Analyzing Appraisal in Major and Bipolar Depression Patients’ Narratives
A corpus-based study of how writers with self-reported major depressive disorder and bipolar disorder express emotion, judge people and behavior, and evaluate treatment, illness, and lived experience.
Purpose and argument
The thesis investigates evaluation in narratives posted to mental-health forums by people who self-reported major depressive disorder (MDD) or bipolar disorder (BD). It asks which forms of Attitude dominate, which emotions recur, how writers assess themselves and others, and which aspects of illness and treatment they evaluate.
The analysis combines Appraisal Theory with corpus methods. Rather than retrieving every adjective and deciding afterward whether it carries evaluation, the study uses Corpus Query Language patterns derived from “syntactic fields of adjectives” to locate constructions likely to realize Affect, Judgment, or Appreciation. The retrieved examples are then interpreted in context, retaining the qualitative distinctions of Appraisal Theory while scaling the search to millions of words.
Corpus design
Narratives were collected from public mental-health forums, deduplicated, cleaned of HTML, and compiled as separate MDD and BD corpora. BootCat supported corpus construction; Sketch Engine provided part-of-speech tagging, concordancing, frequency information, and CQL retrieval. The corpus is used to study language patterns, not to infer diagnoses from text.
Analytical framework
Affect
Emotional states and reactions, including pleasure, misery, fear, confidence, and disquiet.
Judgment
Evaluation of people and behavior, especially capacity and propriety.
Appreciation
Evaluation of things, events, treatment, and illness through complexity, valuation, and quality.
CQL searches targeted patterns such as adjective + to + verb, adjective + for, adjective + to, and forms of be + adjective + to. Concordance lines were then annotated for Appraisal category, polarity, subcategory, and discourse context.
Central findings
- Judgment dominated both corpora. Writers most often evaluated their own and others' capacity and propriety, connecting mental-health experience to competence, responsibility, support, and stigma.
- The second-ranked system differed. Appreciation was more frequent than Affect in MDD narratives; Affect was slightly more frequent than Appreciation in BD narratives.
- Emotional profiles converged. Pleasure, misery, and fear were the three most frequent emotions in both corpora. Disquiet followed in MDD; confidence followed in BD.
- Evaluated phenomena were concrete. Complexity, valuation, and quality captured recurring assessments of illness, medication, relationships, work, study, and treatment.
- Syntactic retrieval was productive but not automatic annotation. Across both corpora, 76% of retrieved instances were attitudinal evaluations and 24% were not applicable, showing that syntax efficiently narrows the search while contextual human analysis remains necessary.
Key results visualized
| Rank | MDD narratives | BD narratives |
|---|---|---|
| 1 | Judgment | Judgment |
| 2 | Appreciation | Affect |
| 3 | Affect | Appreciation |
Implications, limitations, and future work
The thesis argues that narrative language can help reveal concerns around stigma, capacity, relationships, medication, work, and access to support. It recommends mental-health communication that treats illness as manageable, involves families where appropriate, confronts stigma in schools and workplaces, and recognizes expressive writing as potentially valuable alongside clinical care.
Its limitations include self-reported diagnoses, unknown demographic balance, forum-specific selection effects, and the interpretive demands of Appraisal annotation. The findings describe these corpora and must not be used to diagnose individuals. Future research could test the patterns across languages, genders, platforms, and clinically verified datasets, with multiple annotators and computational models evaluated against the thesis's fine-grained lexical resources.