Can you spot anxious thinking not in the words someone uses, but in how their sentences link cause to conclusion?
How do cognitive distortions appear as patterns in how people structure sentences?
This explores whether distorted thinking, such as jumping from one bad event to a sweeping conclusion, leaves a detectable pattern in how claims are linked together in what people say or write, rather than in which words they pick.
This explores whether distorted thinking, such as jumping from one bad event to a sweeping conclusion, leaves a detectable pattern in how claims are linked together in what people say or write, rather than in which words they pick. The corpus has one strong result on this and a few adjacent ones. It has nothing on grammar or syntax as such.
The strongest evidence is about anxiety. Causal explanations that run across statements ('this went wrong, so I always fail') predict anxiety better than any individual word does. The stated reason is that anxious thinking involves overgeneralization through reasoning between statements, not through vocabulary. A model that combines word-level and discourse-level signals beats either alone (Why do discourse patterns predict anxiety better than single words?). So the pattern sits in the joints between sentences: how one claim is made to justify the next.
A second line of work treats distortion detection as a reasoning task instead of a keyword hunt. Structured three-stage prompting improves detection by over ten percent compared with zero-shot ChatGPT. It works by separating three questions: is this statement a subjective interpretation, how does it contrast with alternatives, and what underlying schema does it fit. Clinical experts rated the resulting explanations as useful for case formulation (Can structured prompting improve cognitive distortion detection?). That decomposition suggests a distortion is a relationship between a claim and the evidence or alternatives around it, so it can't be read off the claim alone.
One adjacent finding fits this picture. Humans judge arguments partly by whether they like the conclusion, and language models reproduce that content-sensitivity pattern item by item across syllogisms, natural language inference and the Wason task (Do language models show the same content effects humans do?). This is a study of reasoning errors, not of clinical distortions, so treat it as a parallel. It does show that content and logical form get tangled together in ordinary thinking, which is why a distorted argument can read as sound.
What the corpus doesn't have is a catalogue of specific structural signatures, such as 'always' and 'never' phrasing or the shape of a catastrophizing chain. It also has no work on distortions in spoken or informal language. The evidence points to inter-statement reasoning as the place to look, and it does not yet say which patterns mark which distortion.
Sources 3 notes
Causal explanations across statements—not individual words—are the strongest predictor of anxiety because anxious thinking involves overgeneralization through inter-statement reasoning. A dual model combining both representation levels outperforms either alone.
DoT prompting separates subjectivity assessment, contrastive reasoning, and schema analysis to achieve 10%+ improvement over zero-shot ChatGPT. Expert evaluators rated the resulting explanations as clinically useful for case formulation.
LLMs show identical content-sensitivity patterns to humans on NLI, syllogisms, and Wason tasks, with belief-bias signatures matching human error rates item-by-item. This behavioral isomorphism across three independent tasks suggests content and logical form are inseparable in transformer reasoning architecturally.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Empowering Psychotherapy with Large Language Models: Cognitive Distortion Detection through Diagnosis of Thought Prompting
- Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens
- Language models show human-like content effects on reasoning tasks
- Discourse-Level Representations can Improve Prediction of Degree of Anxiety
- Beyond Accuracy: Evaluating the Reasoning Behavior of Large Language Models -- A Survey
- Premise Order Matters in Reasoning with Large Language Models
- Large Language Models Can Infer Psychological Dispositions of Social Media Users
- Mind Your Step (by Step): Chain-of-Thought can Reduce Performance on Tasks where Thinking Makes Humans Worse