and qualitative analyses of content, narrative, and figurative language on 311 opening forum posts. Linguistic patterns indicated a distinction between narrative and figurative functions of language in communication that is sensitive to the concreteness of the content. Individuals were more likely to use narrative to convey concrete psychosocial content, and figurative language to convey abstract psychosocial content. Individuals did not use narrative or figurative language to communicate abstract, technical biomedical content. Although they belong to the biomedical model, concrete physical symptoms receive treatment in a manner more similar to concrete psychosocial content than more abstract, technical biomedical content. Our findings suggest that individ- ualsuse of narrative and figurative language is largely driven by the concreteness or perceptibility of the underlying content rather than the content domain. Keywords Alcoholism . Online communities . Narrative . Figurative language Individuals use language to convey their conceptual models and understanding through the construction of both narrative (Pennebaker and Seagal 1999) and figurative language (Lakoff and Johnson 1980), especially for mental illness (Kleinman 1988; Ochs and Capps 1996; Sontag 1978). This study examines lay individualsuse of narrative and figurative language for communicating understanding of both biomedical and psychosocial concepts surrounding alcoholism in online discussion forums. Our findings suggest that individualsuse of narrative and figurative language is largely driven by the concreteness or perceptibility of the underlying content rather than the content domain. Int J Ment Health Addiction (2013) 11:458473 DOI 10.1007/s11469-013-9431-8 E. Polander : V. L. Shalin Department of Psychology, Wright State University, Dayton, OH 45435, USA E. Polander (*) Department of Psychology, Wright State University, 335 Fawcett Hall, 3640 Colonel Glenn Hwy, Dayton, OH 45435-0001, USA e-mail: [email protected]

Upload: valerie-l-shalin

Post on 10-Dec-2016

214 views

Category:

Documents


0 download

TRANSCRIPT

An Analysis of Narrative and Figurative Languagewithin Online Alcoholism Discussion Forums

Emily Polander & Valerie L. Shalin

Published online: 16 February 2013# Springer Science+Business Media New York 2013

Abstract This study examined lay individuals’ use of narrative and figurative language forcommunicating biomedical and psychosocial concepts in online alcoholism discussionforums. We performed quantitative and qualitative analyses of content, narrative, andfigurative language on 311 opening forum posts. Linguistic patterns indicated a distinctionbetween narrative and figurative functions of language in communication that is sensitive tothe concreteness of the content. Individuals were more likely to use narrative to conveyconcrete psychosocial content, and figurative language to convey abstract psychosocialcontent. Individuals did not use narrative or figurative language to communicate abstract,technical biomedical content. Although they belong to the biomedical model, concretephysical symptoms receive treatment in a manner more similar to concrete psychosocialcontent than more abstract, technical biomedical content. Our findings suggest that individ-uals’ use of narrative and figurative language is largely driven by the concreteness orperceptibility of the underlying content rather than the content domain.

Keywords Alcoholism . Online communities . Narrative . Figurative language

Individuals use language to convey their conceptual models and understanding through theconstruction of both narrative (Pennebaker and Seagal 1999) and figurative language(Lakoff and Johnson 1980), especially for mental illness (Kleinman 1988; Ochs andCapps 1996; Sontag 1978). This study examines lay individuals’ use of narrative andfigurative language for communicating understanding of both biomedical and psychosocialconcepts surrounding alcoholism in online discussion forums. Our findings suggest thatindividuals’ use of narrative and figurative language is largely driven by the concreteness orperceptibility of the underlying content rather than the content domain.

Int J Ment Health Addiction (2013) 11:458–473DOI 10.1007/s11469-013-9431-8

E. Polander :V. L. ShalinDepartment of Psychology, Wright State University, Dayton, OH 45435, USA

E. Polander (*)Department of Psychology, Wright State University, 335 Fawcett Hall, 3640 Colonel Glenn Hwy,Dayton, OH 45435-0001, USAe-mail: [email protected]

Two Models of Alcoholism

Lay individuals encounter both biomedical and psychosocial models of alcoholism. Thebiomedical concepts associated with alcoholism include damaging short and long-termeffects to the liver, brain, cardiovascular system, and nervous system, among other organsand systems (Barclay et al. 2008). Biomedical information related to these concepts includesperceptible physical symptoms of alcohol abuse (e.g., jaundice) along with more abstract,imperceptible underlying human biochemistry (e.g., enzymes), and diagnostic testing (e.g.,liver panels).

While the medical community has made an effort to disseminate informationregarding these consequences, lay individuals often construct their conceptual modelsof health phenomena based on personal experiences (Santos 2006). This reliance onpersonal experience promotes conceptual models that are founded largely on psycho-social concerns, including those that are more concrete (e.g., work and school-relateddifficulties; Jenkins et al. 1992; Miller et al. 2007) or abstract in nature (e.g., religiousissues, financial concerns; Kendler et al. 1997; San Jose et al. 2000). These psycho-social concerns are common and therefore familiar (Kleinman 1988). This personal,experiential knowledge provides the feeling of understanding and expertise in healthmatters without formal medical training (i.e., ‘lay expertise’) (Prior 2003). Becauseexisting knowledge provides a foundation for the apprehension of new information(Greeno 1983), psychosocial experience grounds the apprehension of incoming bio-medical information. Consequently, how effectively an individual integrates incomingbiomedical information depends largely on how well that biomedical informationaligns with existing conceptual models (Lakoff 2006), in this case, psychosociallyoriented. Challenges with integrating and understanding biomedical information mayarise because of the imperceptible, complex nature of some biomedical phenomena(Hmelo et al. 2000).

This analysis suggests a lack of compatibility between biomedical information andlay individuals’ psychosocially oriented models. One aspect of biomedical phenomenathat shares a concrete, perceptible component with some psychosocial phenomena,however, is physical symptoms. Especially in scientific fields, perceptual experiencesfacilitate the acquisition of knowledge in a domain (Kaiser et al. 1985). Lay individ-uals typically assume a direct causal link between experienced, physical symptomsand their underlying bodily state. For example, Sensky (1997) describes the propen-sity for individuals to assign causal attributions to bodily symptoms, especially incases where the symptoms are particularly frequent or salient. Although physicalsymptoms do not always mirror internal processes (e.g., hypertension), individualswill typically attend to perceptually accessible physical cues and subsequently formunderstanding about underlying biological processes (Cioffi 1991).

Narrative and Figurative Language

Narrative, or story-telling, allows an individual to organize information and make sense ofevents, often constructed around one’s personal experiences, that may be otherwise per-ceived as incoherent or uncertain. Narratives provide a means for individuals to communi-cate concrete information surrounding these events and experiences (Wertsch 2008), and toreveal their understanding of this information (Bartlett 1932). Despite the absence ofunanimous agreement on the identification of narrative, many linguists and cognitive

Int J Ment Health Addiction (2013) 11:458–473 459

scientists agree that the use of event-based language (use of causal explanations, chrono-logical sequences, affective language, and transitions) reflects narrative (Herman 2007).

Alternatively, figurative language such as idioms (non-literal expressions regarded asrepresenting an idea different from the literal, compositional meaning; “between a rock and ahard place”) and analogies (non-literal expressions where aspects from one concept supportthe understanding of another concept) also convey understanding. We examine idioms alongwith two types of analogy in this study; simile (direct comparison of two concepts, oftenusing ‘like’ or ‘as’; “busy as a bee”) and metaphor (implicit comparison between twoconcepts in which the characteristics of more familiar, target concepts are used to explainmore unfamiliar, underlying concepts; “life is a journey”). Some level of comprehension ofthe underlying concepts is needed to construct figurative language in order to extract featuresfrom the underlying concepts and map them onto the target concepts (Gentner 2002; Gickand Holyoak 1983; Lakoff and Johnson 1980). These underlying concepts are often abstractin nature and may preclude explanation in literal terms; thus, figurative language predom-inantly functions to communicate abstract concepts (Lakoff and Johnson 1980). Given theabstraction and complexity underlying many scientific principles, science educators oftenuse figurative language to elucidate fundamental concepts in domains such as physics,geoscience, and biology (Gentner 1989).

Language Patterns in Models of Alcoholism

Narrative and figurative language both function to convey understanding. However, weanticipate that lay individuals use these linguistic devices differently depending on the levelof concreteness or perceptibility of underlying content. To examine such patterns, our datareflect a cross between concreteness of content (perceptible/imperceptible) and domain(psychosocial/biomedical). Because figurative language is typically used to elucidate ab-stract concepts, individuals may be more inclined to use figurative language when underly-ing content is abstract or imperceptible, and narrative to convey understanding whenunderlying content is concrete or perceptible. Similar to biomedical phenomena, psychoso-cial phenomena can either be more concrete (e.g., work or school difficulties, disruption ofdaily or leisure activities) or abstract in nature (e.g., success and achievement, religiousissues). Based on the notion that differences in the concreteness of underlying content willdrive narrative and figurative language patterns in lay dialogue, we predicted the followingfor the effect of concreteness in each domain.

1. Concrete psychosocial content will exhibit a positive relationship with narrative and anegative relationship with figurative language because this content is both understoodand perceptible.

2. Concrete physical symptoms will exhibit language patterns consistent with concretepsychosocial content (i.e., a positive relationship with narrative and a negative rela-tionship with figurative language) because this content is similarly both understood andperceptible.

3. Abstract psychosocial content will exhibit a negative relationship with narrative and apositive relationship with figurative language because this content is both understoodand imperceptible.

4. Abstract, technical biomedical content will exhibit negative relationships with bothnarrative and figurative language due to the absence of understanding to support eithermode of expression.

460 Int J Ment Health Addiction (2013) 11:458–473

Standard methods for investigating conceptual models and understanding employinterviews, questionnaires, simulations, or naturalistic observation (Gentner 2002). Wechose to conduct examine language within online discussion forum posts in an effortto capture uninhibited conversation and to avoid the issues accompanying the afore-mentioned methods (e.g., unwanted influence due to leading questions and/or thepresence of a researcher).

Method

Data Collection

We collected all discussion threads (N=1,429) posted from January 2010 to December2011 on three heavily-trafficked forums: www.medhelp.org, www.soberrecovery.com,www.uncommonforum.com. MedHelp and SoberRecovery have sub-forums dedicatedexclusively to the topic of alcohol abuse. Because UncommonForum has a broader‘Addiction’ sub-forum, we only collected threads in this sub-forum with “Alcohol” or“Drinking” in the title. Forum posters had personal alcohol abuse issues and/or werefriends and family of alcohol abusers. Our institutional IRB determined this researchto be exempt because it involved the collection of publically accessible, archived data,and we do not report any information that could identify individuals (e.g., names,location, age).

Conversation in online forums is often influenced by previous dialogue, particularlyin health forums where participating individuals often assume that the opening forumpost is seeking their advice (Ruble 2011; Vayreda and Antaki 2009). To capture theoriginal content of a post and transcend the issue of conversational dependencies, wechose to use opening forum posts, as opposed to entire discussion threads, as the unitof analysis for this study (McDermott and Roth 1978). Starting with 1,429 openingposts, we chose to exclude posts under 200 words (n=1,009) in order to 1) reduce theunwanted effect of post length on percentage measures of linguistic content and 2)exclude posts that were less likely (due to insufficient amounts of content) to includenarrative and/or figurative language, while allowing us to retain a sample sizecomparable to other studies performing content analysis of social media communica-tions (Carr et al. 2012; Donelle and Hoffman-Goetz 2008; Klemm et al. 1998; Ravertet al. 2004). Of those remaining, we also excluded posts predominantly consisting ofoff-topic dialogue (n=109). We retained 311 opening posts.

Software and Data Preparation

To obtain quantitative information for language use in these posts, we used the textanalysis software, Linguistic Inquiry and Word Count (LIWC; Pennebaker et al.2007). Several studies have used LIWC for language analysis of online discussions(Abe 2012; DeAndrea et al. 2010; Dino et al. 2009). LIWC produces output consist-ing of word counts expressed as percentages of the total number of words in a text,thus controlling for the length of the text. Users can run the software’s defaultdictionary or customize their own dictionaries. Before importing the posts intoLIWC, we repaired misspellings in the forum posts with Microsoft Word spell check,and manually replaced abbreviations with their full expression (e.g., changing “gf” to“girlfriend”).

Int J Ment Health Addiction (2013) 11:458–473 461

General Post Characteristics

Forum Orientation Because the orientation of the forum is more meaningful than thespecific forum for the purposes of this study, we categorized posts as originating from themedically oriented site (MedHelp) or either of the non-medical sites (SoberRecovery orUncommonForum). Excluded posts (those that did not meet aforementioned length orcontent requirements) were evenly distributed across forum orientation.

Agency Two raters coded each opening post as containing either 1) a self-reported issue withalcohol or 2) no self-reported issue with alcohol (i.e., a friend or family member’s alcoholissue, general advice or information regarding alcoholism). We calculated interrater agree-ment for these judgments using Cohen’s kappa coefficient.

Post Content: Predictor Variables

We do not assume directionality or causality among the following variables. However, forthe purposes of our regression analyses, we use post content (technical biomedical content,physical symptoms, concrete psychosocial content, abstract psychosocial content) as ourpredictors, and linguistic devices (narrative and figurative language) as our outcomes.

Technical Biomedical Content We developed a customized LIWC dictionary to extract biomed-ical language specific to alcoholism from forum dialogue.We consultedmedical websites to garnerformal, alcohol-related terminology (Mayo Clinic staff 2010a, b, c; MedlinePlus 2010a, b, c;WebMD 2011a, b, c), and obtained informal, alcohol-related slang from a separate sample of 50discussion threads (taken from www.psychforums.com/alcohol-addiction). Our technical bio-medical content dictionary consisted of 321 words, including language referencingbody parts and organs, medicine, medical outcomes, physiology, hepatic terminology,diagnostic testing, and hospital terminology (e.g., “liver,” “pharmacy,” “enzymes”).

Physical Symptoms We developed a customized LIWC dictionary to extract physicalsymptoms specific to alcoholism from the posts. We consulted the aforementionedmedical websites to garner formal, alcohol-related terminology and obtained informal,symptoms-related slang from a separate sample of 50 discussion threads (taken fromwww.psychforums.com/alcohol-addiction). Our physical symptoms dictionary consistedof 160 words (e.g., aches, dizzy, shakiness).

Concrete Psychosocial Content We combined two pre-existing default LIWC (2007) dic-tionaries to capture concrete psychosocial content in the posts, leisure (229 words; e.g.,guitar, team, nap) and work (327 words; e.g., job, school, interview).

Abstract Psychosocial Content We combined four pre-existing default LIWC (2007) dic-tionaries to capture abstract psychosocial content in the posts: achievement (186 words; e.g.,capable, win, accomplish), money (173 words; e.g., owe, debt, cost), death (62 words; e.g.,fatal, grief, demise), and religion (159 words; e.g., faith, god, heaven).

Linguistic Devices: Outcome Variables

Narrative The following measurable dimensions, grounded in the linguistic anthropol-ogy literature, served to identify explicit indicators of narrative within text. We do not

462 Int J Ment Health Addiction (2013) 11:458–473

claim that these explicit measures encompass all features of narrative; rather, thesemeasures comprise a first attempt to detect instances of narrative automatically andsystematically within large datasets, supplemented with qualitative findings. To iden-tify these linguistic indicators of narrative in text using a quantitative approach, wedeveloped a narrative dictionary comprised of causal, chronological, affective, andtransition words. We created this dictionary using three default dictionaries inLIWC2007: causality (108 words; e.g., affected, caused, reason), time (239 words;e.g., frequent, nightly, whenever), and affect (915 words; e.g., afraid, confident,terrified), and a customized dictionary of transitions (91 words; e.g., but, nevertheless,anyway) (Campbell and Dowell 1997). We calculated one, overall, continuouslyvalued narrative percentage for each post by summing the z-scores of the fourdimensions.

Figurative Language Two raters independently coded instances of figurative languagewithin each post. Each rater obtained a percentage of figurative language for eachpost by dividing the number of words included in idiom, simile, or metaphor by thetotal word count for the post. Based on their ranking of posts from highest to lowestpercentage of figurative language, each rater divided the posts evenly into twocategories: high- and low-figurative language. We ascertained interrater agreement ofpost placement for the 100 highest figurative language posts and 100 lowest figurativelanguage posts after two rounds of reconciling differences and recoding. We measuredinterrater agreement using Cohen’s kappa, and used the agreed upon binary classifi-cation as an outcome in subsequent analyses.

Statistical Analyses

We conducted our analyses with PASW Statistics 18 software (IBM Corporation,Somers, NY, USA). Using an independent samples t-test, we first tested for arelationship between narrative and figurative language. We also used t-tests to exam-ine differences in post content (technical biomedical content, concrete psychosocialcontent, abstract psychosocial content, physical symptoms) and narrative between theposts originating from the medically oriented forum (Medhelp) and the non-medicalforums (SoberRecovery and UncommonForum) and between posts containing a self-reported issue versus no self-reported issue. Because we do not have equal samplesizes, we reported t-value significance aligned with Levene’s test for equality ofvariances. As figurative language is a binary variable in this study (i.e., high- andlow-figurative language groups), we tested the relationships between figurative lan-guage and general post characteristics (i.e., forum orientation and agency) usingPearson’s chi-squared test.

To test our hypotheses, we ran a linear regression model (narrative as outcome)and a binary logistic regression model (figurative language as outcome). That is,each of the outcomes was regressed onto each of the predictors (technical biomed-ical content, concrete psychosocial content, abstract psychosocial content, and phys-ical symptoms) in two separate models. Due to the potential effects of forumorientation and agency on our linguistic measures (statistics reported in Resultssection), we included these two variables in our regression models where appropri-ate. While we do not report analyses for the individual components of narrative(i.e., time, causality, affect, transitions), they appear in a separate manuscript (inpreparation).

Int J Ment Health Addiction (2013) 11:458–473 463

Results

We first tested for a relationship between narrative and figurative language. Next, we testedfor differences in linguistic measures by general post characteristics to examine any overalldifferences in post language by forum orientation (medically oriented versus non-medical)or agency (contains self-reported issue versus no self-reported issue). We proceeded tomodel the language patterns with post content using regression analyses. Qualitative obser-vations from our data supported and augmented the quantitative findings. We first presentour regression results for narrative, followed by regression results for figurative language.

Relationship Between Narrative and Figurative Language

There were no differences in narrative between the high- and low-figurative languageposts, t(198)=1.28, p=.20. In other words, the prevalence of narrative in a post isindependent of the prevalence of figurative language in that post, and vice versa.

General Post Characteristics, Post Content, and Linguistic Devices

Differences in post content and language with respect to general post characteristics (forumorientation and agency) are important in their own right, and for identifying potentialcontrols in subsequent modeling. For each general post characteristic, we examine potentialdifferences in the post content and linguistic devices. We report significant findings first,followed by null findings.

Forum Orientation Posts originating from the medically oriented forum (n=153) containedsignificantly more technical biomedical content (t(309)=3.43, p<.01) and physical symp-toms (t(309)=3.39, p<.01) than those from the non-medical forums (n=158). Additionally,use of figurative language was significantly related to forum orientation (χ2 (1)=16.84, p<.01), with posts from the non-medical forums containing more figurative language than themedically oriented forum.

There were no forum-related differences in narrative (t(309)=−.99, p=.32), con-crete psychosocial content (t(309)=−.33, p=.75), or abstract psychosocial content(t(309)=−1.24, p=.22).

Agency Two raters obtained agreement for coding agency (κ=.97). Posts containing self-reported issues (n=232) with alcoholism were higher in concrete psychosocial content(t(309)=2.72, p<.01), abstract psychosocial content (t(309)=2.39, p<.05), physical symp-toms (t(309)=3.71, p<.01) and narrative (t(309)=4.35, p<.01) than posts that did notcontain a self-reported issue (n=79).

There were no agency-related differences in figurative language (χ2 (1)=1.71, p=.19) ortechnical biomedical content (t(309)=−.19, p=.85).

Narrative as Outcome

The following statistics report the effects of each predictor in our linear regression model,with narrative as the outcome. Agency was included in this model because of its significantrelationship with narrative. As predicted, we found a positive significant effect of concretepsychosocial content on narrative. Like psychosocial content, we also found a positivesignificant effect of physical symptoms on narrative. There was no effect of abstract

464 Int J Ment Health Addiction (2013) 11:458–473

psychosocial content on narrative. Also in support of our hypothesis, we found a negativesignificant effect of technical biomedical content on narrative.

Concrete Psychosocial Content as Predictor As predicted, we found a positive, significanteffect of concrete psychosocial content (leisure and work) on narrative, controlling foragency (β=.11, t(310)=1.99, p=.048). Posts higher in concrete psychosocial content werehigher in narrative, independent of agency. The following is an example of a post high inboth concrete psychosocial content (leisure) and narrative:Excerpt 1

Last night i had far too much to drink…I feel terrible. I actually played guitar in mylocal pub for the first time but i was so drunk i couldn't remember the chords or thewords…I think maybe because i’ve spent too much time with my BF [boyfriend] hisdrinking habits [sic] have rubbed of [sic] on me…I'm scared to go back to my localnow. They all seemed to love my playing even though i was terrible and i got a bigcheer so maybe it's not all bad. I hope they give me a gig in the future and i hope ihaven’t ruined my chances.

This individual uses language indicative of time (“last night,” “first time,” “too muchtime,” “future”), affect (“I feel terrible,” “I’m scared,” “love,” “terrible,” “bad,” “hope,”“ruined”), causality (“maybe because”), and transitions (“but,” “or,” “even though,” “somaybe,” “and”) to elucidate how drinking affects musical performance.

Physical Symptoms as Predictor Similar to concrete psychosocial content, we found apositive significant effect of physical symptoms on narrative, controlling for agency(β=.26, t(310)=4.45, p<.01), as predicted. Posts high in physical symptoms were also highin narrative, independent of agency. Excerpt 2 presents an example of a post high in bothsymptoms and narrative:Excerpt 2

I despretely [sic] want to stop drinking, and I think the only way I can is anabuse[antabuse]. I am currently taking klonopin and drinking on top of it as well, and Ithink that is why [I am] blacking out so much when I drink and I am getting soviolently sick every morning. I have horrible anxiety and the klonopin makes itcompletely disappear [sic] which is a godsend…Anyway, I just wanted to ask othersabout their experiences…

This individual uses language indicative of time (“currently,” “every morning”), affect(“desperately,” “violently,” “horrible”), causality (“the only way,” “that is why,” “klonopinmakes it completely disappear”), and transitions (“and,” “which is,” “anyway”) to relaytheir physical symptoms (“blacking out,” “getting so violently sick,” “horrible anxiety”).

Abstract Psychosocial Content as Predictor In contrast to our prediction, there was nosignificant effect of abstract psychosocial content (achievement, money, death, and religion)on narrative, controlling for agency (β=.04, t(310)=.82, p=.42). In other words, the amountof abstract psychosocial content in a post did not influence the amount of narrative in thatsame post.

Technical Biomedical Content as Predictor As predicted, we found a negative, significanteffect of technical biomedical content on narrative, controlling for agency (β=−.25,t(310)=−4.50, p<.01). In other words, posts higher in technical biomedical content were

Int J Ment Health Addiction (2013) 11:458–473 465

lower in narrative, despite the prevalence of narrative in self-reported posts. Excerpt 3presents an example of a post high in technical biomedical content and low in narrative:Excerpt 3

… I have are a few tiny spider angiomas and my doctor says that those are pregnancyrelated…I have convinced myself that I have alcoholic liver disease and want to get abiopsy done but the gastro said he wouldn't do it…Is it possible to have serious liverdamage with a normal ultrasound, cat scan and blood tests?

This individual does not use language indicative of time or affect to communicatemedical experience. The discussion of biomedical concepts (“spider angiomas,” “doctor,”“pregnancy,” “alcoholic liver disease,” “biopsy,” “gastro,” “liver damage,” “ultra-sound,” “catscan,” “blood tests”) is much more factual and explanatory than narrative.While the individual’s comment about spider angiomas as pregnancy-related does implycausality, we note that this speculation seems to be a recitation of what the doctor suggested(“my doctor says”), and not the individual’s personal inference. Also, while there are a fewinstances of transitions (“and,” “but”), transitions alone are not suggestive of narrative.

Despite their scarcity, posts high in technical biomedical content and also high innarrative demonstrate that it is possible for individuals to communicate technical biomedicalcontent via narrative. Excerpt 4 presents an example:Excerpt 4

9 months ago i was diagnosed to be in the beginning [sic] of the final stage ofCirrhosis. After finding a great Doctor to help me and communicating with himhonestly I have made a most amazing recovery. My amonia (ammonia) level is undercontrol thanks to a fairly new pill called Xifaxan…The rest of my levels are all in thegood!Wow, I thank God! I don’t hear many recovery stories as successful [sic] as minealthough I sure wish I did. A couple of days ago I had a pretty scary thought, what if iwould never stopped drinking.

Albeit unusual, this post contains a substantial amount of technical biomedical content(“diagnosed,” “cirrhosis,” “doctor,” “ammonia level,” “pill,” “platelet count,” “biliru-bin”) while also containing several components of narrative, such as causality (“undercontrol thanks to”), time (“9 months ago,” “in the beginning,” “after,” “couple of daysago,” “never”), affect (“great,” “honestly,” “amazing,” “good,” “wow,” “thank,” “suc-cessful,” “scary”), and transitions (“as,” “and,” “if,” “although”).

Figurative Language as Outcome

Two coders obtained a Cohen’s kappa of .62 for ranking 100 posts into a high-figurativelanguage group and 100 posts into a low-figurative language group. Cohen’s kappa valuesranging from .60 to .80 are considered to reflect ‘good level of agreement’ (Altman 1991),thus we found .62 to be acceptable for the purposes of this study.

The following statistics report the effects of each predictor in our binary logistic regres-sion model, with figurative language as the outcome. Forum orientation was included in thismodel because of its significant relationship with figurative language. As predicted, wefound a negative significant effect of concrete psychosocial content on figurative language.Like psychosocial content, we also found a negative significant effect of physical symptomson figurative language. Also in support of our hypotheses, we found a positive significanteffect of abstract psychosocial content and a negative significant effect of technical biomed-ical content on figurative language.

466 Int J Ment Health Addiction (2013) 11:458–473

Concrete Psychosocial Content as Predictor As predicted, we found a negative significanteffect of concrete psychosocial content on figurative language, controlling for forum orien-tation (Wald Chi-Square=8.20, p<.01). Posts higher in concrete psychosocial contentcontained less figurative language, independent of forum orientation. The following is anexample of an individual explaining how his job as a prison guard contributed to hisdrinking problem, absent figurative language:Excerpt 5

I am trying to quit drinking and need some advice on how to quit. I never had aproblem in the past until i had a stressful job. I worked as a prison guard. I experiencedmy first anxiety attack at that job. But then was doing like everyone else was, gettingoff work and drinking. So started drinking a lot.

Physical Symptoms as Predictor Similar to concrete psychosocial content, there was asignificant negative effect of physical symptoms on figurative language, controlling forforum orientation (Wald Chi-Square=11.19, p<.01), as predicted. In other words, postshigher in physical symptoms were also lower in figurative language, independent of forumorientation. The following example illustrates the use of literal language for discussingphysical symptoms:Excerpt 6

I spent Sat. night sober (with insomnia and night sweats) and all of Sunday. I foundthat when I got off of work this morning at around 3 am that I was very much awakeand upon trying to sleep at around 5 am that I was very anxious, cold but sweaty,shaking violently, and even possible mild hallucinations.

Abstract Psychosocial Content as Predictor As predicted, we found a positive significanteffect of abstract psychosocial content on figurative language, controlling for forum orien-tation (Wald Chi-Square=14.91, p<.01). That is, posts higher in abstract psychosocialcontent contained more figurative language, independent of forum orientation. The follow-ing is an example of an individual using figurative language, comparing God’s healingprocess to a Band-Aid being ripped off of one’s skin:Excerpt 7

It was like mom pulling off a bandaid. She would pull and the skin and glue wouldpull along with it until finally it was ripped off. There would be pain during theremoval but in the end there would be relief. I guess my point is that sometimes ourdefects are not instantly removed and that we have to do our part by not holding onwhile God removes them from us.

Technical Biomedical Content as Predictor As predicted, we found a negative significanteffect of technical biomedical content on figurative language, controlling for forum orien-tation (Wald Chi-Square=7.38, p<.01). Posts higher in technical biomedical contentcontained less figurative language, independent of forum orientation. The following is anexample of a post containing high technical biomedical content, absent figurative language:Excerpt 8

Going to the doctor next week to get this dizziness looked at further. Had some bloodwork done already and everything looks pretty good. Might be a probem [problem] in

Int J Ment Health Addiction (2013) 11:458–473 467

my spine or my ears. Is it bad that I'm actually HOPING it's one of those two thingsand not a liver or heart disease…

Here the individual uses literal language to communicate biomedical concepts (“doctor,”“blood work,” “spine,” “ears,” “liver or heart disease”).

As with narrative, we did find posts that were high in both technical biomedical contentand figurative language, despite their scarcity:Excerpt 9

The fat from the ethanol is stored right in the liver…causing the liver to have anappearance like marble cheese, with little fat deposits here and there. As these fatdeposits are broken down, your liver appearance becomes much like swiss cheese,little holes are here and there.

Here, the individual uses two analogies to communicate complex biomedical con-cepts by relating the appearance of the liver in its different stages to the appearancesof a presumably more familiar concept, cheese. The individual’s understanding ofliver functioning (and of cheese) supports the creation of these analogies, therebycommunicating a potentially unfamiliar, imperceptible concept (liver functioning)through a more familiar, concrete concept (cheese).

Discussion

The goal of this study was to examine language use patterns related to the biomedical andpsychosocial domains associated with alcoholism, each of which includes concrete andabstract content. Narrative and figurative language are both linguistic devices conveyingunderstanding, sense-making, and coherence; all of which are critical elements that serve toguide subsequent reasoning, emotions, and actions when faced with a complex issue such asalcoholism. However, we found that narrative and figurative language are statisticallyindependent. Moreover, a dissociation between the patterns of narrative and figurativelanguage with content implies separate mechanisms for expressing understanding.

Differences in the linguistic patterns individuals use to convey biomedical versuspsychosocial content do not appear to reflect inherent differences in the nature ofbiomedical and psychosocial concepts. Our findings support a more nuanced interpre-tation. Individuals’ use of narrative and figurative language appears to be driven bythe concreteness or perceptibility of the underlying content. Individuals constructednarratives to communicate their understanding of concrete concepts, in this case bothphysical symptoms and concrete psychosocial phenomena. They typically do not usefigurative language for concrete concepts.

However, individuals treat abstract concepts differently, and in a manner thatdepends upon the domain. Individuals tended to communicate abstract psychosocialconcepts through figurative language as opposed to narrative. They generally did notuse either narrative or figurative language to communicate technical biomedicalcontent, potentially because of the absence of sufficient knowledge to support thesedevices. We proceed to discuss in detail the effects of content concreteness anddomain on language usage, and suggest that the similar conceptual status of concretepsychosocial experience and physical symptoms provides a bridge between psychoso-cial and biomedical domains.

468 Int J Ment Health Addiction (2013) 11:458–473

Concrete Psychosocial Content

Just as there are differences in concreteness among concepts within the biomedical model,these differences also exist within the psychosocial model. Psychosocial concepts related tohobbies, schoolwork, and jobs are concrete, familiar, and encompass many perceptiblecomponents (e.g., guitar, homework, coworkers). Because of this perceptibility, we expectedindividuals to convey their understanding of concrete psychosocial content through narrativeas opposed to figurative language. We found support for our hypotheses regarding theincreased use of narrative and decreased use of figurative language to communicate theseconcrete psychosocial concepts. While individuals may not find it necessary to explainconcrete concepts in terms of figurative language, a device which typically functions toelucidate abstract concepts, the perceptible nature of concrete psychosocial phenomenalends itself to a narrative format (e.g., why the phenomenon occurred, when the phenomenonoccurred, how it feels to experience the phenomenon, the sequence of events leading up tothe phenomenon).

Physical Symptoms

Following the linguistic patterns of concrete psychosocial content, we found similar resultsfor concrete, physical symptoms. As predicted, individuals used more narrative and lessfigurative language to convey physical symptoms. Although physical symptoms belong tothe biomedical model, they differ from other biomedical content because of their concrete,perceptible nature. Much of how individuals form their conceptual understanding isgrounded in their perceptual experiences (Kaiser et al. 1985). This experience-based under-standing, garnered from the perception of physical symptoms, manifested itself in the samelanguage used to convey concrete psychosocial content. Much like hobbies, school, andwork, the perceptible nature of physical symptoms may facilitate their integration intonarrative. However, it is this concreteness and perceptibility that may also preclude the needfor explanation via figurative language.

Abstract Psychosocial Content

Psychosocial phenomena may also be abstract in nature. Relative to the activities thatpervade concrete psychosocial content, concepts surrounding achievement, religion, death,and finances, though frequent and pervasive in our lives, are often less directly perceptiblewith more socially constructed meaning. Because of this, we expected individuals to conveytheir understanding of abstract psychosocial content through figurative language as opposedto narrative. While the absence of an effect of abstract social content on narrative did notsupport our prediction regarding a decrease, the data did support our prediction regarding theincreased use of figurative language to convey abstract psychosocial content. These findingsalign with the function of figurative language as a means for communicating abstractconcepts by relating them to more concrete, familiar concepts (Lakoff and Johnson 1980).

The absence of a relationship between abstract psychosocial content and narrative mayreflect a more versatile function of narrative than predicted. Individuals did occasionally usecausal and chronological language, affect, and transitions to convey understanding ofabstract psychosocial content. Our findings regarding the communication of psychosocialcontent reinforce our claim that while narrative and figurative language may both indicateunderstanding, the dimension of concreteness of underlying content appears to drive the useof one linguistic device over another.

Int J Ment Health Addiction (2013) 11:458–473 469

Technical Biomedical Content

Our data supported our hypotheses regarding the general absence of both narrative andfigurative language to relay technical biomedical content (e.g., organ damage, physiology,medicine). Excerpts 3 and 8 demonstrate that individuals communicated technical biomed-ical content using more literal language. The lack of narrative and figurative language toconvey technical biomedical content potentially reflects its imperceptible nature, which mayimpede the acquisition of experiential knowledge and understanding (Barsalou 2008).

The specialized, technical nature of biomedical concepts potentially hinders the integra-tion of biomedical language into narrative or figurative language. However, exceptions tothe rule found in this sample (Excerpts 4 and 9) demonstrate that, although uncommon, it isindeed possible to relay biomedical information through both narrative and figurativelanguage. Although this may simply reflect imitation of medical sources, individualspossessing understanding of imperceptible, technical biomedical concepts appear capableof integrating that information into narrative and figurative language.

In summary, concreteness rather than domain provides a better explanation for the use ofnarrative and figurative language. Individuals tended to use narrative to convey theirunderstanding of concrete concepts, and figurative language to convey their understandingof abstract concepts. The accessibility of these linguistic devices for psychosocial conceptssupports the lay individual’s psychosocially-oriented model of medical issues (Santos 2006).While lay individuals may use narrative for concrete physical symptoms, they generally lackthe conceptual foundation to articulate technical biomedical concepts using figurativelanguage. The combination of familiar psychosocial content and accessible physical symp-toms results in ‘lay expertise’ (Prior 2003).

Limitations

One criticism of this study involves the use of quantitative methods to analyze language.Human discourse is largely contextual and often contains implicit meaning. Our key-wordbased, quantitative approach cannot fully capture meaning, and we acknowledge the benefitsof combining qualitative and quantitative analysis. Nevertheless, quantitative analyses allowus to systematically identify instances of language in large amounts of text, and to drawpatterns from the data that may not be easily detected with qualitative content analysis.Alternatively, qualitative analyses can be used to illustrate these patterns and identify contextand meaning that is implicit and thus superficially unrecognizable.

Also, quantitative methods limit our extraction of terms to finite dictionaries. Therefore, itis possible to miss terms in the text that should have contributed to our percentages (i.e.,error of omission). However, the potential to make the complementary error (i.e., error ofcommission) also exists. That is, terms may be identified by the software as belonging tocertain category when, contextually, the term should not have been categorized as such (e.g.,“heart” in “he broke my heart” was inappropriately categorized as biomedical content).While the presence of both of these errors may have artificially inflated the variability of ourdata (i.e., data are noisier), their complementary nature suggests that they present no seriousthreat to our conclusions. While the minimization of these errors would result in morereliable data, it is unlikely that it would affect the direction of our findings (and may actuallystrengthen our conclusions).

Lastly, the understanding reflected by narrative and figurative language may or may notbe accurate understanding. While lay expertise often leads an individual to believe theypossess conceptual understanding, that understanding may potentially contain

470 Int J Ment Health Addiction (2013) 11:458–473

misconceptions or inaccuracies (Feltovich et al. 1989). Future research should examine thescientific accuracy of biomedical concepts integrated into narrative and figurative languageto determine the extent to which individuals use these linguistic devices to convey accurateunderstanding.

Practical Implications

Analyses of individuals’ language provide insight into their conceptual models andunderstanding (Carley and Palmquist 1992). Knowledge of these lay conceptualmodels might allow researchers and clinicians to better understand how individualsunderstand and communicate biomedical information. Research has demonstrated thatpatient comprehension of biomedical information, particularly in regards to riskinformation, has a positive impact on medical decision-making (Natter and Berry2005; Spandorfer et al. 1995). However, simply laying new, potentially disparatebiomedical information onto individuals’ existing conceptual models is likely inade-quate. Incoming information requires coherence, correspondence, and connectednesswith individuals’ pre-existing models and understanding (Greeno 1983).

Experiential knowledge may more readily contribute to conceptual understandingbecause experienced, perceptible phenomena are actually situated within the contextthat one comes to understand (Greeno 1998). Patient education systems that caneffectively integrate biomedical knowledge into current lay models may leverage offof the salience and experiential influence of perceptible physical symptoms andconcrete psychosocial phenomena.

Conclusion

Our findings demonstrate an association between language patterns, concreteness ofunderlying content, and domain, with implications for comprehension. However, con-creteness rather than domain per se appears to govern language usage. Accessiblepsychosocial phenomena correlate with the employment of narrative for concrete contentand figurative language for abstract psychosocial content. The treatment of biomedicalphenomena also reflects concreteness. Concrete, physical symptoms appear with narra-tive, just as concrete psychosocial phenomena. However, more abstract biomedicalcontent lacks either narrative or figurative devices in lay language. The absence isnot a fundamental property of the domain, as shown through instances in whichindividuals used these devices to communicate biomedical concepts. Instead, the patternof language use reflects the availability of knowledge.

References

Abe, J. A. (2012). Cognitive–affective styles associated with position on war. Journal of Language and SocialPsychology, 31, 212–222.

Altman, D. G. (1991). Practical statistics for medical research. London: Chapman and Hall.Barclay, G. A., Barvour, J., Stewart, S., Day, C. P., & Gilvarry, E. (2008). Adverse physical effects of alcohol

misuse. Advances in Psychiatric Treatment, 14, 139–151.Barsalou, L. W. (2008). Grounded cognition. Annual Review of Psychology, 59, 617–645.Bartlett, F. C. (1932). Remembering: A study in experimental and social psychology. Cambridge: Cambridge

Univ. Press.

Int J Ment Health Addiction (2013) 11:458–473 471

Campbell, G. M. & Dowell, J. A. (1997). Transition words. Michigan State University. Retrieved from https://www.msu.edu/~jdowell/135/transw.html.

Carley, K., & Palmquist, M. (1992). Extracting, representing, and analyzing mental models. Social Forces, 70,601–636.

Carr, C. T., Schrock, D. B., & Dauterman, P. (2012). Speech acts within Facebook status messages. Journal ofLanguage and Social Psychology, 31, 176–196.

Cioffi, D. (1991). Beyond attentional strategies: a cognitive–perceptual model of somatic interpretation.Psychological Bulletin, 109, 25–41.

DeAndrea, D. C., Shaw, A. S., & Levine, T. R. (2010). Online language: the role of culture in self-expressionand self-construal on facebook. Journal of Language and Social Psychology, 29, 425–442.

Dino, A., Reyson, S., & Branscombe, N. R. (2009). Online interactions between group members who differ instatus. Journal of Language and Social Psychology, 28, 85–93.

Donelle, L., & Hoffman-Goetz, L. (2008). An exploratory study of Canadian Aboriginal online health careforums. Health Communication, 23, 270–281.

Feltovich, P. J., Spiro, R. J., & Coulson, R. L. (1989). The nature of conceptual understanding in biomedicine:the deep structure of complex ideas and the development of misconceptions. In D. Evans & V. Patel(Eds.), Cognitive science in medicine: Biomedical modeling. Cambridge: MIT (Bradford) Press.

Gentner, D. (1989). The mechanisms of analogical learning. In S. Vosniadou & A. Ortony (Eds.), Similarityand analogical reasoning (pp. 199–241). Cambridge: Cambridge Press.

Gentner, D. (2002). Psychology of mental models. In N. J. Smelser & P. B. Bates (Eds.), Internationalencyclopedia of the social and behavioral sciences (pp. 9683–9687). Amsterdam: Elsevier Science.

Gick, M. L., & Holyoak, K. J. (1983). Schema induction and analogical transfer. Cognitive Psychology, 15, 1–38.

Greeno, J. G. (1983). Forms of understanding in mathematical problem solving. In S. G. Paris, G. M. Olson, &H. W. Stevenson (Eds.), Learning and motivation in the classroom (pp. 83–111). Hillsdale: LawrenceErlbaum Associates, Inc.

Greeno, J. G. (1998). The situativity of knowing, learning and research. American Psychologist, 53, 5–26.Herman, D. (2007). Introduction. In D. Herman (Ed.), The Cambridge companion to narrative (pp. 3–21).

Cambridge: Cambridge Univ. Press.Hmelo, C. E., Holton, D., & Kolodner, J. L. (2000). Designing to learn about complex systems. The Journal of

the Learning Sciences, 9, 247–298.Jenkins, R., Harvey, S., Butler, T., & Thomas, R. L. (1992). A six year longitudinal study of the occupational

consequences of drinking over “safe limits” of alcohol. British Journal of Industrial Medicine, 49, 369–374.

Kaiser, M. K., Proffitt, D. R., & Anderson, K. (1985). Judgments of natural and anomalous trajectories in thepresence and absence of motion. Journal of Experimental Psychology: Learning, Memory, andCognition, 11, 795–803.

Kendler, K. S., Gardner, C. O., & Prescott, C. A. (1997). Religion, psychopathology, and substance use andabuse: a multimeasure, genetic-epidemiologic study. The American Journal of Psychiatry, 154, 322–329.

Kleinman, A. (1988). The illness narratives: Suffering, healing and the human condition. New York: BasicBooks.

Klemm, P., Reppert, K., & Visich, L. (1998). A nontraditional cancer support group. The Internet. Computersin Nursing, 16, 31–36.

Lakoff, G. (2006). Whose freedom? The battle over America’s most important idea. New York: Farrar, Strausand Giroux.

Lakoff, G., & Johnson, M. (1980). Metaphors we live by. Chicago: Univ. Chicago Press.Mayo Clinic staff (2010a). Alcoholism: complications. Mayo Clinic. Retrieved from http://www.mayoclinic.com/

health/alcoholism/DS00340/DSECTION=complications.Mayo Clinic staff (2010b). Alcoholism: symptoms. Mayo Clinic. Retrieved from http://www.mayoclinic.com/

health/alcoholism/DS00340/DSECTION=symptoms.Mayo Clinic staff (2010c). Alcoholism: treatment and drugs. Mayo Clinic. Retrieved from http://www.mayoclinic.

com/health/alcoholism/DS00340/DSECTION=treatments-and-drugs.McDermott, R. P., & Roth, D. R. (1978). The social organization of behavior: interactional approaches.

Annual Review of Anthropology, 7, 321–345.MedlinePlus (2010a). Alcoholic liver disease. National Library of Medicine: National Institutes of Health.

Retrieved from http://www.nlm.nih.gov/medlineplus/ency/article/000281.htm.MedlinePlus (2010b). Alcoholism and alcohol abuse. National Library of Medicine: National Institutes of

Health. Retrieved from http://www.nlm.nih.gov/medlineplus/ency/article/000944.htm.MedlinePlus (2010c). Cirrhosis. National Library of Medicine: National Institutes of Health. Retrieved from

http://www.nlm.nih.gov/medlineplus/ency/article/000255.htm.

472 Int J Ment Health Addiction (2013) 11:458–473

Miller, J. W., Naimi, T. S., Brewer, R. D., & Jones, S. E. (2007). Binge drinking and associated health riskbehaviors among high school students. Pediatrics, 119, 76–85.

Natter, H. M., & Berry, D. C. (2005). Effects of active information processing on the understanding of riskinformation. Applied Cognitive Psychology, 19, 123–135.

Ochs, E., & Capps, C. (1996). Narrating the self. Annual Review of Anthropology, 25, 19–43.Pennebaker, J. W., & Seagal, J. D. (1999). Forming a story: the health benefits of narrative. Journal of Clinical

Psychology, 55, 1243–1254.Pennebaker, J. W., Booth, M. E., & Francis, R. J. (2007). Linguistic inquiry and word count [computer

software]. Austin: LIWC.Prior, L. (2003). Belief, knowledge and expertise: the emergence of the lay expert in medical sociology.

Sociology of Health & Illness, 25, 41–57.Ravert, R. D., Hancock, M. D., & Ingersoll, G. M. (2004). Online forum messages posted by adolescents with

type 1 diabetes. The Diabetes Educator, 5, 827–834.Ruble, R. A. (2011). The communication of advice on an online message board for language assistants in

France. Journal of Language and Social Psychology, 30, 396–420.San Jose, B., van Oers, H. A. M., van de Mheen, H. D., Garretsen, H. F. L., & Mackenbach, J. P. (2000).

Stressors and alcohol consumption. Alcohol and Alcoholism, 35, 307–312.Santos, S. (2006). The diversity of everyday ideas about inherited disorders. Public Understanding of Science,

15, 259–275.Sensky, T. (1997). Causal attributions in physical illness. Journal of Psychosomatic Research, 43, 565–573.Sontag, S. (1978). Illness as metaphor. New York: McGraw-Hill.Spandorfer, J., Karras, D., Hughes, L., & Caputo, C. (1995). Comprehension of discharge instructions by

patients in an urban emergency department. Annals of Emergency Medicine, 25, 71–74.Vayreda, A., & Antaki, C. (2009). Social support and unsolicited advice in a bipolar disorder online forum.

Qualitative Health Research, 19, 931–942.WebMD (2011a). Alcohol abuse and dependence: medications. WebMD. Retrieved from http://www.webmd.

com/mental-health/alcohol-abuse/alcohol-abuse-and-dependence-medications.WebMD (2011b). Alcohol abuse and dependence: symptoms. WebMD. Retrieved from http://www.webmd.

com/mental-health/alcohol-abuse/alcohol-abuse-and-dependence-symptoms.WebMD (2011c). Alcohol abuse and dependence: what happens. WebMD. Retrieved from http://

www.webmd.com/mental-health/alcohol-abuse/alcohol-abuse-and-dependence-what-happens.Wertsch, J. V. (2008). Collective memory and narrative templates. Social Research, 75, 133–156.

Int J Ment Health Addiction (2013) 11:458–473 473