bs_bs_banner Australian Journal of Psychology 2012; 64: 199–208 doi:10.1111/j.1742-9536.2012.00053.x Novelty and processing demands in conceptual combination ajpy_53 199..208 Brentyn J. Ramm1 and Graeme S. Halford2 1 School of Psychology, University of Queensland, St. Lucia, and 2School of Psychology, Griffith University, Mt Gravatt, Queensland, Australia Abstract In this article, we sought to isolate the processing demands of combining the concepts of modifier-noun phrases from those of other language comprehension processes. Probe reaction time (RT) was used as an indication of the processing resources required for combining concepts. Phrase frequency (as measured by Google hit rates) was used as a metric of the degree of conceptual combination required for each phrase. Participants were asked to interpret modifier-noun phrases using a sense-nonsense decision (Experiment 1) and a phrase meaning access task (Experiment 2). Experiment 2 also used a lexical decision task to activate the word’s individual meanings. Regression analyses for both experiments indicated that phrase frequency (indicating novelty) predicts a significant portion of the probe RT variance, such that low-frequency phrases required more processing resources than high-frequency phrases, when controlling for associative strength, word frequency, letter length, and lexical-semantic activation. Overall, this study indicates that conceptual combination requires processing resources beyond those of other language processes. Key words: conceptual combination, familiarity, probe reaction time, processing demands, working memory Combined concepts, such as road rage, carbon footprint, and beer garden are a common and everyday occurrence in language. Humans are also adept at interpreting novel combinations such as pet human. Creating new meanings such as these, on-the-fly, would seem to rely upon working memory resources; however, it is not clear whether such demands stem from the integrative processes themselves or merely from the concurrent activation of the concepts in working memory, syntactic processing, etc. The aim of this study was to investigate this question by isolating the processing demands of conceptual combination from those of other language comprehension processes. One advantage of using two word phrases, in contrast with sentences (Daneman & Carpenter, 1983; Just & Carpenter, 1992; MacDonald, Just, & Carpenter, 1992; Miyake, Just, & Carpenter, 1994), is that the processing costs of lexical-semantic activation of the individual words, syntactic processing, and sentence meaning updating can be held fixed, while the amount of conceptual combination required for interpretation varies. In particular, we used phrase frequency as a measure of the amount of integration required for a phrase. It was assumed Correspondence: Brentyn J. Ramm, PhD, School of Philosophy, HC Coombs Bldg, The Australian National University, Canberra, ACT 0200, Australia. Email: brentyn.ramm@anu.edu.au Received 8 November 2011. Accepted for publication 26 March 2012. © 2012 The Australian Psychological Society that interpreting low-frequency phrases (e.g., moth signals) would involve more integrative processing than familiar phrases (e.g., mobile phone), the latter of which are more likely to involve the retrieval of a pre-stored meaning than conceptual combination. Modifier-noun compounds such as zebra crossing can be broadly divided into property interpretations (e.g., a striped crossing) and relational interpretations (e.g., a crossing for zebras) (Wisniewski, 1996). In this study, we focused on modifier-noun phrases that tend to elicit relational interpretations (Gagne & Shoben, 1997). Wisniewski (1997) proposes that creating relation-based interpretations in conceptual combination involves the construction of a scenario based on background knowledge. For example, truck soap can be interpreted as soap used for washing trucks because a truck can be washed, while plant soap is more likely to be interpreted as soap made of plants because a plant does not easily fit into the role of an object that can be washed. Background knowledge is also important in the selection of relations and the elaboration of the combined concept (Medin & Shoben, 1988; Murphy, 1988; Murphy & Medin, 1985). Thus, truck soap could be inferred to be rougher and likely to dissolve grease, while plant soap may be inferred to be more natural, sweetsmelling (e.g., rose, lemongrass etc), and softer on the skin (e.g., it may contain Aloe Vera). Furthermore, access to background knowledge may also be constrained by the historical frequency with which a relation has been activated by the modifier-noun (as proposed by the Competition 200 B.J. Ramm and G.S. Halford Amongst Relations in Nominals theory: Gagne 2001; Gagne and Shoben 1997). For example, LOCATION is a highfrequency relation for mountain (e.g., mountain stream) and may be more likely to be activated (more available) than a low-frequency relation for mountain such as ABOUT (e.g., mountain magazine). While conceptual combination is an essential process in human cognition, there do not appear to be any previous studies that have investigated the role that working memory resources play in combining concepts. It is important to investigate this so that conceptual combination can be related to broader cognitive systems. While all theories hold that conceptual combination involves additional processes to those of semantic and syntactic processing (e.g., Gagne & Shoben, 1997; Medin & Shoben, 1988; Murphy & Medin, 1985; Wisniewski, 1996, 1997), not all cognitive processes impose working memory demands. Working memory has been traditionally divided into short-term memory components (e.g., phonological, visual-spatial, etc) that are non-capacity-limited (but open to decay and/or interference) and a central limited-capacity component (e.g., Baddeley, 1986; Baddeley & Hitch, 1974; Cowan, 2001; Oberauer, 2009; although for a multiple capacities view, see Martin, Shelton, & Yaffee, 1994; Wickens, 1984). Whether conceptual combination imposes processing demands seems to depend crucially upon whether relation activation and the subsequent linking of concepts occurs in parallel with lexical-semantic and syntactic processing, or also draws upon the capacity-limited component. The Semantic Cognition model (Rogers & McClelland, 2004) is an example of a neural network where relations are represented subsymbolically and are processed in parallel. Although it is not a model of conceptual combination, it is at least suggestive of how conceptual combination may occur in parallel with other language comprehension processes. As current models of conceptual combination do not make predictions either way, it is important to investigate this question empirically. In this study, we introduced a secondary task (probe reaction time (RT)) concurrently with standard conceptual combination tasks (dual-task design) to measure the processing resources involved in combining concepts. We expected that the higher the novelty of a phrase, the more processing resources (indicated by longer probe RTs) would be required to combine their concepts, thus providing evidence that conceptual combination is resource-demanding. We also used multiple regressions to statistically control for other factors such as associative strength, word length, and word frequency. EXPERIMENT 1 The present experiment used a sense-nonsense decision to measure the processing time of combining the concepts of novel phrases (e.g., Gagne, 2001; Gagne & Shoben, 1997). In the experiment, an auditory tone (probe) sounded randomly during the sense-nonsense decision. Participant’s latency in a vocal response to the probe (probe RT) was used as an indication of the processing demands of each phrase (e.g., Kahneman, 1973; Kerr, 1973; Lansman & Hunt, 1982; Maybery, Bain, & Halford, 1986; Posner & Boies, 1971; Posner & Klein, 1973). These studies have shown that latencies in probe RT do not necessarily mirror those of a primary task. For example, it has been found that processing negative premises takes significantly longer than positive premises during transitive inference (Maybery et al., 1986), and processing four letters takes significantly longer than processing one letter (Shwartz, 1976). Despite this difference in response times for these primary tasks, no difference was found in probe RT. On the other hand, Maybery et al. (1986) used latencies in probe RT to show that premise integration during transitive inference imposes more informationprocessing demands than holding both of the premises in working memory. The assumption behind the task is that as working memory has limited resources, latency in responding to the beep is indicative of the amount of resources available at that point in the primary task. That is, the greater the demands of the primary task, the fewer resources will be left over for responding to the secondary task, and so probe RT will increase (Posner & Boies, 1971). The auditory probes appeared after the visual onset of phrase during the process of conceptual combination. Based upon previous findings of average gaze durations for reading single words of 239 ms (Just & Carpenter, 1980) and 244–324 ms (Miyake et al., 1994), two probe onset times of 400 and 500 ms were used. It was expected that this would place probes around the beginning of the combination process. Two probe onset times were used so as to reduce probe predictability. As a measure of phrase frequency, we entered each phrase into Google in quotes and recorded the hit rate returned (for a previous study using Google hits as indication of familiarity, see Wisniewski & Murphy, 2005). It was expected that high-frequency phrases would require less conceptual combination than low-frequency phrases, and thus fewer processing resources would be drawn upon for the former than for the latter phrases. While the phrases used in this study were all considered to be novel by Gagne and Shoben (1997), we felt that some were clearly more familiar than others (for example, financial headache and flu pills vs moth signals and sap remedy). One advantage of this measure was that for all phrases, both words are processed together, and thus joint lexical-semantic activation and syntactic processing were held constant while the amount of conceptual combination varied. However, a limitation to using phrase frequency as a measure of the amount of conceptual combination is that it may also overlap with factors such as: (1) associative links © 2012 The Australian Psychological Society Processing demands and concept combination between the two words; (2) the frequency of the two words; and (3) letter length. Later, we use regression analyses to assess the unique contribution of phrase frequency to processing demands (probe RT). Method Participants Sixty-seven first-year psychology students participated for course credit. Design The experiment consisted of four blocks of trials: (1) Probe Only (28 trials); (2) Sense-Nonsense Practice (20 trials); (3) Sense-Nonsense Test (160 trials); and (4) Probe-Only (28 trials). Procedure The experiment was computer-based, and participants were tested in groups of up to four. Instructions were displayed on the screen preceding each block. The trial sequence was blank screen (1,000 ms), fixation cross (1,000 ms), blank screen (1,000 ms), and phrase (3,000 ms). Each phrase appeared in the centre of the screen, and participants were asked to decide if it was sensible or nonsense (p and q keys respectively) as quickly as possible. Participants completed two blocks of sense-nonsense decisions, and the 56 test phrases (obtained from Gagne & Shoben, 1997) were randomly distributed between these blocks. Each sensenonsense block consisted of 160 trials, with probes occurring randomly on 56 of the trials, that is, 28 sensible and 28 nonsense phrases (see Appendix I). Thus, probes occurred on 35% of occasions. On a probe trial, a 100 ms beep (probe) sounded through headphones 400 or 500 ms after the appearance of the word-pair stimuli. Participants were required to respond by saying ‘beep’ into a microphone as quickly as possible when the tone sounded. Probe RT was recorded using a verbal button box that measured the latency of the vocal response from the onset of the beep. The sensitivity of the verbal button box was adjusted to suit each participant’s voice. The Probe Only blocks consisted of the same procedure as the earlier conditions, except that a neutral symbol ‘********’ appeared in place of the word pairs, and participants were not required to respond to the symbol. This condition, in which no words are processed, was used as baseline for the primary task probes. Measures of phrase characteristics Phrase frequency. Phrase frequency was used as a measure of the amount of integrative processes that are required for © 2012 The Australian Psychological Society 201 each phrase. To reduce the large variance obtained in the hit rates, we transformed these scores into logs of ten. Associative strength. We obtained the associative strength between each phrase’s words from Nelson, McEvoy, and Schreiber (1998) (we only used forward association, i.e., from modifier to header). According to these free association norms, none of the phrases were composed of associated words, and thus, this variable was not considered in the following analyses. Word frequency. As a measure of word frequency, we obtained Google hit rates for each individual word. The hit rates of the modifier and header noun were summed for each phrase, and we again transformed these scores into logs of ten. Letter length. The length of the word is another factor that may affect processing demands of interpreting a phrase. We obtained this by simply summing the number of letters in each phrase. Results Thirteen participants were removed for scoring 50% or less on overall accuracy for sensible phrases. The remaining sample size was 54 participants. The average accuracy for the sense-nonsense decision for the remaining participants was 70.3%. Because of experimenter error, the data for the probe-only (baseline) blocks were not recorded. The data analyses were performed only on correct trials and only on trials in which a probe occurred. Preliminary analyses showed that the distribution of scores was significantly positively skewed for both probe RT (z = 4.0, p = .001) and sense-nonsense RT distributions (z = 3.11, p = .005). To correct for the skewness, we used the median probe RT and sense-nonsense RT score for each phrase (for previous use of median scores for Probe RT, see Maybery et al., 1986). As there was no main effect of probe onset time (400 and 500 ms), this variable was collapsed for the following analyses. The correlations of the predictor variables and dependent variables for Experiment 1 are displayed in Table 1. The correlations and analyses were conducted across items (i.e., the phrases) rather than participants. Probe RT A multiple hierarchical regression was used to examine whether phrase frequency predicted probe RT after controlling for word frequency and letter length. The variables word frequency and letter length were entered in the first step, and phrase frequency was entered in the second step. In the first step, the combined variables explained a significant 202 B.J. Ramm and G.S. Halford Table 1 Experiment 1 Pearson’s r correlations of dependent variables (probe reaction time (RT), sense-nonsense RT, and accuracy) with predictor variables Probe RT Response time Accuracy Phrase frequency Word frequency Letter length -0.57 p < .001 -0.45 p = .001 0.27 p = .043 -0.47 p < .001 -0.31 p = .022 0.13 p = .431 0.37 p = .005 0.42 p = .001 0.013 p = .925 portion of the probe RT variance (r2 = 0.31, F(2,53) = 11.75, p < .001). At the second step, phrase frequency made a significant additional contribution (b = -0.40, r2 change = 0.07, F(1, 52) = 6.11, p = .017). The only other significant individual contributor to the final model was letter length (b = 0.23, t(52) = 2.04, p = .047). Overall, then, it was found that the less frequently a modifier-noun phrase appears on the Internet, the more processing resources that are required to integrate its constituent meanings, and this is the case when controlling for the frequency of the phrase’s individual words and letter length. Sense-nonsense RT Similar analyses were used to examine the relationship between phrase frequency and sense-nonsense RT, controlling for letter length and word frequency. Letter length and word frequency were added at the first step and combined to explain a significant portion of the sense-nonsense RT variance (r2 = 0.24, F(2,53) = 8.28, p = .001). At the second step, phrase frequency made a significant additional contribution (b = -0.36, r2 change = 0.06, F(1, 52) = 7.34, p = .041). In this final model, letter length also made a significant unique contribution to sense-nonsense RT (b = 0.32, t(52) = 2.65, p = .011). Thus, it was found that phrase frequency made a unique contribution to sense-nonsense RT when controlling for letter length and word frequency. Sense-nonsense accuracy When sense-nonsense accuracy was used as the criterion, letter length and word frequency added at the first step did not explain a significant portion of the accuracy variance (r2 = 0.02, F(2,53) = 0.48, p = .619). At the second step, however, phrase frequency made a significant unique contribution to predicting accuracy scores (b = 0.40, r2 change = 0.07, F(1,52) = 4.25, p = .044). EXPERIMENT 2 The results of Experiment 1 indicated that the less frequently a phrase appears on the Internet, the more processing resources are required to interpret it. However, as it is subjective and ambiguous as to whether a phrase is sensible or not, this task included considerable decision uncertainty that may have imposed additional processing loads to that of conceptual combination. The present experiment assessed the resource demands of processing modifier-noun phrases using a phrase meaning access task in which participants pushed a button when they had generated an interpretation for the phrase. This eliminates the uncertainty of the decision. In the previous experiment, phrase frequency was used as a measure of amount of conceptual combination; particularly, it was expected that more familiar phrases such as financial headache would require less combinatory processes than more novel phrases such as mountain magazine. In the present experiment, we also employed a lexical decision task to gain an independent measure of the lexical-semantic activation of the individual words (i.e., when processed one at a time). The mean sense-nonsense decision time from Experiment 1 was 1,201 ms, suggesting that conceptual combination may be a relatively late-occurring process. To sample processing demands from a slightly later period in the processing of the phrases, we changed the probe onset times from 400 and 500 ms (Experiment 1) to 500 and 600 ms (Experiment 2). It should be noted that probe RT took an average of 657 ms, thus a probe onset of 600 ms should approximately sample processing demands ranging from 600 to 1,257 ms. We also added highly familiar phrases (e.g., stop sign) to increase the range of familiarity to novelty for the stimuli. It was expected that if the combination of concepts is capacitydemanding, then the frequency of the phrase would show a negative correlation with phrase probe RT even when controlling for lexical-semantic access, associative links between the two words, letter length, and the frequency of the phrase’s individual words. Method Participants Data were collected from 61 first-year psychology students who participated for course credit. © 2012 The Australian Psychological Society Processing demands and concept combination Design The experiment consisted of six blocks of trials: (1) Probe Only (28 trials); (2) Lexical Access Practice (20 trials); (3) Lexical Decision Test (112 trials); (4) Phrase Meaning Access Practice (20 trials); (5) Phrase Meaning Access Test (112 trials); and (6) Probe-Only (28 trials). The order of the Lexical Access and Phrase Meaning Access blocks was counterbalanced between subjects. Materials The novel test phrases were the same 56 modifier-noun phrases from Gagne and Shoben (1997) (e.g., tax pressure) used in Experiment 1. We also selected 56 familiar modifiernoun phrases (e.g., mobile phone) from the Oxford English Dictionary. These phrases were randomly distributed between the Phrase Meaning Access and Lexical Access test blocks for each participant, such that each block had 28 novel phrases and 28 familiar phrases. There were also 14 novel and 14 familiar filler phrases, and 28 word–non-word phrases in each block. The word–non-word pairs (e.g., thunk kitten) were constructed by changing the vowel of existing words and by the pairing of random syllables. Non-words appeared equally often in the modifier and head-noun positions (see Appendix II). Procedure Except for the following details, the procedure was the same as Experiment 1. For the Phrase Meaning Access block, participants were asked to read the two words and to push the P key on the keyboard (or the Q key for left-handers) as quickly as possible when they had accessed a meaning for the phrase as a whole. If one of the strings was a non-word, then they were asked to refrain from pushing the button. For the Lexical Decision block, participants were asked to push a button if both of the strings were words and again to refrain from responding if one of the strings was a non-word. The probe sounded randomly on 28 of the trials of each test block (25% of occasions), half of which were novel and half of which were familiar modifier-noun phrases. Measures of phrase characteristics Phrase characteristics were measured using the same method as Experiment 1. Phrase frequency. As expected novel phrases appeared with significantly lower frequency on the Internet (M = 3.97, standard deviation (SD) = 1.17) than familiar phrases (M = 6.57, SD = 0.69), t(110) = 14.29, p < .001. Associative strength. According to Nelson et al.’s (1998) free association norms, many familiar phrases such as beach © 2012 The Australian Psychological Society 203 sand were composed of associated words (e.g., there was a 39% probability that participants would produce sand given beach), while none of the words of the novel phrases such as chocolate plant were associated. A between-groups t-test revealed that familiar phrases (M = 0.057, SD = 0.09) were significantly more associated than novel phrases (M = 0.0), t(110) = 4.71, p < .001. Word frequency. A between-groups t-test revealed that novel phrases were composed of significantly lower frequency individual words (M = 16.42, SD = 0.95) than familiar phrases (M = 16.96, SD = 0.79), t(110) = 3.24, p = .002. Lexical-semantic access. We used probe RT for a lexical decision to the phrases as a means of controlling for the processing demands of lexical access to the two individual words, thereby isolating the demands of conceptual combination from lexical-semantic activation. Results Two participants were removed from the data set for not following the instructions. The only accuracy score for this experiment was the requirement that participants refrain from making a response when one of the words of the phrase was a non-word (that is the lexical decision task for both novel and familiar phrases). A further two participants were removed for scoring outside two SDs in accuracy for the lexical decision task. For the remaining 57 participants, the overall average accuracy rate on word–non-word trials was 87.83%. To reduce skewness in probe RT and phrase meaning access scores, we again used the median scores for each phrase. Preliminary analyses showed that probe RT was significantly faster for the probe-only task (baseline) (M = 475 ms) than for the experimental conditions (M = 565 ms), showing that probe RT was effectively indicating processing load differences, t(61) = 6.08, p < .001. As there was no main effect or interactions with probe onset time (500 and 600 ms), this variable was collapsed. The correlations of the predictor variables and dependent variables for the experiment are displayed in Table 2. Pearson’s r correlations revealed that phrase frequency, associative strength, and word frequency were all negatively correlated with phrase meaning access probe RT. It was also found that lexical probe RT to the phrases and letter length were positively correlated with phrase meaning access probe RT. The purpose of the following two multiple regressions was to investigate the unique contribution of phrase frequency to the variance of phrase probe RT and phrase meaning access response time. 204 B.J. Ramm and G.S. Halford Table 2 Experiment 2 Pearson’s r correlations of dependent variables (phrase probe reaction time (RT) and phrase meaning access response time) with predictor variables Phrase probe RT Response time Phrase frequency Lexical probe RT -0.61 p < .001 -0.75 p < .001 0.36 p < .001 Lexical response time Word frequency Association strength Letter length 0.76 p < .001 -0.43 p < .001 -0.48 p < .001 -0.21 p = .025 -0.25 p = .009 0.31 p = .001 0.36 p < .001 Note. Cells are left empty for variables that were not used as predictors in the multiple regression analyses. Phrase probe RT In the first regression using phrase probe RT as the criterion, the variables lexical probe RT, associative strength, word frequency, and letter length were entered in the first step, and phrase frequency was entered in the second step. In the first step, the combined variables explained a significant portion of the phrase probe RT variance (r2 = 0.28, F(4,107) = 10.49, p < .001). In the second step, phrase frequency made a significant additional contribution (b = -0.54, r2 change = 0.13, F(1,106) = 24.14, p < .001. In this final model, letter length (b = -0.20, t(106) = 2.52, p = .013) also made a significant unique contribution to predicting phrase probe RT. Overall, then, it was again found that the more novel a modifier-noun phrase, the more processing resources that are required to combine its concepts, and this is the case when controlling for lexical-semantic access, the associative strength between the phrase’s words, the frequency of the phrase’s individual words, and letter length. Phrase meaning access RT In the second regression, phrase meaning access RT was used as the criterion. The variables letter length, associative strength, word frequency, and lexical response time were entered in the first step, and phrase frequency was entered in the second step. In the first step, the combined variables explained a significant portion of the response time variance (r2 = 0.61, F(4,107) = 41.84, p < .001). At the second step, phrase frequency made a significant additional contribution (b = -0.44, r2 change = 0.07, F(1,106) = 44.51, p < .001). In this final model, a significant unique contribution to response time was also made by lexical response time (b = 0.41, t(106) = 4.71, p < .001). Thus, for phrase meaning access RT, it was found that phrase frequency again made the largest contribution, although a unique contribution was also made by lexical response time. GENERAL DISCUSSION Even though conceptual combination plays a crucial role in cognition, to our knowledge, no previous studies have directly investigated the resource demands involved in this process. In the present study, we sought to isolate the processing capacity-demands of combining the concepts of modifier-noun phrases from that of other processes such as lexical-semantic and syntactic processing. In both experiments, phrase frequency (as measured by Google hit rates) was used as a metric of the amount of conceptual combination required for each phrase. We also controlled for associative strength, word frequency, and letter length. In Experiment 2, a lexical decision task was also used as an independent measure of the lexical-semantic activation of the two words. Both experiments indicated that conceptual combination imposes more processing loads as the frequency of the phrase decreases, when controlling for these factors. Overall, our findings clarify how conceptual combination occurs within working memory in that they provide converging evidence that the more combinatory processes that are required to interpret a phrase, the more processing resources that will be involved in constructing the representation. One likely source for the processing demands of conceptual combination beyond that of the lexical-semantic processing of the individual words is that conceptual integration requires the simultaneous processing of both concepts and their linking via a relation. One way in which this process could be modelled is as the binding of concepts to a relational symbol (e.g., Goodwin & Johnson-Laird, 2005; Halford, Wilson, & Philips, 1998). For instance, interpreting mountain magazine may involve the selection of a relational symbol in which mountain and magazine play appropriate roles (?(mountain, magazine)). This would be an example of a binary relation, which have been found to impose more processing demands than unary relations such as the binding of an attribute to a single concept (e.g., ripe(apple)) (Halford et al., 1998). The hypothesis that relational processing contributes to the processing demands of conceptual combination is also consistent with findings that the binding/relating of representations is capacity-limited for a wide variety of cognitive functions such as reasoning and general intelligence (e.g., Halford, Wilson, & Philips, 2010; Oberauer, 2009; Oberauer, Süß, Wilhelm, & Wittmann, 2008). However, as we did not isolate relational processing from non-relational processes, we acknowledge that our findings © 2012 The Australian Psychological Society Processing demands and concept combination only support the general conclusion that combining concepts does not take place in parallel with other language comprehension processes. Other non-relational processes may also contribute to the processing costs of combining concepts. For instance, it is likely that low-frequency phrases (e.g., moth signals) require more background knowledge in constructing a scenario (e.g., moth signals may be a flight pattern used in a mating ritual) and elaborating the properties of the new concept (e.g., such moths may be more likely to be social insects than ordinary moths). An alternative interpretation of the present results is in terms of a central attentional bottleneck, where certain central processes must proceed one at a time rather than by the sharing of limited processing resources (e.g., Pashler, 1994). In fact, both processing resources and bottlenecks play an important role in a number of prominent models of working memory (see Oberauer & Göthe, 2006). However, given the large amount of evidence that many sentence level processes draw upon a limited working memory capacity (e.g., Daneman & Carpenter, 1980, 1983; Just & Carpenter, 1992; Miyake et al., 1994), we felt that a processing resources interpretation (or both together) was the most likely explanation for our results. Future research directly investigating whether an individual’s general working memory capacity affects their ability to interpret novel modifier-noun phrases would help to clarify this issue. Further studies are also needed to isolate the processing demands of individual subprocesses in conceptual combination, such as relation competition and availability (Gagne & Shoben, 1997), scenario construction (Wisniewski, 1997), and accessing background knowledge and elaboration of the representation (Medin & Shoben, 1988; Murphy, 1988; Murphy & Medin, 1985). In conclusion, the present study suggests that working memory resources should be considered an essential factor in any complete model of conceptual combination and suggests some potentially fruitful lines of research. REFERENCES Baddeley, A. D. (1986). Working memory. New York: Oxford University Press. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47–89). New York: Academic Press. Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87–185. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466. Daneman, M., & Carpenter, P. A. (1983). Individual differences in integrating information between and within sentences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9(4), 561–584. © 2012 The Australian Psychological Society 205 Gagne, C. L. (2001). Relation and lexical priming during the interpretation of modifier-noun combinations. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27(1), 236–254. Gagne, C. L., & Shoben, E. J. (1997). Influence of thematic relations on the comprehension of modifier-noun combinations. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23(1), 71–87. Goodwin, G. P., & Johnson-Laird, P. N. (2005). Reasoning about relations. Psychological Review, 112, 468–493. Halford, G. S., Wilson, W. H., & Philips, S. (1998). Processing capacity defined by relational complexity: Implications for comparative, developmental, and cognitive psychology. Behavioral and Brain Sciences, 21(6), 803–864. Halford, G. S., Wilson, W. H., & Philips, S. (2010). Relational knowledge: The foundation of higher cognition. Trends in Cognitive Sciences, 14(11), 473–517. Just, M. A., & Carpenter, P. A. (1980). A theory of reading: From eye fixations to comprehension. Psychological Review, 87(4), 329– 354. Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99(1), 122–149. Kahneman, D. (1973). Attention and effort. Englewood Cliffs, NJ: Prentice-Hall. Kerr, B. (1973). Processing demands during mental operations. Memory & Cognition, 1, 401–412. Lansman, M., & Hunt, E. (1982). Individual differences in secondary task performance. Memory & Cognition, 10(1), 10–24. MacDonald, M. C., Just, M. A., & Carpenter, P. A. (1992). Working memory constraints on the processing of syntactic ambiguity. Cognitive Psychology, 24(1), 56–98. Martin, R. C., Shelton, J. R., & Yaffee, L. S. (1994). Language processing and working memory: Neuropsychological evidence for separate phonological and semantic capacities. Journal of Memory and Language, 33(1), 83–111. Maybery, M. T., Bain, J. D., & Halford, G. S. (1986). Informationprocessing demands of transitive inference. Journal of Experimental Psychology: Learning, Memory and Cognition, 12(4), 600–613. Medin, D. L., & Shoben, E. J. (1988). Context and structure in conceptual combination. Cognitive Psychology, 20(2), 158–190. Miyake, A., Just, M. A., & Carpenter, P. A. (1994). Working memory constraints on the resolution of lexical ambiguity: Maintaining multiple interpretations in neutral contexts. Journal of Memory and Language, 33(2), 175–202. Murphy, G. L. (1988). Comprehending complex concepts. Cognitive Science, 12(4), 529–562. Murphy, G. L., & Medin, D. L. (1985). The role of theories in conceptual coherence. Psychological Review, 92, 289–316. Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (1998). The University of South Florida word association, rhyme, and word fragment norms. Retrieved from http://w3.usf.edu/FreeAssociation/ Oberauer, K. (2009). Design for a working memory. In B. H. Ross (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 51, pp. 45-100). San Diego, CA: Elsevier Academic Press. Oberauer, K., & Göthe, K. (2006). Dual-task effects in working memory: Interference between two processing tasks, between two memory demands, and between storage and processing. European Journal of Cognitive Psychology, 18(4), 493–519. Oberauer, K., Süß, H.-M., Wilhelm, O., & Wittmann, W. W. (2008). Which working memory functions predict intelligence? Intelligence, 36, 641–652. Pashler, H. (1994). Dual-task interference in simple tasks: Data and theory. Psychological Bulletin, 116(2), 220–244. Posner, M. I., & Boies, S. J. (1971). Components of attention. Psychological Review, 87, 391–408. 206 B.J. Ramm and G.S. Halford Posner, M. I., & Klein, R. (1973). On the functions of consciousness. In S. K. Kornblum (Ed.), Attention and performance IV (pp. 21–35). New York: Academic Press. Rogers, T. T., & McClelland, J. L. (2004). Semantic cognition: A parallel distributed processing approach. Cambridge, MA: MIT Press. Shwartz, S. P. (1976). Capacity limitations in human information processing. Memory and Cognition, 4(6), 763–668. Wickens, C. D. (1984). Processing resources in attention. In R. Parasuraman & D. R. Davies (Eds.), Varieties of attention (pp. 63–102). New York: Academic Press. Wisniewski, E. J. (1996). Construal and similarity in conceptual combination. Journal of Memory and Language, 35(3), 434–453. Wisniewski, E. J. (1997). When concepts combine. Psychonomic Bulletin and Review, 4(2), 167–183. Wisniewski, E. J., & Murphy, G. L. (2005). Frequency of relation type as a determinant of conceptual combination: A reanalysis. Journal of Experimental Psychology: Learning, Memory and Cognition, 31(1), 169–174. APPENDIX I Experiment 1 materials Sensible test phrases (Gagne & Shoben, 1997) adventure months air power birth rash cable brake chocolate plant city committee coal country college publication cooking treatment cream bread family cow financial headache finger nerve fish gland flu pills gas cloud growth range gutter rodent heat iron historical elder home light horse toy job conflict juvenile swing manual utensils marine antique marital instincts maternal pains morning prayers moth signals mountain magazine murder film musical town nose sound office charge oil moisturizer olive season paper model party toast pine dust sap remedy servant language smoke problem sports pain star drama steam equipment student scandal summer money tax pressure traumatic tale urban album vapour drops vegetable cube voice vote winter breeze wood allergy Sensible filler phrases acid burn alligator knife alpine forest anatomy guide ankle bump ant powder aquatic mammal arm exercise art fund ash bin baby smile bicycle dent billing network bitter peel boar roast bone weapon bronze statue budget list butter pudding cabin trail cabinet photo cake oven casino loss chalk artist cider drum cigarette ban coin game cookie feast criminal file crowd rumble dinner music drink swirl enemy strength farm wedding fiancée vow forest tiger fracture swelling gallows trial garbage rake gestation stage glass pen gold drill harvest curse heart design history conference honey perfume ideological doctor incense sneeze industrial lock jail years lawyer story letter folder lightening rod margin note meditation corner mining engineer minnow stream mirror cloth mitten knitter monkey puzzle mouse phobia muscle tear neighbourhood bar number science ocean warning office cup oval sport parking indecision parrot veterinarian planet mass plastic report pole height political sticker potato sauce property law radio communication rain lake reform meeting restaurant review rice picker robin attack royal massage sand shield show title silk bug social lounge song review spine inflammation storm noise tank transport television shelf tennis rage tile pattern tooth pain torch holder tractor shaft train junk truck factory trumpet store vacation humidity valley village war parade water bird wool string © 2012 The Australian Psychological Society Processing demands and concept combination 207 APPENDIX I Continued Nonsense phrases acquaintance monster angel tube anvil taste anxiety canoe applause bridge apple gravel argument brush balcony coal balloon threat band noose basin friend bath carriage beetle sponge bell shirt bleach moustache bolt mushroom boss well box birthday brain kiosk brass blood bun nuisance bus temple cabbage cord cage smog captive feather carrot necklace chicken light child metal chin mission clam hose clown banana colour ring compass dock cork reading costume digestion counter study courage rack credit scissors crystal moisture cub keyboard day grave debt rib desert spray dew fire diagram axe discretion snow doctor soil dog crayon dwarf ray ear smirk edge gel egg hook elastic dawn electric favour elephant tyre fabric scholar fear shelf felt barbecue fiction ink fire dolphin flower prisoner fox smoke fuzz kettle galaxy sail © 2012 The Australian Psychological Society gin fibre grape golf grit hospital groceries energy hammock paper hate reed helium carnivore highway sheppard hog cotton horn canine horse pope hour ice idea linen jar disease knot key lemonade marriage lion fin lip hammer lizard cotton lung beer map grass mask toe melon punishment mental cart mine meow mineral girl moon fork nail fountain nut satire olympic pineapple pack phone park laser partner climate peace mosquito pear wind pebble thread penny coat piano cook pizza whip plan shop plank dose plate dark platform diet poem revolt pork mine possum sword postage cat pounce library princess salmon punch tablet puppet vault purple grace rug pie salt secretary saw game scotch shoe sheep gymnast shelf river shell cop ship vision shoe disability sky switch slime race slug school song virus space cherry spear rose spot team spring cave stone match stork stitch straw ladder sugar gallery sun pot tantrum eagle tap skull teacher arms television fur thick responsibility thigh candidate tile hood time tiger tin weather tooth seed traffic bed trail curtain truck blindness trumpet bank vanilla comet velvet account violin swan wall trousers war toddler wave cookie wing laugh word sprinkle 208 B.J. Ramm and G.S. Halford APPENDIX II Experiment 2 materials Familiar test phrases academic year art gallery bank statement beach sand boat deck book shelf brand name bronze medal bus ticket business hours centre stage chicken feed circus tent coffee cup Familiar filler bath towel brain damage camp ground cave man civil rights fountain pen ground floor immune system infant mortality lamp post mirror image night school nursery rhyme pirate ship pocket knife prime minister private life rock music rope ladder snow boot song bird sound wave space shuttle stainless steel sugar cane theme park time bomb world record computer game cricket match crystal ball desert island fire truck hand grenade insurance policy intelligence score junk mail kitchen sink labour camp leather jacket lemon juice middle class Novel filler anatomy guide aquatic mammal ash bin budget list butter pudding casino loss cider drum enemy strength fracture swelling grain law jail years margin note mining engineer minnow stream mouse phobia muscle tear ocean warning planet mass political sticker reform meeting restaurant review rice picker spine inflammation tank transport vacation humidity valley village wax burn wool string mobile phone parking space place mat police station primary school prison bars public toilet question mark rain water raw material remote control report card room temperature rubber band screen saver sex drive skeleton key slave labour social security speed camera stock market stop sign table tennis tea leaves tobacco pipe tomato sauce traffic jam treasure chest Word non-word pairs bear raplan bitter pog bont brush breath bostle cigarette insede cike oven clock diot clothes lacone courage ritch criminal follod crish helmet curdail sport dince floor gold druct gont monkey guff machine harch harvast heak design helmet collife humson forest industrial leck jold snack lamade marriage lant powder leif shovel mactus veterinarian milk bafire mitten complun mupe road pabio knife package telmer pallow ban paper ganem perfume voney princess famon pule height relice fabric ribin knitter river jeg rocket frew royal missage salad attick sameg comic soofer justice stanex oyster surface humstag suttid stain sweat charote syrup hension tennis pone thruel guard trevol trunk wemach pump wheat motch wrist selact yater incense Note: The Novel Test phrases used in Experiment 2 were the same as the ‘Sensible Test Phrases’ used in Experiment 1 (see Appendix I). © 2012 The Australian Psychological Society