Semantic activation in LSD: evidence from picture naming

The experiment used a within-subjects, single-blind design in which ten healthy volunteers completed picture naming under placebo and under intravenous LSD. Placebo and LSD sessions were separated by at least one week and placebo always occurred first; participants were blind to condition but the investigators note that the conspicuous subjective effects of LSD can break blinding. Doses on the LSD day varied between 40 mcg and 80 mcg, as this protocol was part of a broader pharmacokinetic and tolerability project. Task onset on the LSD day varied between approximately 180 and 300 minutes post-injection. Plasma concentrations of LSD were not analysed in the dataset provided. Participants were nine males and one female (mean age 34.2 ± 7.4 years), mostly native English speakers, all with prior psychedelic experience. Screening included psychiatric interview, routine blood and urine tests, ECG, blood pressure and heart rate, and a neuropsychological exam. Ethical approvals and regulatory licences were obtained and participants gave written informed consent. Stimuli comprised 24 pictures drawn largely from prior object-naming work, organised into three semantic categories (body parts, clothing, vehicles) with 8 items per category. The design implemented 12 block types repeated twice in pseudorandom order, producing a total of 768 trials. Blocks were constructed to create graded semantic contexts: same-category blocks, near-category blocks (two semantically closer categories paired), and far-category blocks (pairing a distant category with one of the others). Half of the vehicle-category trials served as fillers. On each trial a fixation cross (300 ms) preceded a blank screen (450 ms) and then the target picture (up to 4000 ms or until speech onset). Responses were recorded as .wav files using PsychoPy and response latencies were extracted with an automated Praat script and manually checked for two participants. Accuracy and error types were coded manually. Participants rated subjective drug effects periodically throughout the session. For reaction-time analyses the investigators excluded filler blocks and first presentations of each item, all trials with errors (243 trials), trials with sound disturbances that could impair voice-onset detection (569 trials), and outliers with latencies <250 ms or >2000 ms (37 trials). Mean reaction times were analysed with a 2×3 ANOVA (drug condition × block type), treating subjects and items as random effects (F1 and F2). Error data, which were non-normally distributed, were analysed using Wilcoxon signed-rank tests by subjects and items.

Naming latency analysis excluded a substantial number of trials for errors and technical disturbances; the investigators report 243 error trials, 569 trials with sound disturbances, and 37 latency outliers. The 2×3 ANOVA showed a reliable main effect of block type across both subjects and items, indicating the expected semantic context effect: far-category blocks produced faster naming latencies than near and same-category blocks. Specifically, pairwise tests found far < near (t1(9)=2.79, p=.021; t2(15)=3.08, p=.008) and far < same (t1(9)=2.84, p=.019; t2(15)=3.88, p=.11). There was no significant difference between near and same conditions. The main effect of drug condition was significant only in the items analysis (F2(1,15)=16.47, p<.001) but not in the subjects analysis (F1(1,9)=1.77, p=.217). The interaction drug × block type did not reach significance in either analysis by items or subjects. Error analyses used categorical coding for filled pauses, hesitations, self-corrections, and full lexical substitutions, with lexical substitutions further classified as same-category or different-category. Overall error counts were low. Wilcoxon tests indicated that same-category lexical substitutions were significantly more frequent under LSD than under placebo (subjects: Z1=2.082, p=.038; items: Z2=2.253, p=.006). No significant drug effects emerged for self-corrections, filled pauses, hesitations, or different-category lexical substitutions. When examining block-type effects, comparisons suggested that same-category substitutions tended to be higher in the same-category blocks than in near or far blocks, with some tests approaching or reaching significance; however, an interaction between drug and block type was significant only by subjects (Z1=2.191, p=.028) and not by items (Z2=1.363, p=.173), implying that the interaction may not generalise across items. The authors also note that neither task onset time nor LSD dose correlated with the main outcomes in their sample. Subjective reports indicated participants were experiencing near-peak drug effects at the beginning of the naming task on the LSD day, but plasma LSD concentrations were not available to confirm pharmacokinetic status at testing.

Family and colleagues interpret the pattern of results as evidence that LSD increases the spread of semantic activation. They highlight that the classic semantic context effect (far faster than same) was replicated regardless of drug, but that LSD selectively increased full lexical substitutions for semantically similar items. This selectivity is taken as consistent with broader activation within the semantic network: semantically proximate competitors receive stronger activation and are therefore more likely to be erroneously produced. The investigators consider whether reduced self-monitoring under LSD might account for increased full substitutions; some participant comments suggested difficulty sustaining attention or focus under the drug (for example, "sometimes I'll feel a bit slow because my brain has been off somewhere else and I have to bring it back to focus"). However, they note that hesitations and self-corrections—indices of early error detection—were not reduced under LSD, which complicates a simple self-monitoring deficit account. Reaction times did not differ reliably between drug and placebo, leading the authors to suggest that error measures may be more sensitive than latencies for detecting drug-induced shifts in semantic activation in production tasks. Limitations acknowledged by the investigators include the small sample size, variable dosing, variability in task onset times post-administration, the single-blind design (with the potential for broken blinding), and the absence of plasma LSD measurements. They also note that the highly controlled, repetitive picture-naming paradigm may under-sample spontaneous language processes that psychedelics preferentially affect, and suggest that more naturalistic speech tasks or diverse stimuli could reveal broader semantic effects. For future work the authors recommend combining controlled behavioural paradigms with more naturalistic language measures and direct neuroscientific indices (for example, N400 event-related potentials) to test semantic spread more objectively. They view the present pilot findings as broadly consistent with prior psilocybin results and with the hypothesis that psychedelics increase the breadth and flexibility of cognition, while stressing the need for replication and more finely controlled pharmacokinetic and neurophysiological measurements.

The study's principal conclusion is that LSD appears to increase spread of semantic activation, as evidenced by a selective rise in same-category lexical substitution errors during picture naming. The authors caution that further work is required to establish the reliability and specificity of this effect, and to disentangle contributions of altered self-monitoring versus genuine changes in semantic activation. They advocate using more naturalistic language measures and electrophysiological methods in future studies to improve ecological validity and to provide more direct tests of semantic processing under psychedelics. Overall, the findings are presented as consistent with the broader view that psychedelics alter cognitive breadth and flexibility.