Gray matters: Age-related differences in
context-dependent idiom processing
Name : Amélie la Roi (s1781960) Contact details : amelielaroi@gmail.com
Program : Research Master Language and Cognition Date : May 31, 2016
1 TABLE OF CONTENTS
Table of contents 1
Abstract 3
1. Introduction 4
2. Cognition in young and old age 5
2.1. Working memory 5
2.1.1. Age-related decline in working memory 7
2.2. Inhibition 8
2.2.1. Age-related decline in inhibition 10
2.3. Language and cognition: how aging influences language 11 3. Language in old age: when meaning is not straightforward 13
4. Models of idiom processing 15
5. Using the ERP methodology to study idioms in context 16
5.1. The N400 17
5.2. The P600 18
5.3. Previous research on idiom processing and comprehension 19
6. Research questions and hypotheses 22
7. Lexical decision experiment 25
7.1. Methods 25
7.1.1. Participants 25
7.1.2. Materials and Design 25
7.1.3. Procedure 28
8. Results lexical decision experiment 28
8.1. Accuracy 29
8.2. Reaction times 30
9. Discussion lexical decision experiment 32
10. EEG experiment 34
10.1. Methods 34
10.1.1. Participants 34
10.1.2. Materials and Design 35
10.1.3. Procedure 35
10.1.4. Reading Span test 36
2
10.1.6. Verbal fluency test 38
10.2. EEG recording 39
10.3. ERP analysis 39
11. Results EEG experiment 40
11.1. First time window: 200-300 milliseconds 41
11.2. Second time window: 300-400 milliseconds 44
11.3. Third time window: 400-500 milliseconds 46
11.4. Fourth time window: 500-800 milliseconds 49
11.5. Fifth time window: 800-1100 milliseconds 52
11.6. Individual differences 56
12. Discussion EEG experiment 57
12.1. N400 57
12.1.1. Idiom processing 57
12.1.2. Context-dependent idiom processing 60
12.2. P600 62
12.2.1. Idiom processing 62
12.2.2. Context-dependent idiom processing 63
13. Conclusion 64
14. Acknowledgements 65
15. References 66
3 Abstract
How does age-related cognitive decline affect elderly adults’ ability to use context when processing idioms? Research has demonstrated that people’s cognitive functions decline with age. Compared to younger adults, elderly adults have shown impaired inhibition skills (e.g., Hasher, Stolzfus, Zacks, & Rypma, 1991), as well as reduced working memory capacity (e.g., Van der Linden et al., 1999). In the processing of idioms, such as the Dutch in de soep lopen, literally meaning “to walk in the soup” but figuratively “to fail”, cognitive inhibition helps us in suppressing the idiom’s literal meaning to reach its figurative interpretation. Furthermore, working memory enables us to use contextual information to predict an upcoming idiom, facilitating activation and interpretation of its intended, non-literal meaning.
4 1. Introduction
The process of aging is often associated with gaining wisdom. It is true that elderly adults have demonstrated to have more world knowledge (Light, 1992). However, the more detrimental effects of aging are even more prominent. When people grow older, their cognitive functions decline. Although some researchers suggest that it is merely a slowing in processing that drives age-related cognitive decline (e.g., Fisk & Sharp, 2004), many studies report a cognitive decline due to aging in functions other than processing speed. A well-known aspect of cognition that is frequently studied in research on cognitive aging is episodic memory loss (e.g., Craik, 1977). Other prominent factors in cognitive aging research are working memory (e.g., Van der Linden et al., 1999) and cognitive inhibition (e.g., Hasher et al., 1991).
The process of aging also affects elderly adults’ language. Although some aspects of language, such as vocabulary knowledge (Salthouse, 1993), have been demonstrated to be preserved when people age, other linguistic abilities show impairment. Research on language in old age suggests that particularly higher order language processes, such as context-dependent language processing (Federmeier et al., 2003) are affected by age-related decline. As contextual information enables people to predict and pre-activate upcoming words, thereby facilitating the processing of those words (e.g., Federmeier & Kutas, 1999), a deficiency in the ability to effectively use context may impair elderly adults’ language processing skills to a large degree.
A supportive context can become especially helpful when a linguistic expression can be interpreted in multiple ways. This is for example the case in idioms, such as the Dutch “tegen de lamp lopen”, meaning “to get caught” or “in de soep lopen”, meaning “to fail”. In order to derive the intended interpretation of the idiom “tegen de lamp lopen” (“to get caught”), its literal meaning (“to walk against the lamp”) needs to be suppressed. The cognitive function inhibition helps us in suppressing an idiom’s literal, non-relevant, meaning. However, as inhibitory skills decline with age (e.g., Hasher et al., 1991), elderly adults may experience difficulties in idiom processing and comprehension. Given the fact that idioms form an important part of the language we use every day (Jackendoff, 1995; Sprenger, 2003), impaired idiom processing may hamper the ability of elderly adults to successfully communicate, which in turn affects their quality of life. For elderly adults, contextual information may thus be especially important to derive the correct interpretation of idioms. The question therefore is to what extent elderly adults are able to use rich contextual information to facilitate the processing and comprehension of idioms.
5
regarding online language processing (Luck, 2014), the ERP methodology is a suitable technique to study the fast processes involved in the processing of language in context. The following sections will first provide some theoretical background on the functions of cognition that play a role in context-dependent idiom processing and that are under the influence of age-related decline. This theoretical overview will be followed by a discussion of previous studies describing the influence of cognitive decline due to aging on language in general. The third section considers the question how age-related decline affects the processing of figurative language, in particular idioms. Different models of idiom processing will be explicated, followed by a discussion of the ERP components relevant to an online study of the processing of idioms in context. Finally, the various theoretical perspectives and experimental findings will be combined to formulate the research questions and hypotheses. The present study investigates how elderly compared to younger adults use contextual information to process idioms. Due to age-related cognitive decline, elderly adults are expected to be less able to benefit from a supportive context to facilitate the processing of idiomatic expressions.
2. Cognition in young and old age 2.1. Working memory
In 1968, the scientists Atkinson and Shiffrin suggested that the overall cognitive function of memory could be divided into two components: a temporary short-term memory (STM) system and a long-term memory (LTM) system. They stated that the STM could be considered as an “antechamber” of LTM. In addition, they argued that the STM system operates as a working memory, serving as a workplace in which operations underlying all kinds of complex cognitive processes could take place. Later, the notion of working memory was further investigated by Baddeley and Hitch (1974). They developed a model of working memory consisting of three components: the phonological loop, the visuo-spatial sketchpad, and the central executive (see Figure 1).
Figure 1. Model of working memory as proposed by Baddeley and Hitch (1974), consisting of three components: 1) central executive 2)
phonological loop 3) visuospatial sketchpad. The phonological loop and visuospatial sketchpad support the central executive, which controls attention and behavior.
6
subvocal rehearsal component of the phonological loop. As opposed to the phonological loop handling incoming phonological information, the visuo-spatial sketchpad uses visual, spatial, and kinesthetic information to construct mental representations. These representations can be temporarily stored and manipulated (Baddeley, 2003). In the end, the phonological loop and visuospatial sketchpad together provide the input for the central executive, which is assumed to exclusively control the division of attention between the phonological loop and the visuospatial sketchpad, without having any storage function (Baddeley & Logie, 1999).
The three-component model originally proposed by Baddeley and Hitch (1974) seemed to clearly describe the structure of working memory. However, it could not sufficiently explain how information processed in the verbal-acoustic and spatial-temporal subsystems could finally enter into the LTM system. Furthermore, the amount of information entering into the subcomponents exceeded the storage capacity of both systems, which meant that there needed to be another location where this information could be stored (Baddeley, 2003). This assumption was based on research on the retention of prose passages, which were initially considered to be retained in LTM. However, patients experiencing severe deficiencies in LTM appeared nevertheless to be able to immediately recall prose passages (Baddeley & Wilson, 2002). This pointed to the existence of a location different from LTM where information could be temporarily stored. To account for this finding, Baddeley (2000) proposed to add a fourth working memory component to the existing three-component model of Baddeley and Hitch (1974): the ‘episodic buffer’ (see Figure 2).
The episodic buffer is assumed to be responsible for storing multi-modal information from the different systems within the STM and LTM. The central executive binds all this information into a single code, called ‘episodes’, which can be temporarily stored in the episodic buffer before entering LTM (Baddeley, 2000). Although elderly adults have been shown to be able to use the
Figure 2. Current model of working model as proposed by
7
episodic buffer, for example for the immediate recall of prose passages (Baddeley & Wilson, 2002), working memory in general declines with age. The next section presents the findings of previous research that show an age-related decline in working memory and indicate how the working memory decline affects elderly adults’ language.
2.1.1. Age-related decline in working memory
The working memory is one of the cognitive functions that is affected by decline due to aging. In 1988, Salthouse, Kausler, and Saults already demonstrated that people’s ability to remember the identity or position of seven target items in a 5-by-5 matrix significantly declined with age. Performance on tasks involving both processing and storage of information also proved to be subject to age-related differences. In such a task participants for example answered questions about sentences and at the same time had to remember the last word of each sentence for later recall. Apart from decline in working memory itself, working memory scores seemed to play a crucial role in explaining age-related variance in other cognitive measures as well (Salthouse, 1991). Subsequent investigation of the relationship between age and working memory performance suggested that the link between the two was largely mediated by processing speed. The researcher argued that a decline in processing speed hampered elderly adults’ ability to rapidly encode or activate relevant information (Salthouse, 1992; 1994; 1995), affecting participants’ performance on working memory tasks.
8
comprehension and verbal long-term memory. Moreover, the influence of speed and resistance to interference proved to be mediated by working memory as well.
Finally, in an event-related potential study Federmeier and Kutas (2005) found that elderly adults’ reduced ability to benefit from rich contextual information to predict, and therefore facilitate processing of, upcoming words, could be ascribed to their lower reading span scores which are thought to indicate working memory capacity. Thus, in contrast to Dede et al. (2004), who did not find an effect of working memory on online syntactic processing, the findings of Federmeier and Kutas (2005) suggest that working memory does have an influence on online semantic processing. In conclusion, the results of the studies described above indicate the importance of working memory in constructing the meaning of one or multiple sentences. As in context-dependent idiom processing people also need to construct and derive the correct meaning of the sentence, working memory is expected to play a role in to the process of processing idioms in context.
2.2. Inhibition
Another function of cognition that is subject to age-related decline is inhibition. Inhibition belongs to the larger set of executive functions, which enable people to control and regulate all forms of incoming information (Miyake & Friedman, 2012). Many researchers have tried to classify the processes underlying the construct of inhibition as a whole. Harnishfeger (1995) for example categorized inhibitory processes based on three dimensions: 1) intentional/unintentional or conscious/unconscious 2) behavioral/cognitive 3) inhibition/resistance to interference. This last dimension was used to distinguish between active inhibitory processes applied to the contents of working memory (Wilson & Kipp, 1998), called inhibition, and processes preventing irrelevant information from actually entering working memory, described as resistance to interference. Based on the dimensional classification system of Harnishfeger (1995), Nigg (2000) constructed four categories of inhibition relevant to the field of psychology: 1) interference control, suppressing interfering information competing with other stimuli or resources 2) cognitive inhibition, keeping irrelevant information out of WM, 3) behavioral inhibition, suppressing prepotent responses, and 4) oculomotor inhibition, serving suppression of saccadic eye-movements.
9
also relevant to the present study, as it important to use the correct inhibition task to measure the type of inhibition assumed to be involved in context-dependent idiom processing. To examine which type of inhibition was empirically measured by which task, Friedman and Miyake (2004), carried out a latent-variable analysis on three major types of inhibitory processes: Prepotent Response Inhibition, Resistance to Distractor Interference (DI), and Resistance to Proactive Interference (PI). Friedman and Miyake (2004) chose these three categories based on the categories of inhibition described in the literature and the taxonomy of Nigg (2000), combining Nigg’s behavioral and oculomotor inhibition into one category named Prepotent Response Inhibition. Prepotent Response Inhibition refers to the ability to intentionally inhibit an automatic or dominant response, such as naming the green color of the letters forming the word ‘red’ in the Stroop task (Stroop, 1935). Participants thus need to suppress the dominant tendency to read the word instead of naming the color in which the word is printed. Resistance to DI involves the ability to suppress interference of task-irrelevant information present in the external environment. This function is tested in tasks like the Eriksen flanker task (Eriksen & Eriksen, 1974) requiring participants to identify a target, for example the letter ‘H’, while ignoring distractors accompanying the target, such as the letter ‘K’. Finally, Resistance to PI describes the ability to ignore information from memory that was previously relevant to the task at hand, but in the meantime has become task-irrelevant. An example of a task testing this ability is the AB-AC-AD task (Rosen & Engle, 1998). After acquiring a first set of cue-response pairs, participants need to learn and recall a new list in which the cues from the previous list are combined with new response words. Participants thus need to suppress intrusions of the initially learned pairs in order to recall the correct, newly formed pairs.
10 2.2.1. Age-related decline in inhibition
Although the different taxonomies of inhibition-related functions described above differ from each other to some extent, they have one thing in common, namely the fact that all of the mentioned processes seem to require executive control. Based on studies on patients with frontal lobe damage, who experience deficits in executive functioning (West, 1996), researchers argue that the frontal lobe region is important in executive processes. This finding is of significant relevance to the present study on context-dependent idiom processing in normally aging adults. Since in the brain of normally aging adults the prefrontal area is actually first and the most affected by age-related decline (Scheibel & Scheibel, 1975; Cabeza, Nyberg, & Park, 2005), this may explain elderly adults’ reduced capacities with respect to processes involving executive functioning (e.g., Van der Linden, Beerten, & Pesenti, 1998). Furthermore, as these executive functions, in particular inhibition, are important to facilitate the processing of idioms in a supportive context, age-related decline in executive functioning could cause elderly adults to be less able to benefit from contextual information when processing idioms.
According to Hasher and Zacks’ inhibition deficit hypothesis (1988), elderly adults’ impaired performance on tasks requiring frontal lobe functioning can be ascribed to deficient inhibitory processes in working memory. Their hypothesis is supported by several empirical findings. Hasher et al. (1991) for example found that elderly adults, in contrast to their younger equivalents, showed no slowing in the time needed to name a target letter that was used as a distractor on a previous trial, indicating that the inhibition of a response to a distracting stimulus on a previous trial did not proceed onto the next trial. In other words, the elderly adults did not demonstrate a negative priming effect, while younger adults did. According to Hasher et al. (1991), their findings indicated an age-related reduction in inhibitory capacities in the elderly group. The results of Hasher et al. (1991) are corroborated by other studies, for example showing that elderly adults are less able than younger adults to inhibit information that is no longer relevant to the task at hand to enter memory (Hartman & Hasher, 1991), not even when they are explicitly instructed to do so (Zacks, Radvansky, & Hasher, 1996). Nevertheless, Hasher, Quig, and May (1997) found that providing elderly adults with additional information directing them to the relevant information to be retained helped them to inhibit the irrelevant information.
11
through top-down processes. According to Dempster (1992), the deterioration of the frontal brain area due to aging would cause elderly adults inhibitory capacities to decline. Therefore, elderly adults also process irrelevant information, decreasing the available resources for the processing of important information.
The goal maintenance theory, described by Braver and West (2008) also elaborates on suppression of irrelevant information. Braver and West (2008) argue that age-related decline in the functioning of the lateral (PFC) impairs elderly adults’ ability to focus on information relevant to the goal of their task or behavior. The maintenance of goals and behavior helps in selecting which incoming information is further processed and which is not. Furthermore, the goal maintenance account states that the lateral PFC not only supports selective attention, but also biases the processing of information in other brain areas, such as those responsible for memory retrieval. According to the goal maintenance theory, age-related damage to the lateral PFC thus not only affects inhibition of externally, but also internally generated information.
It thus seems that inhibitory processes play an important role in age-related cognitive decline. However, when Fisk and Sharp (2004) investigated how the executive functions examined by Miyake et al. (2000), one of them being inhibition, would be influenced by the factor age, they found that after correcting for elderly adults’ general slowed speed of processing, the factor inhibition no longer added any individual variance to elderly adults’ performance on tasks of executive functioning. This finding refutes the importance of inhibition in age-related cognitive decline, and rather provides support for the general slowing theory of cognitive aging (Salthouse, 1996). Clearly the debate about what processes lie at the basis of cognitive decline due to aging has not been settled yet.
2.3. Language and cognition: how aging influences language
12
al., 2004; Caplan & Waters, 2005) and reading comprehension, as well as on the processing of sentences with a complex grammatical structure (Norman et al., 1992).
Besides using behavioral experiments to study the effect of aging on language, researchers have also used online research methods to examine the influence of age-related cognitive decline on language. Online research methods are suitable to investigate the fast incremental processes involved in the processing and interpretation of language. Federmeier et al. (2003) for example compared elderly and younger adults’ language processing at a word level, but also investigated whether and how elderly adults used contextual information in the same way as younger adults when interpreting information at a sentence level. No pronounced differences were found at the word level, but elderly adults’ language processing was significantly less affected by message level constraints than that of younger adults. This suggests that elderly adults are less able to make use of contextual information to facilitate language processing. The results of Federmeier et al. (2003) are supported by the study of Federmeier and Kutas (2005), who also demonstrated that in comparison to their younger counterparts, elderly adults benefitted less from contextual information during language processing. Moreover, Federmeier and Kutas (2005) were able to link elderly adults’ reduced ability to efficiently use contextual information in order to facilitate online language processing to their lower scores on a working memory task.
13
adults did not use contextual information in a predictive manner, therefore missing the facilitative effect. Just as in the study of Federmeier et al. (2002), elderly adults with high scores on a category fluency task tended to show a processing pattern similar to that of younger adults, using contextual information in a predictive manner in the processing of language.
In conclusion, the studies above show that age-related decline in cognition not only affects general cognitive functions, but also affects language by decreasing people’s ability to benefit from supportive contextual information in language processing. The next section discusses studies that investigate the processing and comprehension of multiple layer language, in which rich contextual information can be especially helpful to derive the correct meaning of these expressions. However, as the ability to benefit from contextual information requires cognitive functions that are subject to age-related decline, elderly adults may be less able to profit from a supportive context to process and interpret language with multiple meanings, in particular idioms.
3. Language in old age: when meaning is not straightforward
When a word or sentence has multiple meanings, contextual information becomes particularly useful in deriving the correct interpretation of language. This is for example the case in ambiguous words. In younger adults, correctly interpreting lexically ambiguous words has been linked to working memory and being able to keep multiple meanings active in mind (Miyake, Just, & Carpenter, 1994). Given the age-related decline in working memory (Van der Linden et al., 1999), elderly adults may have difficulties deriving the intended interpretation of an ambiguous word. In this case, contextual information supporting one particular meaning of the ambiguous word more than the other can help finding it’s correct interpretation. Meyer and Federmeier (2010) indeed showed that elderly adults processed ambiguous words less efficiently than their younger counterparts. However, both younger and elderly adults used contextual information to select the meaning of an ambiguous word, although the elderly group was less able to, if necessary, revise the word’s meaning initially selected. Meyer and Federmeier (2010) ascribed the elderly adults’ deviant strategies for the interpretation of ambiguous words to their reduced inhibition skills. Elderly adults who scored high on a measure of inhibition showed an interpretation pattern similar to that of the younger adults group.
14
interpretation, a process requiring cognitive inhibition. Inhibition is also involved in the comprehension of metaphors, such as “Lawyers are sharks” (Glucksberg, Newsome, & Goldvarg, 2001). The metaphorical meaning of a metaphor is constructed by transferring the meaning of the so-called source, in this case “sharks”, to the target, “lawyers” (Lakoff, 1993). In an online study on the processing of metaphors, Weiland, Bambini, and Schumacher (2014) found that the interpretation of metaphors was facilitated when metaphorical expressions were preceded by a prime related to the literal meanings of the metaphor. Based on this finding, the researchers suggest that when interpreting a metaphor, people initially activate its literal meaning, which is subsequently suppressed in order to derive the metaphor’s figurative interpretation. Since inhibitory capacities have been shown to decline with age (e.g., Hasher et al., 1991), elderly adults may have difficulties with the interpretation of figurative expressions, for example those involving irony or metaphors. However Hasher et al. (1997) showed that elderly adults benefit from additional information to process upcoming words. Therefore, it is also conceivable that contextual information can support elderly adults’ ability to derive the intended meaning of figurative language.
A special type of figurative language involving inhibition of the literal meaning are idioms such as the Dutch in de soep lopen (literally “to walk into the soup”, figuratively “to fail”) (Galinsky & Glucksberg, 2000). Idioms can be characterized as having an invariable form. Furthermore, idioms are an essential part of the language we use every day (Jackendoff, 1995; Sprenger, 2003). What distinguishes idioms from other forms of figurative expressions is that the idiomatic meaning of an idiom cannot be derived from the meanings of its constituent parts (Weinreich, 1969). For example, the meaning of in de soep lopen does not have anything to do with “soup”, nor with “walking”. As a consequence, an idiom’s literal meaning necessarily needs to be suppressed in order to understand the idiom as a whole. This assumption is supported by research involving neuro-imaging methods. Compared to literal processing, subjects activate additional areas in the PFC during the processing of idioms (Lauro, Tettamanti, Cappa, & Papagno, 2008; Zempleni, Haverkort, Renken, & Stowe, 2007). The PFC has been argued to play an important role in executive functioning (e.g., Hasher & Zacks, 1988). Given the decline in executive functions, and more specifically in inhibition in the elderly, they may have difficulties deriving an idiom’s intended interpretation.
15
memory as a whole (e.g., Gibbs, 1994) or whether the listener first interprets the idiom literally and only after realizing that this interpretation does not fit into the context, derives the idiom’s idiomatic interpretation (e.g., Grice, 1975). The following section discusses the different models that have been proposed to explain how people process and interpret idioms.
4. Models of idioms processing
Models describing how people derive the non-literal meaning of an idiom can be roughly divided into three categories. On the one end are models such as the direct access model (Gibbs, 1994) stating that an expression’s figurative meaning is stored in the mental lexicon as a whole and is therefore retrieved directly without activating literal word meanings. The idiom list hypothesis (Bobrow & Bell, 1973) also presumes a special type of processing for idioms. According to this view, idioms are stored in a special list apart from the normal lexicon. In order to access these idioms, the listener must be in an “idiom mode” of processing which differs from regular literal processing. Thus, the processing of an idiom starts with literal analysis, but soon switches to the “idiom mode” of processing. However, process making the listener switch to the idiom mode remains undefined.
As opposed to models assuming a special manner of processing for idioms, the standard pragmatic model (Grice, 1975) states that listeners first need to fully access and reject an expression’s literal meaning, before obtaining its figurative meaning. Where the standard pragmatic model presumes consecutive processing of respectively an idiom’s literal and figurative meaning, the lexical representation hypothesis (Swinney & Cutler, 1979) states that literal and figurative meaning activation of an idiom take place simultaneously. Idiomatic expressions are processed in the same way as literal expressions. To account for the multiple meanings of an idiom, Swinney and Cutler (1979) argue that these are processed in a way comparable to lexical ambiguity processing.
16 configuration hypothesis (Cacciari & Tabossi, 1988), idioms are processed literally until a certain activation threshold for the idiom’s figurative meaning has been reached. After accessing the ‘key’ of the idiom, the point at which the idiom is recognized as such, literal processing stops. Finally, models adopting a hybrid representation of idioms state that idioms have their own lexical concept node in the mental lexicon from which activation can spread to the lemmas building the idiom (Cutting & Bock, 1997). In terms of Sprenger, Levelt, and Kempen (2006), idioms can be considered as superlemmas that co-activate meanings that are related to the meaning of the idiom as a whole, but also the meanings of the single lemmas that build the idiom.
The extensive number of models described above illustrates that researchers have not reached consensus yet about how idioms are processed in the human brain. By using the online ERP methodology, the results of the present study can indicate whether and to what degree younger and elderly adults activate literal word meanings during idiom processing, providing empirical evidence for one model or the other. Moreover, the findings of the present study can provide insight in how constraining contextual information influences idiom processing in younger as compared to elderly adults. The following section elaborates on how the ERP methodology will be used to study context-dependent idiom processing in elderly adults.
5. Using the ERP methodology to study idioms in context
17 5.1. The N400
The ERP component N400 has originally been found by Kutas and Hillyard (1980). They discovered that the voltage of ERPs in response to sentences containing a semantically anomalous word was more negative compared to correct sentences. This cross-conditional difference in voltage peaking around 400 milliseconds following stimulus onset was called the N400 effect, which should not be confused with the N400 component that is found in response to any meaningful stimulus. Since the discovery of Kutas and Hillyard (1980), the N400 effect has been found in numerous manipulations, linguistic as well as non-linguistic (Kutas & Federmeier, 2011). Several of these findings are relevant to the investigation of idioms. For example, the N400 is thought to reflect processes of word retrieval and integration (Kutas & Federmeier, 2011). Nevertheless, although several studies have used the N400 effect as a measure of literal word integration processes (e.g., Rommers et al., 2013), in their Retrieval-Integration account Brouwer, Fitz, and Hoeks (2012) argue that the N400 only indexes memory retrieval of all features (semantic, syntactic, as well as pragmatic) of the word to be processed. According to Brouwer et al. (2012), subsequent integration processes are reflected by the P600 ERP component, which will be discussed in more detail below. In idiom research, the N400 as an index of the process of word retrieval and perhaps integration can be used to examine how the meaning of an opaque idiom (an idiom whose meaning cannot be derived based on the meaning of its constituent parts) is stored in and retrieved from semantic memory. Furthermore, as the strength of the N400 effect has also been shown to be modulated by working memory load (Gunter, Jackson, & Mulder, 1995), it could be used to investigate how effortful the processing of idiomatic sentences is for younger compared to elderly adults.
Besides reflecting word retrieval and possibly integration, the N400 component has been shown to be influenced by contextual information. Contextual information serves to pre-activate features of expected words, facilitating processing of these specific words, but also of words sharing features with the expected word (Kutas & Federmeier, 2000). In the N400, this facilitative effect is reflected as a decrease in the component’s amplitude. With respect to idioms, the context-dependent N400 effect is relevant to study whether and to what extent contextual information pre-activates the non-literal meaning of opaque idioms, reducing the component’s amplitude.
18
2011). In the processing of idioms, this characteristic of the N400 could indicate how fast people predict an upcoming idiomatic expression, and thus retrieve its intended meaning from semantic memory, compared to literal expressions. If prediction of an idiom is fast, this increases the cloze probability of the idiom’s key and therefore reduces the N400 amplitude in response to the key.
5.2. The P600
Another well-known ERP component is the P600, which actually represents a family of components with a late positivity. Originally, the P600 was considered to be an index of syntactic reanalysis, such as in sentences containing a syntactic violation (e.g., Hagoort, Brown, & Groothusen, 1993) or garden-path sentences (e.g., Osterhout, Holcomb, & Swinney, 1994). However, the earlier mentioned Retrieval-Integration account of Brouwer et al. (2012) states that the P600 represents processes of integration following lexical retrieval from long-term memory. Brouwer et al. (2012) emphasize the importance of the construction of a mental representation of the linguistic input. In their “mental representation of what is being communicated” (MRC) hypothesis, they state that the amplitude of the P600 component reflects the amount of processing effort that needs to be put into “the construction, revision, or updating of a mental representation of what is being communicated.”
The account of Brouwer et al. (2012) is compatible with the findings of Burkhardt (2006), who found a P600 effect in response to sentences that were effortful to process due to the introduction of new referents or the fact that inference of existing references from previous context was required. The extra processing costs entailed by the complex sentences were indexed by an increased P600 effect, while no syntactic reanalysis was involved. A study of Regel et al. (2011) also demonstrated a P600 effect for complex as compared to control sentences. In this study, the complexity of the stimuli consisted of figurative use of language in the form of ironic expressions. Based on their results, Regel et al. (2011) concluded that the increased amplitude of the P600 in response to ironic sentences reflects the additional effort required for the pragmatic processing of irony.
19
mechanisms underlie the N400 and P600 effect, both components may play a role in context-dependent idiom processing.
5.3. Previous research on idiom processing and comprehension
Various experimental studies have aimed at uncovering the processing steps involved in the interpretation of idioms, but no consensus has been reached yet. Some studies support the idea that an idiom’s meaning is directly retrieved from the mental lexicon without processing of literal word meanings. The findings of Rommers et al. (2013) for example indicated a qualitative difference between the processing of literal and idiomatic sentences, suggesting direct access of the idiom’s non-literal meaning. In literal, but not idiomatic sentences, integration of an incorrect word in the sentence was facilitated when the incorrect word was semantically related to the correct word compared to an unrelated incorrect word. Rommers et al. (2013) argue that the absence of a semantic facilitation effect in idiomatic sentences, indexed by a decrease in N400 amplitude for related target words, supports the view that the meaning of an idiom is retrieved from memory as a whole. If idiom processing would also involve step-by-step processing of the literal meaning of the idiom’s constituents, a semantic facilitation effect would have also been found for incorrect target words related to the literal meaning of one of the idiom’s constituents. However, according to Rommers et al. (2013) the absence of a semantic facilitation effect indicates that in idiom processing, the process of literal word meaning integration is “switched off”, suggesting that the non-literal meaning of an idiom is directly retrieved from memory. Taken from the perspective of the Retrieval-Integration account of Brouwer et al. (2012), the results of Rommers et al. (2013) indeed indicate that the non-literal meaning of an idiom is quickly accessed in and retrieved from the mental lexicon. However, Brouwer et al. (2012) would not draw any conclusions about the consequences of Rommers et al.’s findings (2013) for the theory on literal word integration processes in idiom processing, as these would rather be reflected in the later P600 component.
20
agreement or spelling within the lexical item represented by the idiom. However, the Retrieval-Integration account of Brouwer et al. (2012) could explain the P600 following idiomatic, but not literal sentences, as reflecting difficulties in integrating the idiom’s meaning in the sentence, making it more difficult to form a mental representation of an idiomatic sentence’s meaning compared to of a literal sentence’s meaning. In turn, this interpretation corresponds to the findings of Regel et al. (2011) who found a P600 in response to ironic compared to literal sentences. This led them to interpret the P600 as reflecting increased effect involved in the pragmatic processing of irony, which could finally also play a role in the processing of idioms.
Proverbio, Crotti, Zani, and Adorni (2009) also applied the ERP methodology to study idiom processing and additionally tried to localize brain activation patterns for idiomatic and literal language. The investigators failed to find any activation in the left inferior frontal gyrus (LIFG), a brain area assumed to be involved in suppressing an idiom’s literal meaning, during the processing of idiomatic sentences. Based on these results, Proverbio et al. (2009) concluded that idiomatic meanings are directly accessed without any involvement of inhibition of literal word meanings. The conclusion of Proverbio et al. (2009) is consistent with the fMRI study of Raposo, Moss, Stamatakis, and Tyler (2009). They found that when participants read literal sentences containing an action verb, such as “kick”, areas in the motor cortex were activated. However, no motor cortex activation was found for idiomatic sentences containing the same action verb, for example “kick the bucket”, indicating that the literal meanings of the words composing the idiom were not processed.
21
processing in which, in the process of comprehension, activation spreads from the single lemmas composing the idiom to the lexical node of the idiom as a whole.
Additional support for (initial) literal word meaning activation in idioms comes from an eye-tracking study by Holsinger (2013). Holsinger presented participants with a screen showing four words: one word related to an idiom’s figurative meaning, another word related to an idiom’s literal meaning, and finally two unrelated distractor words. After reading the words aloud, participants listened to a sentence with either a literal or idiomatic contextual bias. While listening to the sentences, participants’ eye gazes to the four words on the screen were recorded. Even while listening to idiomatic sentences, in early stages participants looked significantly longer to the word related to the idiom’s literal meaning than to the distractors. Holsinger (2013) argues that his findings indicate that literal word meanings are activated at least in early stages of idiom processing.
In an ERP study, Canal, Pesciarelli, Vespignani, Molinaro, and Cacciari (2015) compared ERPs time-locked to the first, second, and third word of a literally plausible idiom. They found no differences in N400 amplitude between idioms presented in an idiomatic as compared to those presented in a literal context up until the third word. Canal et al. (2015) interpreted these findings as indicating activation of the literal meanings of an idiom’s constituent words at least until the reader or listener has recognized the idiom as such. Moreover, an additional cross-modal priming task carried out by Canal et al. (2015) showed that when a target was related to an idiom’s last constituent noun, participants responded faster compared to when targets consisted of unrelated words, irrespective of the context in which the idiomatic expression was presented. This suggests that in both literal and idiomatic contexts, literal word meanings are activated.
22
component reflecting effortful pragmatic processing (Regel et al., 2011) could also indicate whether and how contextual information influences the processing of idioms in younger compared to elderly adults. Finally, as the processes underlying the N400 and P600 components have been related to cognitive functions, such as working memory (e.g., Federmeier & Kutas, 2005), the use of these components can provide insight in the cognitive factors driving potential age-related differences in context-dependent idiom processing. Taking together the literature and previous experimental findings brings us to the following research questions and hypotheses.
6. Research questions and hypotheses
To investigate how cognitive aging affects elderly adults’ ability to use contextual information to facilitate the processing of idioms, the following research questions are formulated:
1. Do elderly adults use contextual information in the same way as younger adults when processing idiomatic expressions?
2. How are differences in context-dependent processing of idioms between young and elderly adults reflected electrophysiologically in the N400 and P600 effect?
3. What role do the cognitive abilities of younger and elderly adults play in the N400 and P600 effect in response to the context-dependent processing of idioms?
Given on the one hand the role of cognitive functions in predictive language processing (e.g., Federmeier & Kutas, 2005) and the processing of idioms (e.g., Lauro et al., 2008) and on the other hand the age-related decline in cognitive functions (e.g., Hasher et al., 1991; Van der Linden et al., 1999), elderly adults are hypothesized to be less able to use context to predict the non-literal meaning of upcoming idioms. Electrophysiologically the differences between elderly and younger adults will be reflected in multiple ways. The hypothesis that elderly adults are less able to suppress an idiom’s literal meaning due to a decline in inhibition should result in more equal levels of activation of literal word meanings in the literal and idiomatic sentences in the elderly adults. Therefore, the expected decreased N400 effect for idioms read by younger adults, driven by reduced literal word meaning activation in idiom processing, will be less strong in elderly adults.
23
activation of literal word meanings less strong, rendering literal word integration processes less necessary, and eventually causing these processes to be “switched off” faster (see also Rommers et al., 2013). The contextual facilitation effect in the N400 expected for younger adults is however expected to be weaker in elderly adults, because this population has been shown not to use contextual information predictively in language processing (Federmeier et al., 2002; 2005; 2010). The Retrieval-Integration account of Brouwer et al. (2012) would also predict an increased contextual facilitation effect for idiomatic sentences following a strongly constraining context sentence. However, from the perspective of Brouwer et al. (2012) contextual information would not promote the switching off of literal word integration processes, but only facilitate retrieval of an idiom’s non-literal meaning from memory by pre-activating this meaning.
One thing that should be noted concerning the N400 effect in the elderly adults is that the this effect has been demonstrated to show a general decline with age (Kutas & Iragui, 1998). Nevertheless, as no consensus has been reached yet about the processes driving this age-related decrease in the strength of the effect (Wlotko, Lee, & Federmeier, 2010), it is not expected to significantly interfere with the results of the present study.
In the P600, differences between younger and elderly adults are expected to appear with respect to the context effect. As the P600 has been argued to reflect the extra effort invested in pragmatic processing of language (Regel et al., 2011), a strongly constraining context preceding an idiomatic sentence may facilitate this pragmatic processing, resulting in a reduced P600 effect for idiomatic sentences preceded by a strongly constraining compared to a neutral context. Considered from the perspective of Brouwer et al. (2012), a decreased P600 effect for idioms in a strongly constraining context would indicate that the supportive context makes it easier to construct representation of meaning of an idiomatic sentences. In other words, integration of an idiom’s non-literal meaning into a sentence is facilitated when the sentence is preceded by a strongly constraining context. Nevertheless, since the ability to use contextual information predictively in language processing is affected by age (Federmeier et al., 2002; 2005; 2010), the context facilitation effect in the P600 predicted in younger adults is expected to be weaker in the elderly group.
24
measures of cognition. However, as in the present study the elderly subjects are highly educated and education has been shown to form a protective factor against age-related decline (e.g., Gollan, Salmon, Montoya, & Galasko, 2011; Karp et al., 2004), it could be that the elderly participant group’s cognitive functions are more preserved compare to those of the average population of elderly adults.
The following sections will discuss the methodology and the results of the lexical decision experiment that was carried out with younger adults in addition to the EEG experiment. The goal of the lexical decision experiment was to pre-test the influence of an additional context sentence on literal compared to idiomatic sentence processing in a behavioral measure. Lexical decision has been shown to be a robust measure of lexical access dependent on context (Swinney, 1979). Therefore, asking participants to perform lexical decision on the critical word in a literal and an idiomatic sentence preceded by either a strongly constraining compared to a neutral context sentence provides insight in how quickly the reader activates and retrieves the non-literal meaning of an idiom from the mental lexicon. In turn, this informs us about whether an idiom is processed in the same way as a literal expression, or that the idiom’s idiomatic meaning is directly retrieved from memory as a whole. More importantly, the lexical decision experiment could demonstrate whether a supportive context preceding the idiomatic sentence can facilitate the processing of the idiom’s non-literal meaning, speeding up participants’ lexical decision latencies in idiomatic sentences following a strongly constraining context sentence.
25 7. Lexical decision experiment
7.1. Methods
7.1.1. Participants
For the lexical decision experiment a group of 16 monolingual Dutch students (mean age: 22;3, range: 19;8 – 29;2, male: 3) was recruited from the University of Groningen. The participant group consisted of students of the bachelor program Dutch Language and Culture, the master program Multilingualism, and the Research Master program Language and Cognition. Students took part in the experiment on a voluntary basis. Participants did not report any language impairments. An overview of the participants’ characteristics can be found in Table 1.
Table 1. Characteristics of the participants in the lexical decision experiment.
Measure Data
n 16
Age 22;3 (19;8 – 29;2)
Sex male: 3
Years of Education 17;10
Hours of reading per week 10 - 20 Hours of social activity per week 10 - 20
7.1.2. Materials and Design
The lexical decision experiment had a 2 x 2 x 3 within-subjects design. The first factor included in the design was Context (High/Neutral), which described whether the context sentence was high or neutral in the degree of contextual constraint. Contextual constraint reflects the degree to which contextual information reduces the number of possibilities for upcoming words. The second factor was Idiomaticity (Literal/Idiomatic), specifying whether the test sentence containing the critical word was literal or idiomatic. Finally, the factor Condition (correct/incorrect/pseudo), specified the type of critical word in the test sentence. An example of stimuli in the different conditions can be found in Table 2 in the Appendix. The critical word is printed in bold.
26
meaning of the upcoming correct critical word, thereby increasing response latencies if the correct word was substituted by an semantically incorrect word.
The materials of the lexical decision task were adapted from the materials of Rommers et al. (2013). Their set of materials comprised literal as well as idiomatic sentences and had been pre-tested on familiarity, transparency, semantic relatedness, cloze probability, and plausibility. In the idiomatic sentences the critical word consisted of the last noun of the idiomatic expression. In the literal sentences the critical word was the same noun as in the idiomatic sentence, but presented in a literal context. It was made sure that the critical word was never the last word of the sentence to avoid potential sentence wrap-up effects to influence the data.
The study of Rommers et al. (2013) included a condition in which the critical word was replaced by a different word of the same semantic category to investigate literal word activation in idiomatic compared to literal expressions. As the present study, in contrast to Rommers et al. (2013), was as not specifically aimed at examining literal word activation, the semantically related condition was not relevant for our study and was therefore removed. While this condition was eliminated, another condition, in the form of a context manipulation, was added. The context manipulation consisted of two types of context sentences preceding the sentence containing the critical word. One of the types of context sentences, the high constraint context sentence, provided the reader with strongly constraining contextual information regarding the content of the following idiomatic or literal sentence and its critical word. The other type of context sentence was neutral, not giving a clear indication of what would follow in the next sentence. To prevent the experimental results from being influenced by varying complexity of the context sentences, all the sentences had a length of nine words. Moreover, the context sentences contained the same number of constituents: a subject NP, a transitive verb with its corresponding direct or prepositional object, and a PP. The order of the constituents was varied to avoid making the context sentences look too artificial, which could affect the ecological validity of the context manipulation.
27
pseudoword, then, renders the mean orthographic LD between this pseudoword and the twenty words in the lexicon that most closely resemble it. The candidate that showed the lowest OLD20 difference from the critical word was selected to replace the critical word in question. If two candidates’ OLD20 value differed to the same amount from the critical word, the deciding factor was the variable Neighbors at distance 1 (in the Wuggy generator abbreviated as Ned1). This factor reports how many words can be created from the generated candidate by means of inserting, substituting, or deleting one letter (Keuleers & Brysbaert, 2010). The candidate with the lowest difference in Ned1 compared to the correct critical word was selected. Mean OLD20 difference between pseudowords and critical words of the experimental items was 0.07 (sd = 0.11) with a mean Ned1 difference of -0.36 (sd = 3.07). After selection of the most suitable candidates, the list of pseudowords was semi-randomly divided over the items. In this way, a correct critical word was not replaced by a pseudoword that was highly similar to it, possibly confusing participants and increasing reaction times exceedingly.
To guarantee an equal division of items over the lists, six additional items with the same structure as the existing items were created and added to the materials of Rommers et al. (2013). Here, an item refers to a sentence pair consisting of a context sentence followed by a test sentence. As the experiment had a 2 x 2 x 3 within-subjects design, there were twelve item types. The total set of materials was randomly divided over 16 lists, controlling for repetition of items within a list and ensuring that each item appeared in all conditions. This means that each participant was presented with the same critical word only once. For example, when a participant was presented with a correct idiomatic sentence preceded by a strongly constraining context, the same participant would also presented with the literal sentence matching the idiomatic one. However, the literal sentence would be preceded by a neutral context sentence and the critical word would be substituted by a semantically incorrect word.
28 7.1.3. Procedure
The session’s procedure was based on the experimental procedure of Rommers et al. (2013). E-Prime 2.0 Professional software was used to present the lexical decision task. Participants were seated at a table in a quiet room. A 14 inch laptop with a resolution of 1366 x 768 was placed at a distance that the subject deemed comfortable. Reaction times (RTs) were recorded by E-Prime through two designated keys ([z] and [m]) on the laptop’s keyboard.
After signing for informed consent, participants were provided with oral instructions that were repeated to them in written form on the screen. They were informed that they would see two sentences, the first one being presented in its entirety and the second one word by word. One of the words in the second sentence would be presented in red. Participants had to judge whether or not this red word was an existing Dutch word by pressing one of the two indicated keys on the keyboard. Participants were told to respond as quickly as possible without making too many errors. Which key a subject had to press to judge the critical word as being a Dutch word was counterbalanced over the total group of participants to control for effects of preference of response hand.
At the beginning of the session, participants were presented with ten practice items to familiarize themselves with the experiment. Each trial started with a fixation cross remaining on the screen for 1000 milliseconds. Subsequently, the first sentence, which was either a neutral or a strongly constraining context sentence, was presented as a whole in the center of the screen for 4500 milliseconds in black letters (Tahoma font, size 18) against a white background. The context sentence was followed by another 1000 millisecond fixation cross. After this, the sentence containing the critical word was presented word by word. Words were displayed in black in the center of a white background, this time in font size 24. Each word remained on the screen for 300 milliseconds with a 300-millisecond blank screen in between. The experiment was split up into six blocks of 26 sentence pairs, which lasted approximately six minutes each. After each block participants were allowed a short break.
8. Results lexical decision task
29 8.1. Accuracy
A linear mixed effect model analysis using the package lmerTest (Kuznetsova, Brockhoff, & Christensen, 2016) was carried out in R (R Core Team, 2015). The model making the best prediction regarding participants’ accuracy scores for the various conditions only includes the factor Condition as a fixed effect and a random intercept for Subject (see Table 3 for the model’s coefficients and z- and p- values). Whether the sentence containing the critical word was literal or idiomatic does not significantly influence participants’ accuracy scores.
Comparing the three levels of the factor Condition based on Tukey contrasts indicates that subjects are more accurate in responding to correct critical words (mean = 97%) than to semantically incorrect critical words (mean = 81%). Moreover, accuracy scores for trials in which the critical word was a pseudo word (mean = 95%) are also significantly higher compared to accuracy scores for trials in which the critical word was a semantically incorrect word. No difference in accuracy was found for correct critical words and pseudo words (p = .276). The distribution of accuracy scores per critical word type is shown in Figure 3.
Table 3. Best fit model predicting accuracy scores Fixed effects
Factor Estimate Std. Error z-value p-value
Intercept 4.2307 0.4578 9.241 < .001
Condition (incorrect) -2.3481 0.3558 -6.600 <.001
Condition (pseudo) -0.5606 0.3695 -1.517 .129
Random effects
Factor Random effect Variance Std. Dev.
30
8.2. Reaction times
A log transformation was performed on the reaction time data to approximate a normal distribution. Subsequently, a five-point summary was computed to detect outliers in the distribution of the data. Reaction times more than one-and-a-half times the interquartile range above or below the median were labeled as outliers and removed from the data (5.3%). Table 4 presents the average reaction times per condition for the remaining data.
Table 4. Mean reaction time (sd) per condition in milliseconds
Constraint
Idiomaticity Condition High Neutral
Idiom correct 577 (88) 566 (58) incorrect 674 (91) 691 (79) pseudo 677 (87) 678 (82) Lit correct 599 (95) 610 (78) incorrect 713 (84) 672 (69) pseudo 689 (105) 694 (96)
Figure 3. Mean proportions of accuracy scores per
31
The reaction time data of the lexical decision task were also analyzed based on a linear mixed effect model analysis carried out in R (R Core Team, 2015) using the package lmerTest (Kuznetsova et al., 2016). Table 5 gives a summary of the best fit model’s coefficients with concomitant p-values. The model with the best fit to the reaction time data includes Idiomaticity and Condition as predictors of the log transformed reaction times in the various conditions of the lexical decision task, as well as random intercepts for Subject. Figure 4 shows that reaction times are slower for critical words embedded in a literal sentence (mean RT = 662) compared to an idiomatic sentence (mean RT = 646).
Table 5. Summary of the best fit model predicting reaction times in the lexical decision task. Fixed effects
Factor Estimate Std. Error t-value p-value
Intercept 6.3382 0.0243 260.68 < .001
Idiomaticity (lit) 0.0273 0.0099 2.76 .006
Condition (incorrect) 0.1594 0.0138 11.55 < .001
Condition (pseudo) 0.1516 0.0120 12.59 < .001
Random effects
Factor Random effect Variance Std. Dev.
Subject Intercept 0.0076 0.0870
Residual 0.0353 0.1879
Figure 4. Mean reaction times per sentence type in
32
As the factor Condition consisted of three levels, Tukey contrasts were calculated using the package multcomp (Hothorn, Bretz, & Westfall, 2008). The results show that subjects were faster at responding to a correct critical word (mean RT = 587) than at a semantically incorrect critical word (mean RT = 684). Furthermore, reaction times in response to correct critical words are also significantly faster than reaction times in response to pseudo words (mean RT = 674). Subjects do not respond differently to semantically incorrect critical words compared to pseudo words. The response-time distributions for the three levels of the factor Condition are shown in Figure 5.
The data of the lexical decision task thus show that the context sentences do not differentially influence literal word activation in literal and idiomatic sentence processing when this is measured behaviorally. However, although participants did not have more difficulty with processing idioms compared to literal expressions or vice versa, the faster reaction times to idiomatic in comparison to literal sentences suggest that people quickly activate and access an idiom’s non-literal meaning.
9. Discussion lexical decision experiment
In the lexical decision experiment, the lexical decision latencies served as a behavioral measure of the effect of the context sentences that were added to the materials of Rommers et al. (2013). The hypothesis was that a strongly constraining context would promote the pre-activation of an idiom’s non-literal meaning, thereby facilitating the processing of the idiom. Regarding the lexical
Figure 5. Mean reaction times per type of critical
33
decision task, the expectation following from the hypothesis was that reaction times for lexical decisions in idiomatic sentences would be faster compared to reaction times in literal sentences. More importantly, a preceding strongly constraining context sentence was expected to speed up lexical decision latencies in idiomatic sentences even more compared to idiomatic sentences preceded by a neutral context sentence.
The experiment did not show an effect of context on participants’ reaction time, although an effect was expected. The null effect for contextual constraint suggests that the degree to which younger adults predicted upcoming words was not influenced by either the preceding strongly constraining or neutral context sentence. This could indicate that the information provided in the high constraint context sentence was not informative enough to pre-activate the words in the following test sentence and modulate semantic processing at a behavioral level by speeding up reaction times for lexical decision on critical words. However, it is also conceivable that the sentences taken from the research of Rommers et al. (2013) alone, which in the present research were the test sentences, already provided the reader with sufficient contextual information, thereby making the information in the preceding context sentence redundant. A third possibility is that participants did not attentively read the context sentence. In contrast to the EEG experiment, in which the control questions required the participants to pay attention to the context sentence, the information provided by the context sentence in the lexical decision experiment was not strictly necessary to successfully perform lexical decision on the critical word in the test sentence, as no control questions regarding the content of the context sentences were asked.
Besides the fact that according to the hypothesis an effect of context was expected in the reaction time data, the reaction times in the lexical decision experiment were expected to be shorter for the idiomatic as compared to the literal conditions, consistent with Rommers et al. (2013). The results met this expectation, providing support for models arguing for a quicker access of meaning for idioms compared to literal expressions. However, the faster reaction times for idioms alone cannot distinguish between models stating that literal word meanings are not accessed at all, such as the direct access model (Gibbs, 1994), and models that assume activation of literal word meanings in idioms, but only in early stages of processing, before the idiom has been recognized as such (e.g., the configuration hypothesis, Cacciari & Tabossi, 1988).
34
The next section will discuss the methodology and results of the EEG experiment. This experiment will need to show whether the contextual manipulation influences the online processing of idioms differently compared to the online processing of literal expressions, even though the effect of contextual constraint was not found in the offline lexical decision task. More importantly, the central question investigated in the EEG experiment is whether context will have a different influence on the online processing of idioms by younger compared to elderly adults.
10. EEG experiment 10.1. Methods
10.1.1. Participants
For the EEG experiment, twenty-eight healthy elderly participants were recruited from the Senioren Academie Groningen. Recruitment was done here to ensure a high level of education in the elderly group, which was important for a valid comparison to a younger control group consisting of students. All the elderly adults were right-handed, native, monolingual speakers of Dutch with no history of language or neurological disorders. For their participation in the experiment, the elderly adults received a small gift. Twenty-eight students of the University of Groningen and the Hanze University of Applied Sciences Groningen served as a young control group. The students received a gift voucher in return for their participation in the study. An overview of each groups’ characteristics can be found in Table 6.
Table 6. Characteristics of the participants in the EEG experiment.
Measure Young Old
n 25 25
Age 22;6 (18;2 – 28;0) 68;3 (61;9 – 74;3)
Sex m: 6 m: 15
Years of Education 17;10 16;3
Hours of reading per week 10 - 20 10 - 20
