Liwc Linguistic Inquiry
LIWC is the canonical closed-vocabulary instrument for turning text into psychologically interpretable numeric features: it counts words, stems, phrases, punctuation, and related entries against curated dictionaries, then reports category proportions such as affect, cognition, social orientation, pronoun use, and time focus. Its enduring value is not that it “understands” language, but that it offers a transparent, stable, theory-linked measurement layer for Behavioral Language Analytics—one that is now best treated as one component in a triangulated pipeline rather than as a standalone psychological oracle.
Coverage note: verified through May 19, 2026.
Definition and scope
LIWC, or Linguistic Inquiry and Word Count, is a dictionary-based text-analysis system developed in the Pennebaker research tradition. In its standard use, LIWC takes written or transcribed text, compares tokens against a curated dictionary, and outputs variables corresponding to psychologically meaningful categories. Tausczik and Pennebaker’s 2010 review describes LIWC as a “transparent text analysis program” that counts words in psychologically meaningful categories and links word use to attentional focus, emotionality, social relationships, thinking style, and individual differences. CMU School of Computer Science
The core idea is deliberately simple: instead of training a model to infer latent states from arbitrary text, LIWC defines a set of categories in advance—such as function words, affective processes, cognitive processes, social processes, and time orientation—then measures how often words from those categories occur. Earlier LIWC-era papers described roughly 80 categories; LIWC-22 expands and restructures the system into Basic and Expanded dictionaries with many updated variables, while preserving the central word-count architecture. CMU School of Computer Science
That architecture places LIWC in the family of Closed-Vocabulary Text Analysis methods. It is not an open-ended semantic model, a topic model, an embedding model, or a large language model. It is a measurement instrument: a curated mapping from lexical forms to psycholinguistic variables.
Why LIWC mattered
LIWC became influential because it made psychological language analysis reproducible at scale. Before tools like LIWC, researchers interested in language, personality, emotion, coping, or deception often relied on human coding, small hand-labeled corpora, or qualitative interpretation. LIWC offered a middle path: less expressive than human reading, but faster, more reliable, and more easily compared across studies.
Pennebaker and King’s 1999 paper, “Linguistic Styles: Language Use as an Individual Difference,” is a key early foundation. They tested whether language use could reflect personality style using computerized word-based text analysis across daily diaries, writing assignments, journal abstracts, essays, Thematic Apperception Test coding, self-reports, behavior, and health markers. Their conclusion was intentionally modest but important: despite modest effect sizes, linguistic style appeared to be an independent and meaningful way of exploring personality. PubMed
The later Annual Review article by Pennebaker, Mehl, and Niederhoffer framed natural word use as a route into personality, social processes, situational fluctuations, and intervention effects. It also emphasized particles—pronouns, articles, prepositions, conjunctions, and auxiliary verbs—as psychologically informative features, precisely because they are often produced automatically and carry social-cognitive structure. annualreviews.org
This is the conceptual move that made LIWC important for Behavioral Language Analytics: the psychologically interesting signal is not only in topical content, but also in small linguistic habits. “I,” “we,” “you,” “the,” “but,” “because,” “not,” “will,” and “was” are not merely grammar; in aggregate, they can index attention, self-focus, social relation, temporal orientation, certainty, abstraction, and cognitive work.
How LIWC works
LIWC-22 accepts text in several machine-readable formats, including text files, PDFs, Word documents, CSV files, and spreadsheets. Its core module compares each target text against the LIWC-22 dictionary, counts all words, and calculates the percentage of total words represented in each LIWC subdictionary. The output includes the file name, word count, and percentage of words captured for each language dimension. LIWC
The standard score for a category is therefore:
LIWC category score = category-matching tokens / total words × 100
A score of 2.5 for negative emotion means that 2.5% of the words in the document matched entries in the negative-emotion dictionary. Scores are usually interpreted as relative frequencies, not as absolute psychological states.
LIWC’s dictionary is not a one-word-one-category system. In LIWC-22, an entry can belong to multiple categories. The manual gives the example of “cried,” which increments several dimensions including affect, positive or negative tone-related categories, sadness, verbs, past focus, communication, linguistic, and cognition. The dictionary also supports stems such as hungr*, allowing multiple inflected forms to map into the same category. LIWC
LIWC-22 substantially broadened what the dictionary engine can process. Its dictionary contains more than 12,000 words, stems, phrases, and selected emoticons; the updated engine can accommodate numbers, punctuation, short phrases, and regular expressions, including netspeak and SMS-like forms such as “b4” and emoticons such as “:)”. LIWC
Category structure
LIWC categories are partly linguistic, partly psychological, and partly topical. The psychologically central variables are not all equally abstract. Some are near-surface grammatical categories; others are interpretive dictionaries meant to approximate constructs such as affiliation, achievement, power, cognitive processing, positive emotion, negative emotion, social behavior, and temporal focus.
| Category family | Representative LIWC variables | What it usually measures | Interpretation risk |
|---|---|---|---|
| Summary variables | Analytic, Clout, Authentic, Tone | Composite indices such as formal/logical thinking, status language, genuineness, and emotional tone | These are less transparent than simple word-count categories |
| Function words | Pronouns, articles, prepositions, conjunctions, auxiliary verbs, negations | Linguistic style, attention, social orientation, abstraction | Often powerful but unintuitive; small changes can be overread |
| Affect | Positive emotion, negative emotion, anxiety, anger, sadness | Verbal expression of affective language | Not equivalent to felt emotion |
| Cognition | Cognitive processes, insight, causation, discrepancy, tentativeness, certainty, memory | Thinking style, causal reasoning, epistemic stance | Context strongly changes meaning |
| Social processes | Social behavior, prosocial behavior, politeness, conflict, moralization, family, friends, gender references | Social orientation and relational content | Topic and genre can dominate |
| Time orientation | Past focus, present focus, future focus | Temporal framing and attentional orientation | Strongly dependent on task prompt |
| Expanded topical domains | Culture, politics, technology, work, religion, health, food, death | What the text is about | More content-like; can drift culturally |
LIWC-22’s published reliability table lists categories such as function words, pronouns, prepositions, affect, positive emotion, negative emotion, anxiety, anger, sadness, social processes, family, friends, cognition, achievement, power, and expanded domains such as culture, technology, health, illness, mental health, substances, food, death, perception, and time orientation. LIWC+2LIWC+2
A critical methodological point: LIWC variables are not independent by construction. A single token can increment several categories, and many categories are hierarchical. “Sadness” words are also negative emotion words, emotion words, tone-related words, and affect words. This is useful for multi-level analysis, but it means users should not treat category outputs as orthogonal features without checking collinearity.
Dictionary construction and psychometrics
LIWC dictionaries are not arbitrary word lists. The LIWC-22 manual describes a multi-step development process involving word collection, human judges, candidate-word generation, corpus base-rate analysis, conceptual review, cross-categorization, and psychometric evaluation. Candidate words were evaluated by teams of judges for conceptual fit, and words detrimental to category internal consistency could be removed. LIWC
This gives LIWC a distinctive psychometric profile. The instrument is not just a parser; the dictionary itself is a measurement artifact. Category validity depends on which words are included, which are excluded, how overlapping categories are handled, how stems are defined, and which corpora were used to test base rates and internal consistency. That makes dictionary curation part of the model.
LIWC-22 reports reliability statistics differently from standard questionnaire psychometrics. The manual notes that traditional Cronbach’s alpha can underestimate reliability for language categories because word-use base rates vary heavily; it recommends paying attention to KR-20-style estimates for category internal consistency. LIWC
The LIWC-22 manual is also explicit that validation is complex. It reports that correlations between LIWC affect or emotion categories and self-reported affective feelings often range from about .05 to .40, averaging around .15 to .20, with judge ratings of writing samples often slightly higher. The manual warns users not to expect self-report questionnaires and LIWC scores for “the same” construct to correlate strongly, or sometimes at all. LIWC
That is not a defect unique to LIWC. It is the basic problem of Construct Validity in behavioral measurement: self-report, observer report, lexical behavior, physiological signal, and task performance can all be valid but non-equivalent indicators of a latent construct.
Major empirical findings
Personality and linguistic style
Pennebaker and King’s 1999 work established one of the central LIWC claims: individual differences appear in linguistic style, not only in content. Across several samples, they found internal consistency for many language dimensions, extracted replicable factors from student essays, and compared linguistic profiles with self-reports, behavioral measures, health markers, and projective-test coding. The effect sizes were modest, but the overall result supported the idea that linguistic style is a meaningful individual-difference signal. PubMed
The enduring insight is that personality language markers are often indirect. Extraversion, neuroticism, openness, or conscientiousness do not simply appear as words like “extraverted” or “anxious.” They show up through distributed patterns: emotion words, social words, pronouns, certainty markers, articles, prepositions, tense, and topical regularities. This is why LIWC became a useful baseline for Personality Prediction from Text even after more powerful open-vocabulary and neural methods emerged.
Emotion expression
Kahn, Tobin, Massey, and Anderson’s 2007 paper tested whether LIWC emotion categories validly measure verbal emotional expression. In three experimental studies, LIWC emotion counts distinguished sad and amusing autobiographical memories and reactions to emotion-provoking film clips; the authors found weak relations between LIWC emotion counts and individual-difference measures such as emotional reactivity, dispositional expressivity, and personality. Their conclusion was that LIWC is valid for measuring verbal expression of emotion, not necessarily internal emotion itself. ISU ReD
That distinction matters. LIWC can tell us that a text contains more negative-emotion words, sadness words, anger words, or positive-emotion words. It cannot, by itself, prove that the writer felt more sadness, anger, or joy. The measured construct is lexical expression in a context, not raw affect.
Deception
Newman, Pennebaker, Berry, and Richards’ 2003 paper, “Lying Words: Predicting Deception From Linguistic Styles,” is one of LIWC’s best-known applied studies. Across five independent samples, their computer-based text analysis classified liars and truth-tellers at 67% accuracy when topic was held constant and 61% overall. Compared with truth-tellers, liars showed lower cognitive complexity, fewer self- and other-references, and more negative-emotion words. Stanford University
The combined predictive model used five LIWC categories: first-person singular pronouns, third-person pronouns, negative-emotion words, exclusive words, and motion verbs. Across studies, deceptive communications were characterized by fewer first-person singular pronouns, fewer third-person pronouns, more negative-emotion words, fewer exclusive words, and more motion verbs. Stanford University
This is often overinterpreted. The paper itself notes limits: the model was specific to English and possibly American English, and pronoun-based deception markers may not generalize to languages where pronouns are optional or encoded differently in verbs. The authors recommend building and validating language-specific deception profiles rather than assuming the same markers transfer. Stanford University
The honest summary: LIWC-style deception markers are useful research signals and investigative clues, not reliable lie detectors.
Depression and self-focus
Rude, Gortner, and Pennebaker’s 2004 study examined essays by currently depressed, formerly depressed, and never-depressed college students. Their text-analysis approach computed incidence of words in predesignated categories and found patterns consistent with cognitive and self-focus models of depression, including greater use of negatively valenced language among depressed participants. semanticscholar.org
Later synthesis supports the broad pattern but with important caveats. A 2019 systematic review and meta-analysis found a distinctive pattern in depression research: increased first-person singular pronoun use, increased negative-emotion word use, and decreased positive-emotion word use. It reported small-to-medium relations for first-person singular pronouns and depression, and noted possible publication bias in the negative-emotion literature. Tidsskrift
The clinically relevant constraint is sharp: these language markers are not diagnostic instruments. Depression is heterogeneous, language tasks differ, and positive-emotion language can behave differently near suicidality or relief states. The meta-analysis itself argues for more symptom-level nuance rather than treating all depression as one language profile. Tidsskrift
Methodological strengths
LIWC’s main strengths are transparency, comparability, low variance, and theory alignment.
First, LIWC is auditable. A researcher can inspect category definitions, understand the counting procedure, and reproduce outputs on the same text. This matters in Interpretable Machine Learning contexts where the goal is not only prediction, but also theory testing.
Second, LIWC variables are comparable across studies. Because the same dictionaries produce the same variables, researchers can ask whether first-person singular pronouns, cognitive-process words, affiliation words, or negative-emotion words behave similarly across corpora, tasks, and populations.
Third, LIWC works with relatively small samples compared with many open-vocabulary methods. Open-vocabulary analyses often require large corpora to estimate stable word, phrase, topic, or embedding associations. LIWC can be informative in smaller experimental or clinical datasets because the feature space is predefined.
Fourth, LIWC is especially strong for function-word and style variables. Eichstaedt and colleagues’ comparison of automated text-analysis approaches notes that closed-vocabulary programs can rapidly transform many rare words into 10–100 interpretable variables and that validated dictionaries are suitable for testing specific hypotheses, particularly where function words are central. Johannes Eichstaedt
Critiques and failure modes
Closed-vocabulary blindness
The most basic limitation is that LIWC cannot count what its dictionary cannot see. Words outside the dictionary are invisible to the relevant category unless captured by a stem, phrase, regular expression, or related entry. This creates problems for slang, neologisms, domain-specific language, subcultural terms, technical registers, and fast-moving online discourse.
LIWC-22 improved coverage through phrases, punctuation, regular expressions, netspeak, and updated categories, but it remains a predefined dictionary system. This is a strength for comparability and a weakness for discovery.
Context, sarcasm, and word sense
Tausczik and Pennebaker state the critique directly: computerized language measures remain crude, and systems such as LIWC ignore context, irony, sarcasm, and idioms. Their example is “mad,” which may indicate anger in one sentence, attraction or enthusiasm in another, or insanity in an idiom. CMU School of Computer Science
This is the classic closed-vocabulary failure mode: the same surface form maps to the same category even when the meaning changes. LIWC-22’s phrase support helps with some local context, but it does not solve compositional semantics.
Register and corpus dependence
LIWC category base rates vary strongly by context. The LIWC-22 manual notes that dictionary coverage averages roughly 80–90% across speech, social media, personal writing, and formal writing, but word-use base rates vary substantially depending on what people are talking about and where they are talking. The manual’s examples contrast therapist-office language with ceremonial speech, and notes that different corpora have distinct linguistic fingerprints. LIWC
This matters because a high score may reflect genre rather than psychology. A legal brief, therapy transcript, Reddit post, lab essay, medical note, and political speech can all produce different LIWC profiles because their communicative purposes differ.
Cross-cultural and cross-linguistic limits
LIWC-style categories do not automatically transfer across languages. Deception research made this point explicitly for pronouns: languages differ in whether pronouns are required, optional, or encoded through verb morphology, so English pronoun markers may fail elsewhere. Stanford University
More generally, function words carry different social information across linguistic and cultural systems. A dictionary validated in one language, population, or register is not automatically valid in another. Cross-cultural LIWC work therefore requires local validation, not just translation.
Psychometric properties of the dictionary itself
A LIWC category is a measurement instrument. Its validity depends on lexical choices, category boundaries, base rates, overlap with other categories, and the corpus used during development. This means that dictionary revisions can change what a variable measures. LIWC-22 explicitly reports substantial additions, removals, and restructurings relative to earlier versions, including the Basic/Expanded split and updated cognitive and affective categories. LIWC
For longitudinal research, this creates a versioning issue. A “negative emotion” score from LIWC2007, LIWC2015, and LIWC-22 may be conceptually related, but not necessarily identical.
Closed vocabulary versus open vocabulary
The major alternative to LIWC is Open-Vocabulary Text Analysis, where the feature space is learned from the corpus rather than imposed in advance.
Schwartz et al.’s 2013 PLOS ONE paper, “Personality, Gender, and Age in the Language of Social Media: The Open-Vocabulary Approach,” is the canonical comparison point. The authors analyzed 700 million words, phrases, and topic instances from Facebook messages written by 75,000 volunteers who also completed personality tests. Their open-vocabulary method found connections not captured by traditional closed-vocabulary word-category analyses. PLOS
Their method, Differential Language Analysis, extracts words, phrases, and topics from the data itself, then correlates those features with known attributes such as gender, age, location, or personality. They explicitly contrast this with a priori word-category analysis, where fixed dictionaries constrain what can be found. PLOS
| Dimension | LIWC / closed vocabulary | DLA / open vocabulary |
|---|---|---|
| Feature source | Predefined dictionaries | Corpus-derived words, phrases, topics, embeddings |
| Best for | Theory testing, comparability, small-to-medium datasets, function-word hypotheses | Discovery, prediction, cultural specificity, phrase-level and topic-level patterns |
| Interpretability | High at category level | Variable; word clouds and topics can help but may be noisy |
| Adaptability | Limited by dictionary updates | Adapts to corpus vocabulary |
| Context sensitivity | Low to moderate | Higher, especially with n-grams, topics, embeddings, or LLMs |
| Main failure mode | Missing or misclassifying context-dependent language | Overfitting, multiple comparisons, unstable features, post-hoc interpretation |
| Scientific role | Measurement instrument | Discovery and predictive modeling instrument |
Eichstaedt and colleagues’ 2021 comparison is especially useful because it does not frame the debate as simple replacement. Comparing LIWC, General Inquirer, DICTION, LDA, and DLA on Facebook status updates from 65,896 users, they found that closed-vocabulary methods efficiently summarize concepts and help explain how people think, while open-vocabulary methods reveal more specific and concrete patterns, better handle ambiguous word senses, and are less prone to some misinterpretations. Their recommendation is complementary use. Johannes Eichstaedt
Their later discussion is blunt: closed-vocabulary dictionaries are rigidly defined, insensitive to context and word sense, and unable to accommodate changing word senses over time. But they also argue that closed-vocabulary methods remain desirable because they are parsimonious, comparable across studies, suitable for validated hypotheses, and often effective for function-word patterns. Johannes Eichstaedt
The practical conclusion is not “LIWC or open vocabulary.” It is:
Use LIWC when the construct is theory-defined and the dictionary is validated.Use open vocabulary when the goal is discovery, prediction, or cultural specificity.Use both when the output will inform psychological interpretation.
LIWC, Empath, and the Psyche analysis pipeline
In the Psyche analysis pipeline for this wiki/blog project, LIWC is best understood as a low-weight, interpretable signal source rather than a decisive classifier. As specified here, Psyche uses LIWC alongside Empath, with LIWC contributing 0.10 weight to a triangulated profile. That weight encodes a methodological stance: LIWC is valuable because it is stable, theory-linked, and interpretable, but it should not dominate richer open-vocabulary, contextual, or model-based signals.
Empath is a useful comparison layer. Fast, Chen, and Bernstein introduced Empath as a tool that can generate and validate lexical categories from seed terms using neural embeddings trained on more than 1.8 billion words of modern fiction, then validate categories with crowd filtering. Empath also includes 200 built-in categories and reported high correlation with similar LIWC categories. arXiv
For Psyche Analysis Pipeline, this suggests a division of labor:
| Pipeline role | LIWC contribution | Empath / open-vocabulary contribution |
|---|---|---|
| Stable psycholinguistic baseline | Strong | Moderate |
| Category interpretability | Strong | Moderate to strong |
| New domain discovery | Weak | Stronger |
| Slang/register adaptation | Weak to moderate | Stronger |
| Function-word measurement | Strong | Not primary |
| Cross-method triangulation | Useful as independent low-variance signal | Useful as broader semantic signal |
The 0.10 LIWC weight is defensible if the system treats LIWC as a calibration prior: it can confirm or disconfirm patterns found elsewhere, flag unusually high or low lexical features, and provide reproducible psycholinguistic dimensions. It would be indefensible if used as a direct diagnostic or personality oracle.
LIWC in behavioral-language-analytics literature
LIWC sits at the center of a broader lineage: computerized text analysis, psycholinguistic measurement, computational social science, and now LLM-mediated psychological assessment. It is a bridge between older psychological measurement and modern NLP.
DLATK, the Differential Language Analysis Toolkit, represents one major successor ecosystem. It provides an open-source Python and command-line package for social-scientific language analysis, including tokenization, classification, structured metadata integration, specified units of analysis such as document/user/community, statistical metrics for continuous outcomes, and prediction pipelines. ACL Anthology
LLM-based approaches are the newest pressure on LIWC’s role. Rathje and colleagues’ 2024 PNAS paper tested GPT-based psychological text analysis across multilingual datasets and reported that GPT could detect constructs such as sentiment, discrete emotions, offensiveness, and moral foundations across 12 languages with stronger correspondence to manual annotations than English-language dictionary methods in their evaluated settings. Astrophysics Data System
But the LLM replacement story is not settled. Ziems et al.’s 2024 Computational Linguistics study evaluated 13 LLMs across 24 computational social-science tasks and found that, except in minority cases, prompted LLMs did not match carefully fine-tuned classifiers and were often not strong enough to replace human annotation entirely, though they could support human–AI partnered labeling. ACL Anthology
The social-science methods literature is converging on a cautious view. Abdurahman and colleagues emphasize that LLMs are increasingly used for coding, text analysis, and simulation, but raise concerns about reliability, validity, access, transparency, nondeterminism, model updates, and replicability. Sage Journals Brickman, Gupta, and Oltmanns similarly frame LLMs as promising tools for scalable behavioral assessment, while stressing the need for construct validation and careful experimental design. Sage Journals
Does LIWC still have a role in the LLM era?
Yes, but its role changes.
LIWC should no longer be treated as the frontier method for semantic understanding, contextual interpretation, multilingual text classification, or high-accuracy prediction. LLMs and open-vocabulary models are better suited to many of those tasks. They can reason over context, handle paraphrase, process longer phrases, adapt to new domains, and label constructs that no fixed dictionary anticipates.
However, LIWC retains a strong role in four settings.
First, it remains useful for theory-first measurement. If the hypothesis concerns pronoun use, cognitive-process terms, affect words, temporal focus, or social-reference language, LIWC provides a validated starting point.
Second, it remains useful for auditability. LLM labels can be difficult to reproduce across models, prompts, temperatures, providers, and update cycles. LIWC scores are deterministic and easy to inspect.
Third, it remains useful for cross-study comparability. A LIWC variable is a shared measurement convention. That matters for meta-analysis and cumulative science.
Fourth, it remains useful as a baseline or feature layer in hybrid systems. A strong modern pipeline can combine LIWC, Empath, DLA, embeddings, supervised classifiers, and LLM judgments, then check whether the signals converge or conflict.
The open question is not whether LIWC can beat LLMs at general language understanding. It cannot. The open question is whether psychological text analysis needs stable, interpretable, low-dimensional behavioral measures even when richer semantic models are available. The answer is still yes for many research and engineering contexts.
A good contemporary design treats LIWC as a small, transparent instrument inside a broader Triangulated Language Analytics stack:
LIWC = stable psycholinguistic measurementEmpath = expandable lexical-semantic categoriesDLA / embeddings = corpus-specific discoverySupervised models = task-optimized predictionLLMs = contextual interpretation and flexible codingHuman review = construct validation and error correction
The intellectual honesty is that none of these layers is sufficient. LIWC misses context. Open-vocabulary methods can overfit. LLMs can be unstable and opaque. Human labels are expensive and biased. The strongest systems make disagreement visible rather than hiding it.
Selected primary references
[Pennebaker & King 1999 — linguistic styles as individual differences] established the early empirical case that computerized language features can reflect stable individual differences, while reporting modest effect sizes and validating against multiple behavioral and self-report measures. PubMed
[Pennebaker, Mehl & Niederhoffer 2003 — natural language use] reviewed how everyday words, especially function words and particles, connect to personality, social context, psychological intervention, and linguistic style. annualreviews.org
[Tausczik & Pennebaker 2010 — LIWC and computerized text analysis] summarized LIWC’s creation, validation, and empirical use across attentional focus, emotionality, social relationships, thinking styles, and individual differences, while explicitly warning about context, irony, sarcasm, idioms, and probabilistic interpretation. CMU School of Computer Science
[Boyd et al. 2022 — LIWC-22 development and psychometrics] documents LIWC-22’s dictionary construction, updated processing engine, category revisions, psychometric evaluation, reliability statistics, and validation caveats. LIWC+3LIWC+3LIWC+3
[Kahn et al. 2007 — emotional expression] experimentally validated LIWC emotion counts as measures of verbal emotional expression, while distinguishing expression from individual differences in emotional reactivity or personality. ISU ReD
[Newman et al. 2003 — deception] found LIWC-style linguistic markers could classify deception above chance across five samples, but with limited accuracy and clear cross-language constraints. Stanford University+2Stanford University+2
[Rude, Gortner & Pennebaker 2004 — depression language] examined language in currently depressed, formerly depressed, and never-depressed college students, linking depression to negative valence and self-focus patterns. semanticscholar.org
[Schwartz et al. 2013 — open-vocabulary DLA] introduced a large-scale open-vocabulary alternative using words, phrases, and topics from Facebook language to study personality, gender, and age. PLOS+2PLOS+2
[Eichstaedt et al. 2021 — closed/open method comparison] compared LIWC, General Inquirer, DICTION, LDA, and DLA, concluding that closed- and open-vocabulary approaches are complementary rather than interchangeable. Johannes Eichstaedt
Companion entries
Core theory: Behavioral Language Analytics, Closed-Vocabulary Text Analysis, Function Words, Psychological Construct Validity, Verbal Behavior as Measurement, Psycholinguistic Markers
Methods: Open-Vocabulary Text Analysis, Differential Language Analysis, DLATK, Empath, Dictionary-Based NLP, Triangulated Language Analytics, LLM-as-Judge, Interpretable Machine Learning
Applications: Personality Prediction from Text, Emotion Detection, Deception Detection, Depression Markers in Language, Affective Computing, Computational Social Science, Psyche Analysis Pipeline
Counterarguments and limits: Context Collapse in Text Analysis, Lexical Drift, Cross-Cultural Psychometrics, Register Effects, Construct Validity Crisis, LLM Validity and Reproducibility