LSD and language: Decreased structural connectivity, increased semantic similarity, changed vocabulary in healthy individuals

Language provides a window into thought by revealing both form (structure, connectivity) and content (semantics, vocabulary). Recent computational approaches have linked measures of speech connectedness to cognitive performance and psychiatric states, and semantic and lexical analyses have been used to detect dreamlike, mystical or disordered thought. Prior reports—ranging from case studies to recent computerised analyses—suggest that serotonergic psychedelics (including LSD) alter speech, producing simpler or disorganised structure, shifts in semantics towards dreamlike or mystical content, and greater use of concrete or emotion-laden vocabulary. However, earlier work has been limited by retrospective designs, single time points, or subjective/manual analyses, and a systematic, time-resolved quantification of language under psychedelics is lacking. Wießner and colleagues set out to map how a low-to-moderate dose of LSD (50 μg) affects speech structure, semantics and vocabulary over the acute and sub-acute phases. Using a randomised, double-blind, placebo-controlled crossover design in healthy volunteers, the study assessed two production types (experience reporting and storytelling), multiple analytic components (speech topology via Speech-Graphs, semantic distances via FastText, and word categories via LIWC) and several time points up to +24 hours, with the aim of characterising time-dependent changes in language during and after LSD effects.

The study used a randomised, double-blind, placebo-controlled crossover design. Twenty-five healthy adults were recruited and screened; one dropped out, leaving 24 participants (mean age 35 ± 11 years, range 25–61; 33% women). Inclusion required prior LSD experience and drug/alcohol abstinence windows; exclusion criteria included personal or first-degree family history of psychotic disorders, current psychiatric symptoms, psychiatric medication use, substance use disorder, severe prior complications after psychedelics, relevant medical conditions, pregnancy, and non-native Brazilian Portuguese speakers. Sessions were separated by 14 days and each participant received 50 μg oral LSD in one session and an inactive placebo in the other, with session order counterbalanced. Procedures combined structured experience-report interviews and short storytelling tasks at scheduled time points. Baseline storytelling occurred at +0 h (9:00 a.m.). Drug/placebo was administered at 9:30 a.m. Experience reporting was sampled at +1.5 h (a 5-minute interview; note this measure was only obtained from the sixth participant onward, so n = 19 for that time point) and at +6.5 h (first 5 minutes of a 30-minute semi-structured interview). Storytelling was recorded at +0 h, +2 h, +4 h, +6 h and +24 h. Storytelling stimuli comprised four stimulus types presented in a fixed order: positive (pleasant photos), neutral (IAPS photos), ambiguous (black-and-white Thematic Apperception Test illustrations) and creative (written impossible-scenario prompts). Two parallel stimulus versions were counterbalanced across sessions. Language was analysed along three complementary dimensions. Structure was quantified using Speech-Graphs (word-trajectory graphs where words are nodes and successive-word links are edges), yielding metrics such as Word Count (WC), Nodes, Edges, Average Total Degree (ATD), Largest Connected Component (LCC), Largest Strongly Connected Component (LSC), Density, Average Shortest Path (ASP), Clustering Coefficient (CC) and related measures. Semantics were assessed via semantic distances computed with FastText embeddings trained on a Portuguese Wikipedia corpus; extracted variables included Forward Flow, Flow Steps and Semantic Spread, which index semantic distance and diversity between consecutive or overall words. Vocabulary was measured by proportional word-category sizes using LIWC (Brazilian Portuguese dictionary) across grammatical and psychological categories; word-frequency word clouds were produced for qualitative inspection. Pre-/post-processing details and supplementary variables are reported in supplemental methods (not reproduced here). Statistical analysis used repeated-measures General Linear Models (GLMs) in SPSS. For experience reporting, within-subject factors were treatment (LSD, placebo), time point (+1.5 h, +6.5 h) and variable set (structure, semantics, vocabulary), with treatment order (LSD–placebo, placebo–LSD) as a between-subjects factor. For each storytelling stimulus, analogous GLMs tested treatment, time point (+0 h, +2 h, +4 h, +6 h, +24 h) and variable, again including treatment order. Main effects and interactions (including treatment*time point*variable) were examined; given the exploratory aims, pairwise comparisons were performed for significant effects. Effect sizes were reported as partial eta squared (ηp2). Significance was set at α = 0.05 and post-hoc Bonferroni corrections were applied for the number of time points (experience reporting = 2, storytelling = 5).

Overall, Wießner and colleagues report that a single 50 μg LSD dose altered speech structure, semantics and vocabulary in time- and stimulus-dependent ways, with effects observed during acute and into sub-acute phases (up to +24 h). Results are presented separately for structure, semantics and vocabulary. Structure: In the 5-minute experience-report transcripts there were no significant treatment, period, order or interaction effects. Shorter (1-minute) analyses of experience reports did reveal reduced global (Average Total Degree) and local connectivity (e.g. Repeated and Parallel Edges) under LSD, indicating early changes at the start of reports (this 1-minute analysis had reduced sample size for +1.5 h as noted). For storytelling, significant main treatment effects emerged for positive, neutral and ambiguous stimuli, with lower structure-related metric values under LSD; order effects also appeared (lower values for the LSD–placebo order), suggesting carryover. Pairwise comparisons at +2 h under LSD showed decreased Word Count, Nodes, Edges, ATD, LCC, LSC and Parallel Edges and increased Density for positive stimuli; neutral stimuli at +2 h showed decreased WC, Nodes, Edges, LCC and LSC. Ambiguous stimuli showed decreased WC, Nodes, Edges, LCC and LSC at +2 h and +4 h, with increased Density, and at +6 h decreased ATD and increased Average Shortest Path. The authors interpret these patterns as reduced verbosity, lexicon, dispersion and both global and local connectivity in storytelling under LSD. When controlling for overall verbosity, certain effects for ambiguous stimuli persisted: at +2 h LSD decreased Nodes (p = 0.01), Diameter (p = 0.05) and LCC (p = 0.001) and increased Density (p = 0.045), Clustering Coefficient (p = 0.04) and ATD (p = 0.03). Reported period and interaction statistics included a period effect for neutral stimuli (F(1,19) = 5.28, p = 0.033, ηp2 = 0.22) and an interaction for ambiguous stimuli (F(1,19) = 7.69, p = 0.012, ηp2 = 0.29). Semantics: No treatment, period, order or interaction effects were observed for semantic measures in experience reporting. In storytelling, a treatment effect emerged for creative stimuli: LSD decreased semantic distances (i.e. increased semantic similarity). Specific variables showed reductions at different times: Forward Flow decreased at +2 h, +6 h and +24 h; Flow Steps decreased at +4 h; Semantic Spread decreased at +24 h. These changes indicate greater semantic similarity between consecutive and overall words in creative storytelling during acute and sub-acute phases. Vocabulary: In experience reporting, the primary consistent finding was reduced time- and space-related vocabulary across time points (+1.5 h and +6.5 h), plus increased biology-related words at +6.5 h; order effects (lower values for LSD–placebo) suggested carryover influences on vocabulary. For storytelling, a treatment effect for creative stimuli indicated greater LIWC category use under LSD, while order effects for positive and neutral stimuli again indicated carryover. Pairwise and time-dependent comparisons for storytelling revealed several grammar- and person-related changes: increased dictionary words and verbs at +2 h, increased punctuation at +6 h, and increased third-person plural pronoun they and present-tense focus at +24 h. Positive and neutral stimuli also showed decreased past-focus and increased present-focus words and decreased ingestion-related vocabulary (food-related) at specific time points. The qualitative word-frequency clouds reflected thematic differences across conditions and times but were not subjected to statistical testing. The authors summarise that vocabulary shifts encompassed grammar, person reference, time/space and biological-process categories, and that some changes align with subjective reports of time–space distortions and somatic focus under LSD.

Wießner and colleagues interpret their findings as evidence that a low-to-moderate LSD dose alters language in multiple, stimulus- and time-dependent ways. The strongest structural effects were observed in ambiguous storytelling around and after peak drug effects, whereas semantic and vocabulary changes were most evident in creative storytelling during and after the peak. Structurally, the reduced verbosity, lexicon, dispersion and connectivity are taken to reflect simpler and more disconnected speech under LSD. The authors link these structural changes to cognitive impairments commonly seen with psychedelics—such as disruptions of planning, attention and working memory—and note parallels to reduced connectedness observed in neurocognitive disorders. Semantically, decreased distances (increased similarity) in creative storytelling suggest constricted semantic diversity during acute and sub-acute phases. The researchers propose that task demands and stimulus abstraction may moderate whether psychedelics increase or decrease semantic variability: demanding storytelling tasks here yielded decreased distances, whereas other studies using free association or higher doses reported increased semantic variability. The authors further observe that semantic constriction has been associated with depressive mood, and they tentatively propose that problem-focused verbal interventions during the acute LSD state may be less appropriate than approaches leveraging suggestion or hypnotherapeutic techniques—points raised as hypotheses grounded in their data rather than clinical recommendations. Vocabulary changes are interpreted as reflecting subjective perception shifts: reduced time- and space-related words across experience reports and storytelling are consistent with reported temporal and spatial distortions, while increased biology-related words late in the acute phase may reflect somatic focus. Grammar-related shifts (e.g. more verbs, punctuation changes) are seen as consistent with briefer, more concrete sentences. The authors also highlight persistent order effects and carryover, suggesting that some language changes may last beyond two weeks and that this complicates interpretation. Key limitations acknowledged by the study team include the exploratory nature and multiplicity of analyses (raising false-positive risk), emergence of order/carryover effects, differences in the number and timing of experience-report versus storytelling measurements (reducing comparability), and the fixed order of storytelling stimuli (not randomised), which may have introduced stimulus-order confounds. The investigators therefore present the results as exploratory and emphasise the need for replication. Finally, they propose that automatic, quantitative analysis of free speech could be a practical complement to questionnaires and neuroscientific measures for probing psychedelic experiences and call for further work on neural mechanisms and clinical applications.

In this exploratory crossover study, a single 50 μg LSD dose produced changes in language structure, semantics and vocabulary across production types and time points up to +24 h. The investigators conclude that language under LSD appears to reflect both cognitive impairments in storytelling (simpler, disconnected structure and greater semantic similarity) and subjective perceptual changes in experience reporting (reduced time/space language and increased somatic vocabulary). They suggest that automated language analysis may offer an accessible, unconstrained tool to complement existing measures of psychedelic effects and recommend further research into therapeutic formats and neural mechanisms, while noting caveats such as carryover effects that future studies should address.