Can the words a word keeps company with give it hidden feelings?
Words that people rate as neutral, but that usually appear next to good or bad words in everyday language, still carry automatic positive or negative associations.
Source
Implicit Bias Reflects the Company That Words Keep
Study at a glance
- Design
- Human experiment — Within-subject evaluative priming: verb primes with positive/negative semantic prosody, valence-matched non-prosodic controls (and, in Study 2, strongly valenced verbs) shown for 200 ms before positive/negative target words (Study 1) or pictures (Study 2).
- N
- 78 students in Study 1 and 89 analysed in Study 2 (90 tested, one excluded); explicit valence came from existing norms, not from these participants.
- Population
- Introductory psychology undergraduates at a large Midwestern US university
- Outcome
- Speed of classifying positive vs negative targets after each prime (congruity effect, log reaction time)
Structured fields used in claim comparison tables when every cited study has a complete layer.
What they did
The researchers picked 'semantically prosodic' verbs such as cause, commit and peddle (usually followed by negative words) and restore, gain and provide (usually followed by positive words), which people nonetheless rate as neutral, plus control verbs with the same rated valence but no biased company. In an evaluative priming task, each verb flashed for 200 ms before a positive or negative target that participants had to classify as quickly as possible. Study 1 (78 students) used target words; Study 2 (89 students) used target pictures and added strongly positive or negative verbs as a check that the task works.
What they found
In both studies, verbs with negative semantic prosody sped up classification of negative targets and verbs with positive prosody sped up positive targets, a congruity effect of small-to-medium size in Study 1 (d about 0.3) and medium-to-large in Study 2 (d about 0.6-0.7). Valence-matched control verbs produced no such congruity effect, and in Study 2 the effect for prosodic verbs was statistically indistinguishable from that of strongly valenced verbs. Word-embedding analyses confirmed the prosodic verbs sit near positive or negative words in a large news corpus.
The limits
What it doesn't show
The study measured naturally occurring language patterns rather than manipulating them, so it cannot show that exposure to biased company causes implicit associations; implicit attitudes could equally shape how people use words, as the authors acknowledge. Explicit valence came from norms collected on other people, not from these participants, and only eight prosodic verbs were tested with non-social words, so the jump to social implicit biases such as race or gender is an extrapolation. Samples were US undergraduates in single labs.
Key terms
- Semantic prosody
- The tendency of a word to occur alongside words of a particular emotional tone (e.g., 'cause' with problems and damage), even if the word itself seems neutral.
- Evaluative priming
- An implicit measure in which a briefly shown prime speeds classification of targets with the same valence; the pattern of speed-ups reveals the prime's automatic valence.
- Causal embedding hypothesis
- The idea that the statistical company words keep in a language does not just mirror implicit biases but helps create them.
- Propositional knowledge
- Beliefs a person consciously endorses and can report, such as judging a word to be pleasant or unpleasant.
- Collocate
- A word that regularly occurs near another word in natural text.
Flashcards
0 of 12 answers reviewed
Research intelligence for this paper
See its role on concept claims, tensions it is part of, placement history, and related discoveries.
Quiz yourself
Which verb was used as an example of negative semantic prosody?
Common questions
Why not just use the Implicit Association Test?
An IAT needs two contrasting target categories, and the neutral verbs had no obvious opposite category, so the authors used evaluative priming, which works with single primes.
What do the control verbs add?
They had the same rated pleasantness as the prosodic verbs but no biased company in text. If tiny differences in rated valence drove the effect, the controls should also show it; they did not, pointing to language company instead.
Does this prove language causes prejudice?
No. It shows that implicit associations line up with linguistic company even when conscious ratings do not, which fits the causal account, but the direction of causation was not tested experimentally.
More on Social cognition