Mechanisms of word concreteness effects in explicit memory: Does context availability play a role?

Abstract

One explanation for why concrete words are better recalled than abstract words is systematic differences across these word types in the availability of context information. In contrast, explanations for the concrete word advantage in recognition memory do not consider a possible role for context availability. We investigated the extent to which context availability can explain the effects of word concreteness in free recall (Experiment 1) and item recognition (Experiment 2) by presenting each target word in isolation, in a low-constraint sentence context, or in a high-constraint sentence context at study. Concreteness effects were consistent with previous research, with concrete word advantages in both tasks. Embedding words in sentence contexts with low semantic constraint hurt recall performance but helped recognition performance relative to presenting words in isolation. Embedding words in sentence contexts with high semantic constraint hurt both recall and recognition performance relative to words in low-constraint sentences. The effects of concreteness and semantic constraint were consistent for high- and low-frequency words. Embedding words in high-constraint sentence contexts neither reduced nor eliminated the concreteness effect in recall or recognition, indicating that differences in context availability cannot explain concreteness effects in explicit memory.

Verbal memory performance is known to vary as a function of semantic and lexical properties of the words to be remembered. Word concreteness has marked effects on explicit memory tasks, including free recall and recognition (e.g., Balota & Neely, 1980; Glanzer & Adams, 1985; Holmes & Langford, 1976), but there is no consensus on the mechanisms underlying these effects. In the present study, we examine whether context availability contributes to the effects of word concreteness in explicit memory. Specifically, we examine whether context manipulations at encoding moderate the effects of concreteness on free recall and item recognition performance. We also examine the impact of these encoding contexts and semantic constraint on explicit memory for words.

Word Concreteness Effects in Explicit Memory

The concreteness of a word is the extent to which its referent can be experienced by the senses (Walker & Hulme, 1999). For example, a concrete word like house represents a tangible object whereas an abstract word like truth represents an intangible quality or state of being. Concrete words also more readily invoke familiar mental images than do abstract words. In the current study, we focus on effects of concreteness as defined by concreteness ratings from published sources.

The general finding that concrete words have an advantage over abstract words is known as the concreteness effect. This finding has been reported in memory tasks such as free, serial, and cued recall (e.g., Holmes & Langford, 1976; Richardson, 2003; Romani, McAlpine, & Martin, 2008; Walker & Hulme, 1999) and recognition (e.g, Glanzer & Adams, 1985; Glanzer, Adams, Iverson, & Kim, 1993; Hirshman & Arndt, 1997), as well as in lexical processing tasks (e.g., Schwanenflugel, Harneshfeger, & Stowe, 1988; van Hell & de Groot, 1998, 2008). However, the mechanisms and processes that underlie this advantage remain a subject of debate.

One of the most well known explanations of the concreteness effect is dual coding theory (for a review see Paivio, 1991). This account details two processing systems, a verbal system and an image system. Under this theory, it is assumed that concrete words are processed in both systems whereas abstract words are processed only in the verbal system. As a result of being processed in two systems, and thus having two possible retrieval routes, concrete words are more easily retrieved than abstract words. One challenge to this model was that it could not explain why the concreteness effect was eliminated when stimuli were embedded within contextually rich materials (Wattenmaker & Shoben, 1987). Specifically, concrete target sentences were better recalled than abstract sentences when presented in random orders, however when placed in a coherent paragraph format there was no difference in the recall of concrete and abstract sentences.

This limitation of dual coding theory was the impetus for the development of the context availability framework (Kieras, 1978; Schwanenflugel, Akin, & Luh, 1992; Schwanenflugel et al., 1988; Schwanenflugel & Shoben, 1983; van Hell & de Groot, 1998, 2008). This account posits that comprehension is aided by access to contextual information, such as environmental information, semantic information, or content that serves to provide detail and specify the meaning of the word or the situation in which it is encountered. Episodic contexts are encoded along with the word itself. Benefits of context can come from accessing prior exposure contexts or processing a new context (Schwanenflugel et al., 1988). When it becomes easy to access features of an episodic context, the word has high context availability.

Under the context availability account, the concrete advantage in lexical processing and memory arises due to differential availability of contextual information for concrete and abstract words. Indeed, rated concreteness and rated context availability are highly correlated (De Groot, Danenburg, & Van Hell, 1994). One possible reason for this relationship is that abstract words (of comparable frequency) are associated with more different episodic contexts than concrete words (Schwanenflugel & Shoben, 1983). Because the associations of abstract words are divided across more different contexts, these associations are generally weaker than the associations of concrete words. It is therefore more difficult to access contextual information for abstract words than for concrete words. For example, a concrete word like microscope is almost exclusively with a scientific context, so this association is strong. However, an abstract word like boredom is associated with a variety of contexts (e.g., waiting in line at the department of motor vehicles, in class listening to a monotonous professor, being stuck in traffic, being forced to watch golf on TV, etc.), so each individual contextual association is weaker. This difficulty in accessing contexts associated with abstract words leads to lexical processing and memory advantages for concrete words.

Evidence for the Context Availability Account

The reasoning of the context availability account has been supported by findings showing that the concreteness effect is reduced or eliminated when concrete and abstract words have equal access to contextual information. Two approaches have been used to equalize access to contextual information and test the context availability effect. One approach was to control for participant-rated context availability across word types. Context availability ratings of concrete and abstract words were obtained by asking participants to rate, on a Likert scale, the ease with which they could think of a context or circumstance where the words would be used (e.g., Schwanenflugel et al, 1988). Matching concrete and abstract words on context availability eliminated concreteness effects in lexical processing tasks (Schwanenflugel et al., 1988; van Hell & de Groot, 1998). However, in free recall, matching context availability for concrete and abstract words produced mixed results. In one experiment with intentional learning as the encoding task, controlling for context availability did not eliminate the advantage for concrete words (Schwanenflugel et al., 1992). However, in a second experiment controlling for context availability eliminated the concreteness effect only when the encoding task was to judge the context availability of each word. With intentional learning or imagery instructions, the concrete advantage remained.

The second approach to reducing context availability differences across concrete and abstract words has been to embed the words in the context of sentences or paragraphs. Providing such contexts impacts both lexical processing and memory, and performance varies with the degree of semantic constraint provided by the sentence. Semantic constraint can be thought of as the degree to which a sentence biases processing towards a particular target word. High-constraint sentence frames lead to high predictability and a high probability that a reader will complete the sentence with the intended target word. For example, the sentence; He got the leash and collar to take a walk with his ________ is highly constraining towards the target dog. On the other hand the sentence; The woman needed to go to the store to buy some more ________ is low-constraint because there are many potential targets that could fit with that frame. Semantic constraint in a sentence generates a set of semantic features that increases activation of the target or reduces competition by decreasing activation of competitors, thus facilitating identification and comprehension of the target word (e.g., Griffin & Bock, 1998; Schwanenflugel & LaCount, 1988; Schwanenflugel & Shoben, 1985). Manipulations of semantic constraint serve as a method to increase priming and/or reduce lexical competition in lexical processing. Embedding target words in a sentence context rather then presenting them in isolation speeds both comprehension and production, and this effect is stronger with a semantically constraining sentence context rather than a neutral or incongruent context (e.g., Balota, Pollatsek & Rayner, 1985; Forster, 1981; Gollan et al., 2011; Griffin & Bock, 1998; Rayner, Ashby, Pollatsek, & Reichle, 2004).

Embedding words within a sentence context also impacts explicit memory performance, but the impact of semantic constraint is less clear. Nouns presented in sentence contexts were less well recalled than nouns presented in isolation (Cofer, 1968; Wood, 1970). In a study where the encoding task was to judge whether target words at the end of sentences made sense in their sentence contexts, target words were better recalled when presented in high-constraint sentences than when presented in low-constraint sentences (McFalls & Schwanenflugel, 2002). In recognition, studying words in sentence contexts diminished performance relative to studying words presented in isolation (Schwartz, 1975). Embedding high-frequency words in the context of experiment instructions, an incidental encoding task, reduced recognition relative to isolated presentations, but low-frequency words were equally well recognized after contextualized or isolated presentation (Coane & Balota, 2010). We could find no studies that examined the effects of semantic constraint on later recognition of target words. Thus, with only one study of semantic constraint effects on recall and no studies of its effects on recognition, little is known about how semantic constraint impacts explicit memory.

Providing sentence or paragraph contexts, particularly contexts with high semantic constraint, reduced or eliminated concreteness effects in lexical processing tasks (Schwanenflugel & Shoben, 1983; Van Hell & de Groot, 2008). For concrete words, prior contexts are easily accessed, so the current (experimental) context does not increment accessibility as much as for abstract words where it is more difficult to access a prior context (Schwanenflugel et al., 1988). Providing a sentence context just before the target word makes it easier to access an appropriate context. The disadvantage for abstract words is reduced or eliminated, because it becomes easy to access contextual information for both concrete and abstract words. For either word type, there is the option to simply access the sentence context that is still in working memory. The availability of a current experimental context provides a mechanism for closing the accessibility gap between abstract and concrete words. For the same reasons, when this context is more specific and salient as in high-constraint sentences, the benefit will be greater for abstract words than for concrete words, closing the accessibility gap. It is unknown whether this differential facilitation carries over into the encoding and/or retrieval processes of explicit memory.

The concreteness effect persists in free recall for concrete and abstract sentences. In one study, sentences containing concrete nouns, such as Adult elephants are protected by strong skins, were better recalled than sentences containing abstract nouns, such as Large companies are regulated by strict rules. Not only did participants recall fewer abstract sentences, but they also recalled fewer specific words from abstract sentences (Holmes & Langford, 1976). In another study of sentence recall, when concrete and abstract sentences were organized in a random order there was a concrete advantage, but when sentences were presented in a coherent paragraph context, abstract and concrete sentences were equally well recalled (Marschark, 1985; Wattenmaker & Shoben, 1987). One possible explanation for these results is that while constraint levels were neither manipulated nor controlled, a coherent paragraph context provided the constraint necessary to equate context availability for concrete and abstract sentences and thereby eliminate the concreteness effect. Because semantic constraint was not manipulated at the sentence or paragraph level and the focus was on recalling entire sentences rather than individual concrete or abstract words, these studies provide only limited evidence for the role of context availability in the concreteness effect in explicit memory.

The Present Study

The primary motivation of the present study was to test whether the context availability hypothesis of the concreteness effect can be appropriately extended to explicit memory. Although the context availability hypothesis has considerable support in lexical processing, its support in explicit memory is mixed and limited. When pre-experimental context availability was matched for concrete and abstract words, the only condition that eliminated the concreteness effect was when the encoding task was to judge context availability itself (Schwanenflugel, et al., 1992). The only studies of how sentence context manipulations would impact the concreteness effect used paragraph level manipulations with concrete and abstract sentences (Marschark, 1985; Wattenmaker & Shoben, 1987). These studies did not include manipulations of sentence constraint, which were important for tests of the context availability hypothesis in lexical processing. Therefore, it remains unknown whether studying words in high-constraint sentences would reduce the concreteness effect in the recall of individual target words. Experiment 1 tested the context availability hypothesis by examining the effects of concreteness in recall of individual words that were studied either in isolation, in low-constraint sentences, or in high-constraint sentences.

Because concreteness effects extend to recognition memory, and there have been no previous tests of the context availability hypothesis as an explanation for concreteness effects in recognition memory, we wanted to perform this test. Experiment 2 tested the context availability hypothesis by examining the effects of concreteness in recognition of words that were studied in isolation, in low-constraint sentences, or in high-constraint sentences. Under the context availability hypothesis, providing a context, particularly a high-constraint context, would disproportionately increase context availability for abstract words, thus closing the context availability gap between concrete and abstract words,. Closing this gap by providing a sentence context, particularly a high-constraint sentence context would in turn reduce or eliminate the concreteness effect typically observed for isolated words in recognition memory.

A second motivation of the study was to extend previous research on the effects of semantic constraint on explicit memory performance. While a small number of studies have shown that embedding words in a sentence context hurts recall and recognition performance, there are a number of processing differences between isolated words and sentences other than context that could potentially explain these effects. A more interesting question is whether semantic constraint impacts memory for words embedded in sentence contexts.

The only study to make such a comparison showed better free recall performance for target words at the end of high-constraint sentences than for target words at the end of low-constraint sentences (McFalls & Schwanenflugel, 2002). We contend that this result does not provide a definitive answer to this question for two reasons. First, the encoding task was to judge whether each target word made sense in its sentence context, which draws the focus of attention to the sentence context and therefore the constraint manipulation, because the words would clearly make more sense in the high-constraint contexts than in the low-constraint contexts. Second, the congruent sentences in which the target words made sense were mixed with incongruent sentences in which the target words did not make sense, and this mixture could have changed how the congruent sentences were processed. In Experiment 1, we make this comparison in free recall using a more neutral encoding task, simply reading the sentences aloud and attempting to commit the final target words to memory, and using only sentences in which the target words made sense. Whereas recall is thought to depend to a great degree on inter-item processing, recognition is thought to depend more on intra-item processing (e.g., Hunt & Einstein, 1981). Because semantic constraint seemed likely to alter the balance of inter-item (between-sentence) and intra-item (within-sentence) processing, we thought it would be informative to know whether the effects of semantic constraint on recall and recognition are the same or different. We therefore examined the effects of semantic constraint on recognition memory for the first time in Experiment 2.

Although not the primary focus of the study, we included a word-frequency manipulation for three reasons. First, in the only prior study of context effects in recognition memory, context hurt recognition of high-frequency words but not low-frequency words (Coane & Balota, 2010). This means that it was important to consider whether any context effects found in the present study would also interact with word frequency and whether they would generalize across words of different frequency levels. Second, semantic constraint is known to reduce word-frequency effects in word production tasks (Gollan et al., 2011; Griffin & Bock, 1998) but not in comprehension tasks (e.g., Balota, Pollastek & Rayner, 1985; Forster, 1981; Gollan et al., 2011; Rayner et al., 2004). If we consider the encoding task to be a comprehension task, recall to be a production task, and recognition to be a comprehension task, this pattern of effects suggests that semantic constraint would reduce the frequency effect in recall but not in recognition. If word frequency and semantic constraint indeed interact for recall but not recognition, the implication would be that semantic constraint impacts not only how words are encoded but also how words are retrieved. On the other hand, if word frequency and semantic constraint do not interact for either test task, the implication would be that semantic constraint impacts encoding but not retrieval. Third, the effects of word frequency on recall and recognition go in opposite directions, with high-frequency words being better recalled and low-frequency words being better recognized (e.g., Balota & Neely, 1980; Kinsbourne & George, 1974; MacLeod & Kampe, 1996; Mandler, Goodman, & Wilkes-Gibbs, 1982). These three reasons make it interesting to include a word frequency manipulation to enable tests of its possible interactions with context and constraint, how well effects of context and constraint generalize across more and less familiar words, and how these effects might differ across recall and recognition tasks.

Thus, the goals of the study were to 1) test the context availability hypothesis of the concreteness effect in explicit memory, 2) examine the effects of semantic constraint in explicit memory, and 3) find out whether the effects of context in explicit memory depend on word frequency. These goals were addressed in two experiments that included manipulations of context (isolated words, low-constraint sentences, or high-constraint sentences), concreteness, and word frequency. In each experiment, the study task was to read words or sentences aloud and attempt to commit the target words to memory. The test task was free recall in Experiment 1 and item recognition in Experiment 2.

Experiment 1

Experiment 1 examined the effects of word concreteness, word frequency, and sentence constraint in a free recall task. Previous work has shown that controlling for participant-rated context availability can reduce the concreteness effect in recall tasks, and providing participants with sentences to serve as context can reduce the concreteness effect in lexical processing. Thus, the primary goal of Experiment 1 was to determine whether studying words in high-constraint sentence contexts would reduce the concreteness effect in free recall of individual target words. It was hypothesized that the concreteness effect would be reduced or eliminated in the high-constraint condition as previously reported for lexical processing tasks (van Hell & de Groot, 2008) but remain in the low-constraint conditions as in previous research with lexical processing tasks (van Hell & de Groot, 2008) and free recall (Holmes & Langford, 1976; Marschark, 1985; Wattenmaker & Shoben, 1987), thereby supporting the context availability account. A second goal was to test the effects of semantic constraint on free recall with a protocol that only included congruent sentences and had an encoding task, reading aloud, that would not draw attention to the constraint manipulation.

Method

Power and Sample Size.

We wanted to have 80% power to detect an effect of small-to-medium size (f = .175), which would require 67 participants. Because complete counterbalancing of the assignment of items to contexts and presentation orders required a multiple of 36 participants, 72 participants were recruited.

Participants.

Participants were 72 students (45 women, 27 men) from the University of Texas at El Paso with a median age of 20 (SD = 4.1). Participants were recruited from psychology courses and earned their choice of credit toward a course research requirement or $10. All were proficient speakers of English. Two additional volunteers began the protocol but wished to end the session before it was complete; these individuals had to be replaced because their data were incomplete.

Design.

The experimental conditions formed a 2 (concreteness) X 2 (frequency) X 3 (study context) within-subjects design. The concreteness factor divided stimuli into concrete and abstract words and the frequency factor divided stimuli into high- and low-frequency words. The context factor included words presented in isolation, words presented in low-constraint sentence frames, and words presented in high-constraint sentence frames. The dependent variable was recall proportion.

Materials.

Stimuli were selected from three published sources (Altarriba, Bauer, & Benvenuto, 1999; Miller & Roodenrys, 2009; Tokowicz & Kroll, 2007) in which normed abstract and concrete words were provided. Word frequency information for the stimuli was obtained using CELEX (Baayen, Piepenbrock, & Gulikers,1995). For items to be included as low-frequency words the reported occurrence per million had to be no greater than 40. To be included as a high-frequency word the item needed to have a reported occurrence per million of at least 70. Median word frequencies for each stimulus category are provided in Table 1.

Table 1.

Median Frequencies of Target Words and Mean CLOZE Probabilities for Stimulus Sentences

	Concrete Words		Abstract Words
Item Property	HF	LF	HF	LF
Word Frequencya	156	21	158	25
High Constraint CLOZE	.90	.82	.86	.75
Low Constraint CLOZE	.04	.02	.03	.05

Note: HF = high frequency; LF = low frequency

^aFrequency per million from CELEX (Baayen et al., 1995)

High- and low-constraint sentence frames were developed and pilot tested using a CLOZE procedure (see Griffin & Bock, 1998). Each high-constraint sentence frame was written to make the target word highly predictable. Low-constraint sentence frames were written to avoid biasing prediction toward any particular word and to make target words plausible but not predictable. It should be noted that the low-constraint sentences were not devoid of content or meaning as in some previous studies (e.g., Griffin & Bock, 1998) -- they were simply less restrictive than high-constraint sentences in the range of plausible completions. Target words were always presented as the last word of the sentence in all capital letters. Thirty-two pilot participants drawn from the same population as the experimental sample read sentence frames in which the final word had been omitted, and they were asked to type in the word that they thought would end the sentence. For high-constraint sentences, the 36 sentences of each word type with the highest proportion of expected responses were selected, along with low-constraint sentences matched as closely as possible on target frequency. Table 1 shows the mean proportion of times that participants responded with the intended target word for high-constraint and low-constraint sentences ending with target words of each type, and Table 2 shows examples of the sentences for each word type and constraint level.

Table 2.

Examples of Sentences of Each Type Used in Experiments 1 and 2

Target Concreteness	Sentence Type
Target Frequency	High Constraint	Low Constraint
Concrete Target
High-Frequency	The new kids in town were nervous for their first day of SCHOOL.	The photographer took a picture of the SCHOOL.
Low-Frequency	During the national anthem the soldiers saluted the country’s FLAG.	From where she was sitting she couldn’t see the FLAG.
Abstract Target
High Frequency	As soon as they saw each other the couple fell in LOVE.	The painter said her biggest inspiration is LOVE.
Low Frequency	The young boy wouldn’t listen to anyone because he was so STUBBORN.	Everyone knew the woman was very STUBBORN.

The final stimulus set included 144 target words (36 words for each stimulus category) that were randomly distributed across 12 lists (3 lists for each stimulus category) of 12 targets each. Assignment of lists to study context conditions was counterbalanced across participants using a Latin square.

Apparatus.

Stimuli were presented on a Macintosh computer, and the experiment was programmed using PsyScope X software (Cohen, MacWhinney, Flatt, & Provost, 1993). Participant recall responses were recorded for later transcription using a Sony IC voice recorder.

Procedure.

Participants were tested individually in sessions lasting approximately 45 minutes. After English proficiency screening, participants completed brief language and demographic background questionnaires. For the main experiment, participants sat in front of a computer in a small experiment room. The experimenter explained that the participant’s responses would be recorded for the purposes of later transcription.

Participants then completed 12 study-test cycles in which they were first presented a sequence of 12 words or sentences one at a time, and after the twelfth item, they were asked to recall as many target words as they could remember. Participants were instructed to read each target word or sentence out loud and to commit the target words to memory. The target word was always the final word of the sentence and appeared in all capital letters. Once the participant read the word or sentence aloud, they pressed a button to advance to the next item, which would appear after 500 ms; thus, the presentations were self-paced. The reason for this self-paced study was twofold. First, different participants have different reading rates, and we wanted to make sure they had time to read aloud every sentence. Second, although this aspect of the procedure made the effective presentation rates different for isolated and sentence encoding conditions, the primary reason for more accurate memory with slower presentation rates is increased rehearsal (Bhatarah, Ward, Smith, & Hayes, 2009). In the sentence conditions, the time between presentations of target words was filled with reading the sentences aloud, thus limiting rehearsal, so we did not want participants to have long gaps between words in the isolated conditions that would give them time for many more rehearsals than in the sentence conditions. The study-test cycles were blocked by context and stimulus category such that there were four cycles for each context condition (one for each stimulus category). The orders of the context conditions and stimulus types were counterbalanced across participants.

Results

Recall proportions (see Figure 1) were calculated in each condition for each participant. These values were submitted to a 2 (concreteness) x 2 (frequency) x 3 (study context) repeated-measures ANOVA. More high-frequency words were recalled than low-frequency words, F(1, 71) = 11.21, MSE = .017, p = .001, partial η² = .14, and more concrete words were recalled than abstract words, F(1, 71) = 120.83, MSE = .011, p < .001, partial η² = .63. There was a significant interaction of frequency and concreteness, F(1, 71) = 12.57, MSE = .010, p = .001, partial η² = .15, indicating that the concreteness effect was larger for high-frequency words than for low-frequency words.

Figure 1.

Recall Scores in Experiment 1 as a Function of Concreteness, Word Frequency, and Encoding Context. Error bars represent standard errors of the mean.

There was a significant main effect of study context, F(2,142) = 35.68, MSE = .015, p < .001, partial η² = .33. Planned comparisons (pairwise contrasts on the main effect) showed that targets presented in isolation were recalled more than targets in low-constraint sentences, F(1, 71) = 29.53, MSE = .007, p < .001, partial η² = .45, and high-constraint sentences, F(1, 71) = 57.11, MSE = .009, p < .001, partial η² = .29. Targets in low-constraint sentences were also recalled more than targets in high-constraint sentences, F(1, 71) = 10.98, MSE = .006, p = .001, partial η² = .13. Study context did not interact with frequency, F(2,142) = 2.64, MSE = .015, p = .075, partial η² = .04, or concreteness, F < 1, partial η² < .01. However, the three-way interaction of frequency, concreteness, and context was significant, F(2,142) = 4.07, MSE = .012, p = .019, partial η² = .05. Breaking this effect into simple interactions showed that for low-frequency words, the effect of concreteness was larger for words studied in isolation than for words studied in high-constraint sentences, F(1, 71) = 5.32, MSE = .027, p = .024, partial η² = .07, but for high-frequency words, this interaction was not reliable, F(1, 71) = 2.68, MSE = .021, p = .106, partial η² = .04. For low-frequency words, tests of simple effects showed that the concreteness effect was significant for isolated words, F(1, 71) = 20.12, MSE = .013, p < .001, partial η² = .22, and words in low-constraint sentences, F(1, 71) = 11.61, MSE = .010, p = .001, partial η² = .14, but not for words in high-constraint sentences, F(1, 71) = 2.06, MSE = .010, p = .156, partial η² = .03.

Discussion

As expected, recall proportions were higher for concrete words than for abstract words, demonstrating the typical concreteness effect. Isolated words were better recalled than words presented in the context of sentences, consistent with findings from prior research (Cofer, 1968; Wood, 1970). Words from high-constraint sentences were less likely to be recalled than words from low-constraint sentences. This result stands in contrast to the results of the only previous study to examine the effect of semantic constraint at study on free recall performance. In that study, words from high-constraint sentences were more likely to be recalled than words from low-constraint sentences (McFalls & Schwanenflugel, 2002).

Important methodological differences between the McFalls & Schwanenflugel (2002) study and Experiment 1 likely explain the different results obtained. First, and most importantly, the encoding task differed across studies. In the previous study, participants made semantic decisions about whether the target words made sense in the sentence context, so semantic processing of the target words and semantic integration of the target words with the sentences was an essential part of the encoding task and would provide additional retrieval routes. These processes would likely be enhanced for the high-constraint sentences, giving words in high-constraint sentences more possible retrieval routes than words in low-constraint sentences. In contrast, participants in Experiment 1 simply read sentences aloud and attempted to commit the target words to memory, and there was little basis to expect greater semantic processing of target words in high-constraint sentences. Second, in the previous study, congruent high-and low-constraint sentences in which the target words made sense were mixed with incongruent sentences in which the target words did not make sense, whereas in the present study, only congruent sentences were included.

Three other methodological differences are important to mention, although they seem less likely to explain the discrepancy in findings. There was a marked difference in the level of constraint that defined the low-constraint conditions. The prior study reported the mean CLOZE percentage for low-constraint sentences as 38% (McFalls & Schwanenflugel, 2002), which is much higher than the rate of approximately 4% in Experiment 1. Previous studies that examined the effects of sentence constraint on lexical processing provide support for operationalizing low constraint at this lower level (e.g., Fischler & Bloom, 1979; Griffin & Bock, 1998; Van Hell & de Groot, 2008). Because there was no overlap in the ranges of cloze probabilities for the “low-constraint” item sets used in the two studies, the results are not necessarily contradictory. Second, the mean frequency of the target words in the previous study was comparable to the mean frequency of the low-frequency word set in the present study. Third, in the previous study. all items were studied in one block and recalled in a second block, whereas the present study divided items into 12 study-test cycles of 12 words each and did not mix constraint levels within cycles. Therefore, although Experiment 1 utilized a larger stimulus set, each study-test block was based on about half as many trials as in the previous study.

The two-way interaction of concreteness and context was not reliable as might be expected under the context availability account. However, a significant 3-way interaction of frequency, concreteness, and context revealed that for low-frequency words (but not high-frequency words) there was a significant interaction component for concreteness and context. Specifically, for low-frequency words, the concreteness effect was larger for isolated words than for words presented in high-constraint sentences, where the effect was eliminated. The reduction/elimination of the concreteness effect in the high-constraint condition, though limited to low-frequency words, is consistent with the context availability hypothesis. However, under the context availability hypothesis we would expect this effect to extend to high-frequency words and to the comparison between low-constraint and high-constraint sentence conditions. Therefore, taken together, the results of Experiment 1 do not support the context availability hypothesis of the concreteness effect.

High-frequency words were better recalled than low-frequency words, replicating the typical high-frequency recall advantage., but, the effect of word frequency did not interact with the context conditions. However, as mentioned above, the interaction of concreteness and context was present for low-frequency but not high-frequency words.

Experiment 2

Experiment 2 examined the same factors as Experiment 1 using the same set of word and sentence stimuli but with item recognition as the test task. The major goal of Experiment 2 was to examine whether the impact of context availability on the manifestation of the concreteness effect would extend to item recognition. It was hypothesized that, in accordance with the context availability account, the concreteness effect would be reduced or eliminated in the high-constraint condition relative to the isolated and low-constraint conditions. A second goal was to test for the first time how semantic constraint at encoding might impact recognition memory performance.

Method

Power and Sample Size.

We wanted to have 80% power to detect an effect of small-to-medium size (f = .175), which would require 67 participants. Because complete counterbalancing of the assignment of items to contexts and presentation orders required a multiple of 48 participants, 96 participants were recruited.

Participants.

Participants were 96 students (56 women, 40 men) from the University of Texas at El Paso with a median age of 22.5 (SD = 5.7). Participants were recruited from psychology courses and earned their choice of credit toward a course research requirement or $10. All were proficient speakers of English. Two additional volunteers who completed the protocol were excluded for failure to follow test instructions and replaced to preserve counterbalancing. Ten additional volunteers (not included in participant count above) were screened out for low English proficiency and did not complete the memory protocol.

Design.

The experimental conditions formed a 2 (concreteness) X 2 (frequency) X 4 (study context) within-subjects design. The concreteness factor divided stimuli into concrete and abstract words and the frequency factor divided stimuli into high- and low-frequency words. The context factor included words presented in isolation, words presented in low-constraint sentence frames, words presented in high-constraint sentence frames, and words not presented at study. Dependent variables included hit rates, false-alarm rates, and d’.

Materials.

Experiment 2 utilized the same stimulus set as Experiment 1. Thus, there were 36 high-frequency concrete words, 36 high-frequency abstract words, 36 low-frequency concrete words, and 36 low-frequency abstract words. The recognition task required withholding some items at study to serve as foils at test. Thus the 36 target words of each type were randomly assigned to 4 sets of 9 targets each. These sets were rotated through isolated, low-constraint, high-constraint, and not-studied (foil) conditions across participants using a Latin square. Experiment 2 also utilized the same demographic and language background questionnaire as in Experiment 1.

Apparatus.

Stimuli were presented on a Macintosh computer, and the experiment was programmed using PsyScope X software (Cohen et al., 1993). Responses were recorded using an ioLab Systems button box.

Procedure.

Participants were tested individually in sessions lasting approximately 45 minutes. After English proficiency screening, participants completed language and demographic background questionnaires. For the main experiment, participants sat in front of a computer and button box in a small experiment room. At study, participants completed 3 blocks of 36 trials for each context condition (isolated, low constraint, or high constraint). Within each context block, presentations were blocked by word type. Orders of blocks and target word types within blocks were counterbalanced across participants. Participants were asked to read each stimulus word or sentence aloud and try to commit the target words to memory. The target word was always the final word of the sentence and appeared in all capital letters. After reading each word or sentence, the participant pressed a button to advance to the next item, which would appear after 500 ms.

Immediately after the study trials were completed, instructions were given for the test trials. At test participants were administered a yes/no recognition task. The test trials consisted of two blocks of 72 trials in which all target words were presented, including those that were not presented at study. Of the 144 trials, 108 items were studied, and 36 items were new. Participants were instructed to press one of two buttons on the button box to indicate whether they recognized each word from the study phase or not.

Results

Discrimination.

Hit rates and false-alarm rates (given in Table 3) were used to calculate the detection parameter d’ in each condition for each participant using the equal variance model. Values of d’ (given in Figure 2) were then analyzed using a 2 (concreteness) x 2 (frequency) x 3 (study context) repeated measures ANOVA. A significant effect of frequency indicated that low-frequency words were more accurately recognized than high-frequency words, F(1, 95) = 22.82, MSE = .858, p < .001, partial η² = .19. A significant effect of concreteness revealed that concrete were better discriminated than abstract words, F(1, 95) = 4.01, MSE = .850, p = .048, partial η² = .04. The effects of frequency and concreteness did not interact, F < 1, partial η² < .01.

Table 3.

Mean (SE) Recognition Performance in Experiment 2

Concreteness
Frequency
Study Context	Hit Rate	False Alarm Rate
Concrete
High frequency
Isolated	.626 (.02)	--
Low constraint	.681 (.02)	--
High constraint	.601 (.02)	--
Not presented	--	.218 (.02)
Low frequency
Isolated	.623 (.02)	--
Low constraint	.691 (.02)	--
High constraint	.683 (.02)	--
Not presented	--	.166 (.01)
Abstract
High Frequency
Isolated	.652 (.02)	--
Low constraint	.666 (.02)	--
High constraint	.636 (.02)	--
Not presented	--	.253 (.02)
Low frequency
Isolated	.674 (.02)	--
Low constraint	.721 (.02)	--
High constraint	.705 (.02)	--
Not presented	--	.230 (.02)

Figure 2.

Recognition Scores (d’) in Experiment 2 as a Function of Concreteness, Word Frequency, and Encoding Context. Error bars represent standard errors of the mean.

There was a main effect of context, F(2, 190) = 5.12, MSE = .419, p = .007, partial η² = .05. Planned comparisons (simple contrasts on the main effect) showed that targets in low constraint sentences were better recognized than targets in isolation, F(1, 95) = 8.22, MSE = .231, p = .005, partial η² = .08, and targets in high constraint sentences, F(1, 95) = 8.58, MSE = .146, p = .004, partial η² = .08. Although the full interaction of context and frequency did not reach significance, F(2, 190) = 2.86, MSE = .301, p = .06, partial η² = .03, simple interactions revealed that the low-frequency advantage for high-constraint sentences was larger than for words studied in isolation, F(1, 95) = 4.53, MSE = .759, p = .036 but not larger than for words studied in low-constraint sentences, F(1, 95) = 1.86, MSE = .512, p = .175. Context did not interact with concreteness, F < 1, partial η² = .01, and the three-way interaction was not significant, F(2,190) = 1.04, MSE = .232, p = .355, partial η² = .01.

Discussion

Detection (d’) scores were higher for concrete words than for abstract words, producing the expected concreteness effect. Words embedded in low-constraint sentences elicited higher detection scores than words embedded in high constraint sentences and words presented in isolation. This pattern of performance across context conditions diverged from the results of the only prior studies to examine the effects of context or semantic constraint on recognition, where recognition accuracy decreased when words were presented in sentence contexts (Coane & Balota, 2010; Schwartz, 1975). Although the level of semantic constraint in the Schwartz (1975) study is unknown, inspection of the stimuli for the Coane and Balota (2010) study suggests that the sentences had relatively low constraint. Although there were main effects of concreteness and context, the two factors did not interact as would be expected under the context availability account. Thus, the results of Experiment 2 do not support the context availability hypothesis of the concreteness effect in recognition memory.

Low-frequency words exhibited higher detection scores than high-frequency words, with both higher hit rates and lower false alarm rates, replicating the typical low-frequency advantage in recognition. The low-frequency recognition advantage was greater for words studied in high-constraint sentence contexts than for words studied in isolation. This effect parallels results of a previous study in which the low-frequency advantage was more pronounced when words were presented in a paragraph context than when presented in isolation (Coane & Balota, 2010).

General Discussion

The present study examined the effects of sentence context and semantic constraint on explicit verbal memory performance, with the goal of understanding how context availability might contribute to the well-established effects of concreteness. In the following sections, we elaborate on the nature of sentence context and semantic constraint effects on memory performance and implications of the results for theories of the concreteness effect.

Context and Semantic Constraint in Explicit Verbal Memory Performance

The effects of context were significant in both experiments. In free recall, words studied in isolation were recalled at a higher rate than words presented in low-constraint sentences, and words presented in low-constraint sentences were better recalled than words presented in high-constraint sentences. The negative impact of sentence context on recall is consistent with previous results (Cofer, 1968; Wood, 1970). The explanation given was that embedding words in sentence contexts at study disrupted the formation of new inter-item associations, thereby negatively affecting recall.

The finding of superior recall for target words in low-constraint sentences over high-constraint sentences stands in contrast to the results of a previous study that reported the opposite pattern (McFalls & Schwanenflugel, 2002). However, as explained in the Discussion for Experiment 1, the use of an encoding task that draws attention to the different types of context may have been responsible for the high-constraint advantage in the previous study. In the present study, the reasons for a low-constraint advantage may be twofold. First, high semantic constraint facilitates identification and comprehension of the target word (e.g., Balota et al., 1985; Forster, 1981; Gollan et al., 2011; Griffin & Bock, 1998; Rayner et al., 2004), and this easier processing may reduce attention to the target word or eliminate the desirable difficulties (Bjork, 1994) that would be present and enhance memory in low-constraint conditions. Second, extending the reasoning of two early studies (Cofer, 1968; Wood, 1970), high-constraint sentences ought to elicit even more within-sentence processing than low-constraint sentences due to the stronger semantic connection between the sentence frame and the target word. This increase in within-sentence processing for high-constraint sentences would further reduce between-sentence, or inter-item processing relative to low-constraint sentences and have a stronger negative impact on recall performance.

In recognition, words studied in low-constraint sentences were better discriminated than words studied in isolation, contrary to previous findings (Coane & Balota, 2010; Schwartz, 1975). Experiment 2 showed for the first time that words studied in low-constraint sentences were better recognized than words studied in high-constraint sentences. Given that rehearsal and inter-item processing are not as important for recognition memory as for free recall, any differences in inter-item processing would be unlikely to produce a recognition performance difference. Therefore, a more plausible explanation is that the ease of target word processing afforded by the high-constraint sentence frames eliminates desirable difficulties relative to the low-constraint sentence conditions.

To get a better idea of the processes affected by sentence context and constraint manipulations, we examined interactions of these manipulations with word frequency. As in previous research, high-frequency words were better recalled and low-frequency words were better recognized (e.g., cites). We examined whether words embedded in sentence contexts and words presented in isolation would exhibit different frequency effects, as in one previous study (Coane & Balota, 2010). In Experiment 2, the low-frequency advantage for recognition was greater for targets studied in high-constraint sentences than for targets studied in isolation, consistent with the larger frequency effects in recognition of contextualized relative to isolated words in the previous study. In Experiment 1, no such interactions were observed in recall performance.

We tested whether the interactions of word frequency and semantic constraint seen in lexical processing studies would be paralleled in explicit memory. In contrast to those studies, word frequency effects in recall and recognition did not differ for high- and low-constraint sentence conditions. The lack of interaction in recall suggests that the production processes used at test are not impacted by the constraint manipulation. Instead it would seem that for both tasks, semantic constraint only affected the comprehension processes of encoding, where no interaction would be expected.

Implications for Theories of Concreteness Effects in Explicit Memory

Concrete words were better recalled and recognized than abstract words as in previous research. According to the context availability framework, the concrete advantage arises due to differential availability of associated context information for concrete and abstract words. Specifically, it is assumed that it is easier to access associated context information for concrete words than for abstract words, thereby leading to benefits in lexical processing and memory. Based on this logic, we expected a reduction in the advantage for concrete words when they were embedded in high-constraint sentence contexts at study. In Experiment 1, although the concreteness effect for low-frequency words was smaller in the high-constraint condition than in the isolated condition, this pattern was not observed for high-frequency words, and there was no difference in the magnitude of the concreteness effect for low- and high-constraint sentence conditions. In Experiment 2, there was no indication of an interaction of concreteness and context condition. Thus the present results do not support the context availability framework.

The present results indicate that context availability is not an important factor underlying concreteness effects in explicit memory. As explained in the introduction, the results of prior studies examining whether context availability moderates concreteness effects in free recall have been mixed (Marschark, 1985; Wattenmaker & Shoben, 1987; Schwanenflugel et al., 1992). Although the sentence context manipulation utilized in the current experiment had not been applied in previous studies of concreteness effects on memory, this manipulation did reduce concreteness effects in lexical processing tasks (van Hell & de Groot, 2008), indicating that it is a sufficiently powerful context manipulation to reduce or eliminate concreteness effects.

The two approaches to manipulations of context availability in previous research may tap into different phenomena. In some studies, context availability was treated as a long-standing property of target words based on their pre-experimental history of episodic contexts (Schwanenflugel et al., 1988, 1992; van Hell & de Groot, 1998). Context availability was manipulated by comparing word sets with higher and lower subjective ratings of context availability. In contrast, in the present study and some other studies (Schwanenflugel & Shoben, 1983; Van Hell & de Groot, 2008), the approach was to provide contexts at encoding in an effort to temporarily boost pre-experimental levels of context availability, with the expectation that providing a context at encoding would have a bigger impact on words with lower pre-experimental context availability. It is as yet unknown whether common mechanisms underlie the effects of pre-experimental context availability and the effects of experimental manipulations of context availability.

Although support for the context availability explanation of concreteness effects in lexical processing is compelling, support for the context availability explanation of concreteness effects in explicit memory is minimal. It therefore appears that concreteness effects have different bases in lexical processing and in explicit memory. Based on the results of previous lexical processing studies, the high-constraint sentence frames in the present study should have equalized the ease of processing of the concrete and abstract target words, but that did not equalize memory performance for concrete and abstract words. Concrete and abstract words must differ on some characteristics other than context availability that are important for memory but not for lexical processing. One possibility is that these are characteristics that do not have anything to do with context. For example, some candidate differences might include more precise meanings or more distinctive orthography for concrete words (as suggested by Hirshman & Arndt, 1997) or more possible retrieval routes for concrete words because of imagery (as in the dual coding theory, Paivio, 1991).

Another possibility is that a different contextual variable is responsible for concreteness effects in explicit memory. Context variability is associated with explicit memory performance such that words with low context variability are better recalled and recognized (e.g., Steyvers & Malmberg, 2003). Context variability, which has been implicated in explaining word-frequency effects in both lexical processing (Adelman et al., 2006) and explicit memory (e.g., Buchler & Reder, 2007; Marsh, Meeks, Hicks, Cook, & Clark-Foos, 2006), is another candidate explanation for concreteness effects in explicit memory. Although one previous study showed no interaction of context variability and word concreteness in free recall (Marsh et al., 2006), in that case, context variability was manipulated by changing the external environment in which targets were studied and tested rather than changing the semantic context. That environmental approach to episodic context is quite different from the semantic approach to episodic context addressed by the current study and by Criss et al. (2011). Therefore, it may be fruitful to explore further whether context variability helps to explain concreteness effects.

Conclusions

The effects of context, word concreteness, and word frequency were examined in free recall and recognition memory tasks. Although there were significant effects of concreteness and context in both memory tasks, the magnitude of the concreteness effects in recall and recognition did not differ for words studied in high- and low-constraint sentence contexts. These results therefore do not support extending the context availability hypothesis beyond lexical processing to explain concreteness effects in explicit memory performance. Semantic constraint hurt both recall and recognition performance, which suggests that the facilitating effect of semantic constraint on processing the target word reduces either attention to the target word or the desirable difficulties that would be present in a low-constraint context.