SciELO - Scientific Electronic Library Online

vol.36 issue1Escala de orientações gerais para objetivos de desempenho: Estudos de adaptação para o ensino secundário author indexsubject indexarticles search
Home Pagealphabetic serial listing  

Services on Demand




Related links

  • Have no similar articlesSimilars in SciELO


Análise Psicológica

Print version ISSN 0870-8231On-line version ISSN 1646-6020

Aná. Psicológica vol.36 no.1 Lisboa Mar. 2018 

Acknowledging the role of word-based activation in spontaneous trait inferences

Diana Orghian1, Tânia Ramos2, Joana Reis3, Leonel Garcia-Marques3

1Massachusetts Institute of Technology, The MIT Media Lab, Cambridge, USA / CIPSI, Faculdade de Psicologia, Universidade de Lisboa, Lisboa, Portugal

2New York University, New York, USA / CIPSI, Faculdade de Psicologia, Universidade de Lisboa, Lisboa, Portugal

3CIPSI, Faculdade de Psicologia, Universidade de Lisboa, Lisboa, Portugal




The first goal of the present paper is to call attention to a confounder in studies that explore Spontaneous Trait Inferences (STI). These studies use, most of the times, behavioral descriptions which strongly imply personality traits about the actor of the behavior. However, a potential limitation of this material is the possibility of the trait being activated by specific words in the sentence (word-based priming) and not, or not only, by an inference made based on the comprehension of the behavioral sentence as a whole (text-based priming). This aspect has been recurrently ignored in previous studies. In the present paper, we discuss how the word-based priming may have obscured the interpretation of previous results in the STI literature. A second goal of this paper is to present a potential solution for this problem. We created a set of 122 trait-implying sentences and their correspondent control versions. These control sentences have approximately the same words as the trait-implying sentences, but the words are rearranged in such a way that the sentences no longer imply the target traits. By keeping the words constant, we control for the activation from individual words in the sentences (word-based priming). Thus, differences in trait activation between the two sentences can only be attributed to text-based priming. Researchers interested in investigating STI in the Portuguese language can use these materials in their studies. With this paper, we hope to stimulate a discussion about the mechanisms responsible for the trait activation in STI studies.

Key words: Inference, Spontaneous trait inference, Word-based activation, Rearranged sentences.



O primeiro objetivo deste artigo é chamar a atenção para um artefacto existente nos estudos que exploram o fenómeno das Inferências Espontâneas de Traço (IETs). Esses estudos utilizam, na maior parte das vezes, descrições comportamentais que implicam fortemente um traço de personalidade acerca do ator do comportamento. No entanto, uma potencial limitação deste tipo de material é a possibilidade de os traços ficarem ativados devido à presença de palavras específicas na frase (primação com base na palavra) e não, ou não apenas, devido a uma inferência feita a partir do comportamento descrito na frase como um todo (primação com base no texto). Este aspeto tem sido recorrentemente ignorado na literatura. Neste artigo, discutimos como a primação com base na palavra pode estar a afetar a interpretação dos resultados existentes na literatura. O segundo objetivo é fornecer uma solução para esta limitação. Criámos, assim, um conjunto de materiais constituídos por 122 frases implicativas de traços, acompanhadas pelos seus traços correspondentes (implicados nas mesmas) e por frases controlo. As frases controlo têm aproximadamente as mesmas palavras que as frases implicativas de traços, mas rearranjadas de tal forma a não implicarem mais os traços críticos. Mantendo as palavras contantes, consegue-se controlar para a ativação vinda de palavras específicas da frase. Assim, diferenças de ativação entre os dois tipos de frases, pode ser atribuída a inferências verdadeiras que considerem a frase como tudo. Investigadores interessados em investigar as IETs na língua portuguesa, podem utilizar este material nos seus estudos. Com este artigo, esperamos estimular a discussão acerca dos mecanismos responsáveis pela ativação de traços personalidade em estudos sobre IETs.

Palavras-chave: Inferências, Inferências espontâneas de traço, Ativação com base na palavra, Frases rearranjadas.


Truth effect

Individuals have the fascinating talent of making inferences and reading in between the lines. This is an efficient and effortless skill which plays a crucial role in our comprehension of the world. If someone says that he or she likes apples, for instance, we automatically know that the person is talking about eating apples. We do not need to retrieve all the possibilities which match the same context (e.g., “she likes the color of apples”, “she likes catching apples”), instead, the appropriate meaning immediately comes to our mind. Inferential thinking is useful in many different contexts; one of them is the social context. If you know that “John did not smoke at home while his roommate was trying to quit”, it is very likely that you will infer that John is a “considerate” person. Research in social cognition has shown that such trait inferences occur spontaneously, that is, in the absence of any specific intention to form impressions or to infer traits (e.g., Winter & Uleman, 1984; for a review see Uleman, Rim, Adil Saribay, & Kressel, 2012).

Inferring personality traits from behavioral descriptions seems intrinsic to our comprehension, but from a methodological standpoint, researchers encounter recurrent difficulties. One of them has been in demonstrating that the trait inference is part of the natural understanding of the behavior, that is, that the inference automatically occurs when the behavioral information is being encoded. Spontaneous Trait Inferences (STI) research has put much effort into creating better paradigms to demonstrate how and when trait inferences occur. But less attention has been paid to the type of behavioral descriptions used. Specifically, the question about what the processes are through which the trait may be activated based on the behavioral descriptions has not been fully considered. In the present paper, we discuss how the sentences used to elicit trait inferences can activate the target traits without a trait inferential process being necessarily involved. Note that while it is usually assumed that the trait is inferred as a result of the comprehension of the behavioral sentence meaning, a plausible alternative explanation for the activation of the trait is a word-to-word activation process. That is, the activation of the trait might result from associations with specific words in the sentence, and not from the understanding of the sentence as a whole. Take, for example, the following behavioral description employed by Carlston and Skowronski (1994) in one of the most cited papers in the STI literature: “I am 18 years old and a doctor. I received my medical degree from Harvard. In my spare time, I enjoy doing research at the Mayo Clinic”. This is a behavioral description which implies the trait “intelligent”. However, there are several words in the sentence (“degree”, “Harvard” and “research”) which may activate the trait “intelligent”. Thus, in order to activate the trait, participants do not need to process and understand the meaning of the sentence, and do not need to make any text-based inferences. Not controlling for word-to-word associations in the material used to study STIs might thus lead to systematic confounds. Importantly, word-based priming cannot be considered an inference, but its outcome perfectly mimics that of a text-based inference.

The present paper has two main goals. The first goal is to discuss the potential role of word-based priming on STI results, and to explain in detail why this issue should be a concern to researchers. The second goal is to present a possible solution to this problem. Specifically, we created a set of materials that controls for word-based priming, which can be used by researchers conducting studies with Portuguese speaking participants. We start by discussing the importance of distinguishing word-based priming effects from true trait inferences, both at the theoretical and methodological levels. We then show how this issue has been neglected in many of the past studies in the trait inference literature. Finally, we present a set of sentences which were initially validated for the Portuguese language, and which can be further developed and applied by researchers in future STI studies.


Word-based priming: The problem of confounding

Keenan, Potts, Golding and Jenning (1990; see also Keenan & Jennings, 1995) called attention to the problem of how word-based priming might contaminate the results found in studies examining inferences during text comprehension. They noticed that two sources of priming, word-based priming and sentence-based priming (i.e., text inferences), had been confounded in many previous studies in the text comprehension field, rendering the evidence for inferences somewhat inconclusive. These authors highlighted that word-based priming is based on intra-lexical associations and it is insensitive to the meaning of the whole sentence, whereas an inference is based on the meaning of the text combined with the perceiver’s knowledge of the described situation. Thus, it was crucial to investigate whether results which had been taken as evidence of these more elaborated, higher level inferential processes, were not simply a result of word-based associations. Moreover, even though this issue has been systematically explored in the text comprehension literature, it has been mostly neglected in the STI research.

Within the STI research, the main aspect we want to emphasize is that the trait activation, usually interpreted as an inference made from the behavior, might be a consequence, not of the (spontaneous) interpretation of the behavioral description, but merely a result of the presence of certain word(s) which are semantically associated with the trait. In fact, it is common for trait-implying sentences to contain words highly associated with the trait concept. And, even if the sentence is processed in a shallow manner (if the participant is under cognitive load, tired or unmotivated for the task), the person might still be able to pick up the presence of these words and activate the trait without the need to deeply process the meaning of the sentence. The need of experimentally controlling for these associations is crucial, not just because of their inference mimicking power, but also because they might interact with sentence-based inferences in unknown ways.

One possible solution to overcome this confounder would be to eliminate from the sentence all the words which individually activate the trait. Then we could be sure that word-based priming effects were not responsible for differences in trait activation. However, it is very difficult, if not impossible, to eliminate all these word-to-word associations (see Forster, 1981; Keenan et al., 1990; Kintsh & Mross, 1985). A more viable solution is to equate the word-based associations present in the trait-implying behavioral descriptions with the control versions, by using the same words, but rearranged in such a way that the meaning of the control sentence changes and no longer implies the target trait. It is then possible to verify whether the trait is more activated in the inference version than in the rearranged version. Because the trait-implying sentence and control versions have similar words, word-based priming effects should be of similar magnitude in both sentences. Any differences in the activation of the trait between the two sentences would be attributed to the processing of the meaning of the sentence as a whole, that is, to text-based inferences. The set of materials which will be presented in this paper is essential for the implementation of this second solution.

As previously said, neither the problem, nor the solution, are new for the text comprehension researchers (McKoon & Ratcliff, 1986; Potts, Keenan, & Golding, 1988). For example, in order to study the occurrence of predictive inferences, McKoon and Ratcliff (1986) used a probe recognition task during which the participants were presented with paragraphs containing predicting sentences (“The director and the cameraman were ready to shoot close-ups when suddenly the actress fell from the 14th story”, where “dead” should be the inferred prediction) and paragraphs containing rearranged sentences (“Suddenly the director fell upon the cameraman, demanding close-ups of the actress on the 14th story.” where “dead” should not be inferred any longer). Note that roughly the same words were used in both sentences. After reading the last sentence in the paragraph, a target word was presented. In critical trials, the participants had to indicate, as quickly and accurately as possible, whether the target word (the predictive event “dead” in the example above) was part of the paragraph. If the predicted event was inferred during the reading of the predictive sentence, then the correct response (“no”) should be slower and more inaccurate, than in the case in which the predicted event followed a rearranged sentence. This was exactly what was found. In this case, results cannot be accounted for by word-based associations because both sentences have approximately the same words.


The word-based priming problem in STI research

In this section, we present the main paradigms used to study STIs and discuss the potential role of word-based priming on the results obtained with each paradigm. The first paradigm used in the STI field was the cued-recall paradigm (Claeys, 1990; Uleman, Moskowitz, Roman, & Rhee, 1993; Winter & Uleman, 1984; Winter, Uleman, & Cunniff, 1985). In this paradigm, participants are presented with trait-implying behaviors, under memory instructions (i.e., they are asked to memorize the material for a later unspecified memory test). After a distractor task, participants are asked to recall the previous behaviors under various cue conditions. The central assumption is that if the trait is inferred during the encoding of the sentence, it will be an effective cue for the recall of the behavior, compared to a no-cue condition.

Although none of the studies with the cued-recall paradigm used rearranged sentences, Winter and Uleman (1994) expressed concern about the potential interference of word-based associations. These authors conducted a series of rigorous pre-tests which controlled for some of the word-based activations in the sentences. For example, they controlled for semantic associations with the actors of the sentences, who were described via their occupations (e.g., reporter, librarian), by eliminating those actors who were strongly associated with the target trait. The authors wanted to rule out the possibility of traits presented at test activating the actor in the sentence based on semantic backward associations, thus increasing the likelihood of retrieving the sentence, independently of any inference processes which might be taking place at encoding. Critically, while these researchers pre-tested the materials extensively, they did not explore all the possible associations between the trait and the words in the sentences, remaining inconclusive the extent to which word-based associations affect the results in the cued-recall task. This problem could have been overcome by using rearranged sentences to test whether the recall of the trait-implying sentence is better than the recall of the rearranged version, when the trait is provided as a cue.

Other common paradigms which have been applied to explore STI are the savings in relearning (Carlston & Skowronski, 1994) and the false recognition (Todorov & Uleman, 2002). In both paradigms participants are initially presented with a series of photos of actors, each one paired with a trait-implying behavior. In the false recognition task, the previously seen photos are presented at test, paired with trait-words. Some of the pairs are created by pairing the actor with the trait previously implied by his behavior (match pairs), while other pairs are created by presenting the actor with a trait previously implied by the behavior of another actor (mismatch trials). The participant is instructed to indicate whether the trait was part of the sentence presented previously with that actor. A higher rate of false recognitions of the trait in the match than in the mismatch condition is taken as evidence of occurrence of STI during the encoding of the behavior, that is, as evidence that the trait became attached to the actor’s representation in memory. In the savings paradigm, in a second phase, participants are presented with face-trait pairs. Some of these pairs are “relearning pairs” (faces are paired with the corresponding implied traits) while others are new (faces are paired with new traits). In a final phase, the photos from the second phase are presented as cues and participants have to recall the corresponding traits paired with the photos. Results typically show that the recall of the traits is superior for relearning pairs, than for new pairs (the so called “savings effect”), indicating that participants had inferred the traits in the initial phase of the study. The false recognition and the savings paradigms are currently the most frequently used methods in STI research. However, studies using these paradigms never included rearranged control versions of the trait-implying sentences. As such, the degree to which performance in these tasks is affected by specific intra-lexical associations between the to-be-inferred traits and words from the sentences remains unclear.

To our knowledge, the only STI study which originally included rearranged sentences was conducted by Uleman and collaborators, with the recognition probe paradigm (Uleman, Hon, Roman, & Moskowitz, 1996). The recognition probe paradigm was borrowed by the authors from the text comprehension literature (e.g., McKoon & Ratcliff, 1986) and has been popular since then in the STI literature (e.g., Ham & Vonk, 2003; Newman, 1991, 1993; Ramos, Garcia-Marques, Hamilton, Van Acker, & Ferreira, 2012; Uleman et al., 1996; Van Overwalle, Drenth, & Marsman, 1999; Wigboldus, Dijksterhuis, & Van Knippenberg, 2003; Wigboldus, Sherman, Franzese, & Van Knippenberg, 2004). Uleman and collaborators (1996) first applied this paradigm to the study of STI, by using both types of sentences: trait-implying sentences (“He took his first calculus course when he was 12 years old” – a sentence which implies the trait “smart”) and control sentences which used roughly the same words but do not imply the trait (“He took his first calculus course when he was 42 years old.” – a control sentence which should not imply the trait “smart”). Results showed that participants made more errors (Experiment 1) or took more time to provide a correct response (Experiment 2) after reading trait-implying, than control rearranged, sentences.

Note, however, that Uleman and colleagues’ paper (1996) is an exception and not the rule. As far as we know, none of the following studies (Ham & Vonk, 2003; Newman, 1991, 1993; Ramos et al., 2012; Uleman et al., 1996; Van Overwalle et al., 1999; Wigboldus et al., 2003, 2004) using the recognition probe paradigm have included rearranged controls, making it impossible to know whether they were dealing with real trait inferences or not. The control trials in many of the recent studies using the probe paradigm are neutral paragraphs with the only requirement being to not imply the trait, but without controlling for the individual words included. This kind of neutral control does not lead to the inference of the trait from the sentence as a whole, however it does not allow researchers to disentangle real inference from word-based priming effects.


A possible solution: The use of control sentences

In the present work, we present a set of materials which researchers can use in future studies in order to control for the impact of word-based priming effects on STI occurrence. We created a set of trait-implying behavioral descriptions and control versions of those descriptions. The control versions were created in such a way that the control sentence incorporates as many words from the trait-implying sentence as possible.

Three different pre-tests were conducted to create these materials. In the first pre-test, the goal was to create a large set of pairs of sentences and traits. Participants were presented with a list of traits and were instructed to generate behaviors representative of each trait. We then created rearranged controls for the trait-implying sentences. In the second pre-test, we tested how much the traits were being inferred from the created, rearranged versions. This allowed us to select only those pairs of sentences in which the rearranged counterparts did not lead to the inference of the critical traits. Finally, in the third pre-test, we analyzed how much the traits were implied in the trait-implying sentences when compared to the rearranged controls, and also how these two types of sentences differ in terms of ease of comprehension.


Pre-test 1: Creation of trait-implying and rearranged sentences

In this initial pre-test, the purpose was to obtain trait-implying behavioral sentences representative of a large set of personality traits. In order to create a diversified list of trait-implying behavioral descriptions, a list of 223 traits was initially compiled. The personality traits were translated from Norman Anderson’s list (Anderson, 1968) but traits which we considered to be less frequent in the Portuguese language were not included and other new traits considered common in the Portuguese language, but that were not part of Anderson’s list, were included. Two hundred and ninety-three subjects (115 males) took part in the pre-test. The average age of the sample was 29.31 years old. The pre-test was conducted online, using Qualtrics Survey Software and the participants were recruited using social media tools (e.g., Facebook groups dedicated to data collection for social research) and email invitations. Each participant was presented with 15 personality traits randomly chosen from the initial list of traits and their task was to generate a representative behavior for each one of the presented traits. Participants were instructed to think about people they knew and to give concrete examples of their behaviors. Participants were also instructed to: take no more than 1 minute and 30 seconds per trait; avoid using adjectives; and be as specific as possible in their behavioral descriptions.

Two independent judges analyzed the collected data. Each judge received half of the generated behavioral descriptions. The first step of the analysis consisted in eliminating answers which were not behavioral descriptions (e.g., definitions of the trait or traits’ synonyms), as well as redundancies between participants’ answers. Then, the judges selected 2 or 3 behavioral descriptions which better illustrated each trait (e.g., the sentence “She was in favor of the marriage between people of the same sex.” was generated for the trait “open-minded”), and in cases in which none of the descriptions were behaviors, the judges created a sentence for that trait. Traits with similar behavioral descriptions, usually synonyms, were grouped under the same trait label, such “chata” (“boring”) and “enfadonha” (“dull”). This grouping resulted in a total of 154 traits and their corresponding behavioral descriptions.

Two new judges received half of the stimuli each (77 pairs) and were asked to select the best behavioral description for each trait. At this point, the main concern was to choose the sentence which implied the trait the most. These same two judges also created rearranged control versions for each of the 154 trait-implying sentences. Critically, an effort was made to use all the words from the trait-implying sentences in their rearranged versions. In some sentences keeping exactly the same words was easier (e.g., the rearranged sentence “A nota mais alta que conseguiram tirar na sua cadeira no semestre passado foi 14” of the trait-implying sentence “A nota mais alta que conseguiram tirar na sua cadeira no semestre passado foi 19.”, which implies the trait “exigente”) than in others (e.g., the rearranged sentence “Este ano passou um mês a viajar sozinha.” of the trait-implying sentence “Para não ir sozinha, adiou a viagem para o próximo mês deste ano.” which implies the trait “aventureira”).


Pre-test 2: Rearranged control sentences

The 153 rearranged sentences were presented to four new independent judges. Two of the judges (group A) were asked to write down the first word which came to their mind when reading the sentences and the other two (group B) were asked to evaluate to what extent the sentences presented were related to the critical traits by using a 9-point scale ranging from not related (1) to very related (9). If at least one of the judges from group A generated the target trait or a word which was related to the trait (“esperançoso” when the trait inferred was “optimista”), that pair of rearranged/implying sentences was excluded from the set (based on this criterion, 14 % of the material was excluded). In addition, from the remaining material, if both judges from group B rated the relation between the rearranged sentence and the trait with a rating higher than 5, that pair of sentences was also excluded from the set (8 % of excluded material based on this criterion). This resulted in 122 pairs of trait-implying/rearranged sentences, which are presented in Table 1. In this pre-test the inter-judges reliability in Group B was low, ICC=.360, 95 % CI [.008, .577], and that motivated the next pre-test.


Pre-test 3: Comparing the trait-implying and the rearranged control sentences

Four new independent judges were presented with half of the pairs of sentences (66 trait-implying sentences and 66 rearranged sentences) and the correspondent traits. For each sentence-trait pair, the judges were asked to indicate how well the trait described the person performing the behavior. They were instructed to use a scale ranging from 1 (“the trait does not describe the person at all”) to 9 (“the trait describes very well the person”). Moreover, for each sentence, the judges had to indicate how easy it was to comprehend the sentence, again by using a 9-point scale (1 – not easy at all, 9 – very easy). Each judge was presented with one of the two sentences (the trait-implying or the rearranged), related to the same trait. Thus, 2 judges were presented with a set of sentences (Group C) and the other 2 with a different set (Group D). For the trait ratings, in both Group C, ICC=.868, 95% CI [.811, .907], and Group D, ICC=.818, 95 % CI [.735, .874], we obtained a high inter-rater reliability. As expected, the ratings for how much the trait describes the person in the rearranged sentence are much lower (M=2.84, SD=2.08) than in the trait-implying sentence (M=8.19, SD=1.07), t(121)=22.87, p<.001. For the comprehension, in both Group C, ICC=.354, 95% CI [-.042, .590], and Group D, ICC=.386, 95 % CI [.120, .572], the inter-rating reliability was low. The comprehension ratings for the rearranged sentences (M=7.37, SD=1.88) were significantly lower than for the trait-implying sentences (M=8.68, SD=.70), t(121)=6.71, p<.001. Even though different, the ratings for comprehension are above 5 in both conditions. Table 2 contains the average ratings for each pair of stimuli (trait-rearranged sentence and trait-implying sentence). When using this material, we recommend researchers to select those pairs with similar levels of comprehensibility for the rearranged and the trait-implying sentences.



Evidence for STI has been extensively obtained. However, it is difficult from the existing literature to conclude whether previous results are being influenced by simple word-based priming effects. This is particularly problematic because trait-implying sentences frequently contain individual words which are strongly related to the critical traits, and the presence of these words can activate the traits, regardless of text-based inferences. In the present paper, we have suggested that one efficient way of overcoming this confounder is by comparing the trait-implying sentences with control sentences containing the same words. Any differences in the performance found between the two sentences should be due to real inferences based on processing of the entire meaning of the sentences. We recommend the use of these pairs of sentences in future studies investigating trait inferences in the Portuguese language. It should be noted that we consider the present paper as a first step in approaching this issue. Future studies might be necessary to further test the present material with larger samples, and regarding other variables, such as valence, familiarity, or ease of comprehension (since our results were not totally clear regarding the latter).

One aspect of the material presented here that should be considered is that the rearranged sentences vary in a critical aspect. Sometimes, the resulting rearranged sentence does not imply any trait, that is, neutral. This is the case, for example, of the control sentence “She was singing a popular song on her way to work.” (corresponding to the trait-implying sentence “It was playing a popular song on the radio on her way to work.”), which implies the trait “joyful”. However, other times the rearranged version implies a different trait, which can even be opposite in meaning to the critical trait implied in the trait-implying sentence. For example, the control version for the sentence: “She was in favor of marriage between people of the same sex.”, which implies the trait “open-minded”, is the sentence “She was not in favor of sex before people marry.”, which implies the opposite trait “closes-minded”. This should not be a major problem as long as the critical trait is not activated by the control-rearranged sentence, since that is the trait we are controlling for. If a different trait is being inferred from the rearranged version, this sentence is still a control for the trait inferred in the trait-implying sentence. If the trait activated by the control sentence is the antonym of the trait activated in the trait-implying sentence, the differences in the activation of the critical trait between the two sentences will be even more pronounced than it would be in the case where the rearranged sentence is neutral. One can argue that in such a situation the difference between the two sentences is overestimated, and ultimately one can choose to use only neutrally rearranged controls. And even though we don’t find it problematic to use this kind of controls, we provide in Appendix 1 a list of the traits for which at least one of the two judges, from the second pre-test, generated a trait which is opposite in meaning to the critical trait implied in the trait-implying sentence. This was verified only for 15 traits (the “open-minded” example mentioned above is one of them).

To sum up, our first recommendation for future studies is to include rearranged sentences, besides the trait-implying ones. However, even when including rearranged sentences, because the calculation of STI effects relies on the difference between critical and control sentences, we should guarantee beforehand that word-based activation does not play a significant role in the material. Not controlling the material for this aspect before the actual experiment carries the risk of underestimating effects or obtaining false negatives (e.g., strong associates can lead to a ceiling effect). Thus, our second recommendation is to test the material before the experiment, and only select those pairs of trait-implying and rearranged sentences in which the activation of the trait is significantly stronger in the critical sentence when compared with its control version. In order to pretest the material, we suggest the use of activation measures such as the lexical decision task (e.g., Zárate, Uleman, & Voils, 2001). In the lexical decision task, words (some of them traits) and non-words are presented to the participants, and the activation of the trait can be accessed via the speed of the decision when the trait-implying sentence versus rearranged sentence precedes the presentation of the trait word. If traits are more accessible after reading the trait-implying version than after reading the rearranged version, then it can be deduced that participants generated “true” trait inferences from the trait-implying sentences, beyond any word-based priming effects.

Finally, an important aspect which is still unclear is how word-based priming processes might differentially impact performance in different tasks used to study STI. For that, it would be necessary to clearly understand the mechanisms underlying word-based priming and text-based priming, as well as how both processes interact during text comprehension. For example, if word-based priming is a shorter-lived effect and text-based priming has a longer duration, then the effect of word-based priming would be particularly problematic for activation measures as the probe recognition paradigm. In contrast, memory measures, as the savings in re-learning and the false recognition tasks, should be less affected, or maybe not affected at all, by this contamination. Unfortunately, although there are models which can be used to make predictions about the processes underlying lexical and text level effects (Kintsch, 1988; see also Sharkey & Sharkey, 1992), we still don’t understand how lexical activation emerging from single words interacts with a more elaborate and holistic processing of the text (Stafura & Perfetti, 2014). Thus, knowing to what degree this confounder influences performance in different STI tasks remains an empirical question which needs to be carefully addressed in future studies.

To conclude, the present paper is the first one which explicitly addresses the role of word-based priming in STI studies, discusses how this issue might affect and distort the interpretation of previous findings, and proposes specific steps which should be taken by future research in order to avoid this problem.



Anderson, N. H. (1968). Likableness ratings of 555 personality trait words. Journal of Personality and Social Psychology, 9, 272-279.         [ Links ]

Carlston, D. E., & Skowronski, J. J. (1994). Savings in the relearning of trait information as evidence for spontaneous inference generation. Journal of Personality & Social Psychology, 66, 840-856.         [ Links ]

Claeys, W. (1990). On the spontaneity of behaviour categorization and its implications for personality measurement. European Journal of Personality, 4, 173-186.         [ Links ]

Forster, K. I. (1981). Priming and the effect of sentence and lexical contexts on naming time: Evidence for automatic lexical processing. Quarterly Journal of Experimental Psychology, 33, 465-495.         [ Links ]

Ham, J., & Vonk, R. (2003). Smart and easy: Co-occurring activation of spontaneous trait inferences and spontaneous situational inferences. Journal of Experimental Social Psychology, 39, 434-447.         [ Links ]

Keenan, J. M., & Jennings, T. M. (1995). The role of word-based priming in inference research. In R. F. J. Lorch & E. O’Brien (Eds.), Sources of coherence in reading (pp. 37-50). Hillsdale, NJ: Lawrence Erlbaum Associates.

Keenan, J. M., Potts, G. R., Golding, J. M., & Jennings, T. M. (1990). Which elaborative inferences are drawn during reading? A question of methodologies. In D. A. Balotta, G. B. Flores d’Arcais, & K. Rayner (Eds.), Comprehension processes in reading (pp. 377-402). Hillsdale, NJ: Erlbaum.

Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction-integration model. Psychological Review, 95, 163-182.         [ Links ]

McKoon, G., & Ratcliff, R. (1986). Inferences about predictable events. Journal of Experimental Psychology: Learning, Memory, and Cognition, 12, 82-91.         [ Links ]

Newman, L. S. (1991). Why are traits inferred spontaneously? A developmental approach. Social Cognition, 9, 221-253.         [ Links ]

Newman, L. S. (1993). How individualists interpret behavior: Idiocentrism and spontaneous trait inference. Social Cognition, 11, 243-269.         [ Links ]

Potts, G. R., Keenan, J. M., & Golding, J. M. (1988). Assessing the occurrence of elaborative inferences: Lexical decision versus naming. Journal of Memory and Language, 27, 399-415.         [ Links ]

Ramos, T., Garcia-Marques, L., Hamilton, D., Van Acker, K., & Ferreira, M. (2012). What I infer depends on who you are: The influence of stereotypes on trait and situational spontaneous inferences. Journal of Experimental Social Psychology, 48, 1247-1256.         [ Links ]

Sharkey, A. J. C., & Sharkey, N. E. (1992). Weak contextual constrains in text and word priming. Journal of Memory and Language, 31, 543-572.         [ Links ]

Skowronski, J. J., Carlston, D. E., Mae, L., & Crawford, M. T. (1998). Spontaneous trait transference: Communicators take on the qualities they describe in others. Journal of Personality and Social Psychology, 74, 837-848.         [ Links ]

Stafura, J. Z., & Perfetti, C. A. (2014). Word-to-text integration: Message level and lexical level influences in ERPs. Neuropsychologia, 64, 41-53.         [ Links ]

Todorov, A., & Uleman, J. S. (2002). Spontaneous trait inferences are bound to actors’ faces: Evidence from a false recognition paradigm. Journal of Personality & Social Psychology, 83, 1051-1065.

Uleman, J. S., Hon, A., Roman, R. J., & Moskowitz, G. B. (1996). On-line evidence for spontaneous trait inferences at encoding. Personality and Social Psychology Bulletin, 22, 377-394.         [ Links ]

Uleman, J. S., Moskowitz, G. B., Roman, R. J., & Rhee, E. (1993). Tacit, manifest, and intentional reference: How spontaneous trait inferences refer to persons. Social Cognition, 11, 321-351.         [ Links ]

Uleman, J. S., Rim, S., Adil Saribay, S., & Kressel, L. M. (2012). Controversies, questions, and prospects for spontaneous social inferences. Social and Personality Psychology Compass, 6, 657-673.         [ Links ]

Van Overwalle, F., Drenth, T., & Marsman, G. (1999). Spontaneous trait inferences: Are they linked to the actor or to the action?. Personality and Social Psychology Bulletin, 25, 450-462.         [ Links ]

Wigboldus, D. H. J., Dijksterhuis, A., & Van Knippenberg, A. (2003). When stereotypes get in the way: Stereotypes obstruct stereotype-inconsistent trait inferences. Journal of Personality & Social Psychology, 84, 470-484.         [ Links ]

Wigboldus, D. H. J., Sherman, J. W., Franzese, H. L., & Van Knippenberg, A. (2004). Capacity and comprehension: Spontaneous stereotyping under cognitive load. Social Cognition, 22, 292-309.         [ Links ]

Winter, L., & Uleman, J. S. (1984). When are social judgments made? Evidence for the spontaneousness of trait inferences. Journal of Personality and Social Psychology, 47, 237-252.         [ Links ]

Winter, L., Uleman, J. S., & Cunniff, C. (1985). How automatic are social judgments?. Journal of Personality and Social Psychology, 49, 904-917.         [ Links ]

Zárate, M. A., Uleman, J. S., & Voils, C. I. (2001). Effects of culture and processing goals on the activation and binding of trait concepts. Social Cognition, 19, 295-323.         [ Links ]



Correspondence concerning this article should be addressed to: Diana Orghian, Massachusetts Institute of Technology, The MIT Media Lab, 20 Ames Street, E15-381, Cambridge, MA, USA. E-mail:


Acknowledgments: We thank Helena Palmieri for editing the English of the manuscript.


Submitted: 28/07/2016 Accepted: 28/05/2017

Creative Commons License All the contents of this journal, except where otherwise noted, is licensed under a Creative Commons Attribution License