Wednesday, June 9, 2010

The concreteness effect : why is it much more easy to understand what a "chair" is than "freedom" ?

Well, I would say : "bad start : checked. Pure procrastinating waste of time : checked. Blog still unnoticed : checked".
But, in order to be forgiven, I'll give you an extract of a paper I wrote in the frame of a course about "categorization and induction". This may be a little bit long for a blog note, but still, I think that it's worth it, if it can help somebody...

As I told you in my presentation article, I have a deep interest in the semantic processes. This news is no exception and will develop what the concreteness effect is, in semantic and conceptual prossessing. I dropped a big part of my discussion and a proposal I made (you don't really need this...). Ready ? Let's start !

Introduction to the concreteness effect
There has been a wide number of researches about the concrete-word advantage, or concreteness effect, in the psycholinguistic literature. This effect refers to the idea that people perform better when confronted to a task presenting concrete words and/or sentences (i.e., highly imageable words or sentences) than when presenting abstract words/sentences. Evidences has been reported in lexical decision tasks (e.g., James, 1975; Schwanenflugel & Shoben, 1983; Tolentino & Tokowicz, 2009), recall (e.g., Paivio, 1971), word naming (e.g., Bleasdale, 1987), sentence verification (e.g., Holmes & Langford, 1976), sentence reading (e.g., Schwanenfluegel & Shoben, 1983) and bilingual translation (e.g., De Groot, 1992; Van Hell & De Groot, 1998).
However, while a great number of studies observed different results between the concrete and abstract words presentation, various others did not find any evidence in favor of the advantage of the concrete words over the abstract ones (e.g., Feldman, Pastizzo & Basnight-Brown, 2006; Samson & Pillon, 2004; Tokowiz & Kroll, 2007; Van Hell & De Groot, 1998, 2008). Some researches even showed a abstract advantage over concrete words (e.g., Macoir, 2009; Papagno, Capasso and Miceli, 2009; Tokowicz & Kroll, 2007; Van Hell et al., 1998).
In the following sections, I will first describe various fields of investigations debating the existence or not of the concreteness effect. Second, I will develop the different models that have been proposed to explain the advantage or, sometimes, the disadvantage of the concrete words compared to the abstract ones. Third, the abstractness question will be briefly discussed, according to an embodying cognition point of view. And finally, I will summarize these accounts and address a few propositions for future researches.

Behavioral evidences
Literature of the concreteness effect showed in many experiments that participants tend to respond faster and process with better accuracy the imageable words in word naming and lexical decision tasks (e.g., Binder, Westbury, McKiernan, Possing and Medler, 2005; De Groot, 1989; Paivio, 1991). These results will be further developed in the light of neuroimaging and neuropsychological experiments.

Neuroimaging evidences
Electrophysiology
Despite its low spatial resolution, ERP provides a good way to investigate the temporal evolution of the semantic processes. In that perspective, number of electrophysiological studies used the N400, "a negative deflection peaking about 400 ms after the stimulus onset which reflects both context-dependent and context-independent semantic factors" (Kounios, Green, Payne, Fleck, Grondin, and McRae, 2009, p. 96).
Thus, in the frame of the concreteness effect, Kounios, Holcomb and colleagues (1994, 1998) observed a greater difference in the activation of the N400 during the presentation of concrete words compared to abstract ones while making a semantic categorization task or a lexical decision task (Kounios and Holcomb, 1994) and during a congruency judgment task (Holcomb, Kounios, Anderson and West, 1998). But if the concrete words are easier to process, the reason of this greater activation needs to be clarified. According to these authors, concrete items would activate more features than the abstract ones, thus creating a larger amplitude of the N400 due to a bigger effort. Nevertheless, this still does not explain completely the behavioral results.
In a study of Tolentino et al. (2009), the order of presentation of the abstract and concrete words was investigated. In the abstract-concrete order, responses to the concrete words were more negative, while in the concrete-abstract order, ERP responses did not differ.
Finally, in a recent study of Huang, Lee and Federmeier (2010), the LH has been found to process the concrete and abstract words in a qualitatively similar way while the RH elicits a sustained frontal activity when presenting concrete words (which has been related to mental imagery).


fMRI
There exists a important number of neurobiological theories of the semantic memory that postulate that our conceptual knowledge of concrete words (and, sometimes, to a lower degree, also the abstract ones) are grounded in the sensory and motor systems of our brain (e.g., Barsalou, 1999; Prinz, 2002; Damasio, Tranel, Grabowski, Adolphs, Damasio, 2004; Paivio, 1971; ...). This consideration is of major importance because some models attribute the concreteness effect to the imageable features of the concrete concepts. This means that the conceptual processing of this type of concepts would also involve the sensory and motor systems.
Thus, many functional imaging studies examined the brain activity during conceptual processing. For example, Martin (2007) signals that the visual areas are activated when the subjects execute conceptual tasks about animals and that the motor areas are activated during the conceptual processing of tools. Hauk, Johnsrude and Pulvermüller (2004) also report an activation of the premotor and motor areas during a passive reading of face, arms or legs movements. The gustatory areas are activated during the conceptual processing of food (Simmons, Martin & Barsalou, 2005), as well as the olfactive areas when thinking to odors (Gonzalez, Barros-Loscertales, Pulvermüller, Meseguer, Sanjuan, Belloch & Avila, 2006).
Nevertheless, despite the tremendous number of evidence that the sensorial and motor areas are activated during the conceptual processing of words, little fMRI evidence has been found in the concrete vs. abstract distinction. Several studies did not show any activation differences between these types of concepts (e.g., Grossman, Koenig, De Vita, Glosser, Alsop, Detre and Gee, 2002; Noppeney and Price, 2004) with some finding no activation in the ventral temporal lobe where a difference would have been expected due to the visual specialization of this area and its reciprocal connections with other sensory association areas (e.g., Jessen, Heun, Erg, Granath, Klose, Papassotiropoulos and Grodd, 2000, even if they found a greater activation in the parietal lobes when encoding concrete concepts). However, some studies did still find a greater temporal lobe involvement for the processing of concrete words (e.g., Fiebach and Friederici, 2003; Wise, Howard, Mummery, Fletcher, Leff, Büchel and Scott, 2000; Sabsevitz, Medler, Seideberg and Binder, 2005). For example, Sabsevitz et al. (2005) found an greater activation in a bilateral network of multimodal association areas for concrete words while abstract concepts activated almost exclusively the LH. More generally, Jefferies et al. (2009) performed a meta-analysis of 12 neuroimaging studies which, taken together, showed a greater activity of the occipital, posterior infero- and medial anterior temporal areas for concrete words (these regions are known to be related to visual object recognition, mental imagery and picture-based semantic tasks). However, abstracts words tended to activate more the left superior temporal lobe (which has been related to speech comprehension and intelligibility).
Finally, concrete words have been shown to enhance the activation of the hippocampal structure (Fiebach et al., 2003), a neural circuit known for its involvement in the recollection of memories. This lead Peter and Daum (2008) to discover a greater difficulty to recollect memories of a list of concrete words (learnt in during the first phase of the experiment) than abstract words in older adult population. However, the feeling of familiarity of the presented words did not change between the ages.

Neuropsychological evidences
Brain damaged population
The concreteness effect has also been observed in brain-damaged patients suffering from aphasia and deep dyslexia. Indeed, these subjects made more errors for abstract than for concrete items (e.g., Coltheart, 1980, cited in Jefferies, Patterson, Jones & Ralph, 2009 ; Jefferies et al., 2007). More specifically, as reviewed in Papagno, Capasso and Micelli (2009), aphasic patients had a better performance on concrete over abstract words in "spontaneous speech (Howes & Geschwind, 1964), reading (Coltheart, Patterson and Marshall, 1980), writing (e.g., Bub & Kertesz, 1982), repetition (e.g., Martin & Saffran, 1992), naming (e.g., Frankin, Howard, & Patterson, 1995) and comprehension (e.g., Franklin, Howard, & Patterson, 1994)" (Papagno et al., 2009, p. 1138).

A particular case : the semantic dementia
While the concreteness effect has been reported in the majority of the studies on healthy and brain-damaged subjects, a particular phenomenon has been observed in patients suffering from semantic dementia (SD) : a reverse concreteness effect. In other words, these patients performed better during the processing of abstracts than concrete words, indicating a intriguing preservation of the abstract knowledge (e.g., Yi, Moore, & Grossman, 2007; Macoir, 2009). Neary and colleagues (1998) describe SD as a neurodegenerative disease characterized by a :
severe naming and word comprehension impairment [...] in the context of fluent, effortless, and grammatical speech output; [with a] relative preservation of repetition; and the ability to read aloud and write orthographically regular words. Also there is an inability to recognize the meaning of visual percepts (associative agnosia). This loss of meaning for both verbal and nonverbal concepts (semantics) contrasts with the preservation of visuospatial skills and day-to-day memory (Neary, Snowden, Gustavson, et al., 1998, pp. 1546-1547).

A recent study (Vesely, Bonner, Reilly and Grossman, 2007) also showed a greater degradation of the concrete words during a free-association task compared to abstract ones. Nevertheless, while it would seem pertinent to consider the reverse imageability effect as a new indicator of semantic dementia, this effect seems, up to present, more restricted to some particular cases. For instance, Jefferies et al. (2009) conducted a experiment of synonym judgments on 11 SD subjects without observing any advantage of the abstract items. These authors attribute then this reversal effect as the result of an unusual distribution of atrophy with a spared superior region of anterior temporal lobe and a greater damage in the posterior and the inferior temporal cortex. It also seems important to signal that, even if the concrete knowledge seems more degraded than the abstract one, in some studies, both types of words are significantly touched in SD (e.g., Vesely et al., 2007). These differences may be due to the evolution's stage of the disease. Indeed, to our knowledge, only one research conducted a longitudinal study on a SD patient (Macoir, 2009) and observed that the distinction concrete/abstract was noticeable only at an early stage of the disease, before gradually disappearing.

Explanatory models
Dual coding theory
One of the most famous accounts of the concreteness effect comes from the Dual Coding Hypothesis of Paivio (1971). This model postulates two functionally distinct representational systems : a imagery system for the processing of the non-verbal information carried by the words, and a verbal system for the processing of the linguistic informations.
Thus, according to this hypothesis, concrete words would be mainly represented in the right hemisphere (RH) of the brain, while the abstract words would be represented only in the left hemisphere (LH). ERP results from several studies I previously described would support such a theory in regards to the greater activation of the RH in favor of concrete items during various semantic tasks (Holcomb et al. 1998; Huang et al., 2010; Kounios et al., 1994). The fMRI experiment conducted by Binder, Westbury, McKiernan, Possing and Medler (2005) also support this account by observing a greater activation of the LH while treating abstract words and a bilateral activation during the processing of concrete words. However, a lot of experimental results do not go in the same direction as the Dual Coding theory. For example, Jessen et al. (2000) found a greater activation of the occipital areas (involved in visual processing) while processing abstract concepts than for concretes ones. Also, Fiebach et al. (2003) did not find any processing advantage of the RH for concrete words but reported changes only in the LH with, still, a greater activation for the abstract concepts.
Whether the dual coding theory is right or not about the precise mechanisms underlying the processing of the abstract and concrete words, it has the interest to propose a model where concepts may also be represented by perceptual features. Indeed, by considering the reverse concreteness effect observed in some SD patients, the particular degradation of concrete words may be explained, according to some authors (e.g., Yi et al., 2007), by a loss of visuoperceptual semantic features (partly represented in the ventral temporal lobe, a neural area prone to deterioration due to SD). Thus, a embodied account of the semantic organization might be of particular interest in the light of the (reverse) concreteness effect.

Network accounts
For a long period, models of the semantic organization proposed a localist representation of the concepts whereby the meaning of a word was represented by a single node in a network (Collins and Quillian, 1969). Thereafter, distributed representation models have been proposed in which a concept's meaning is represented over multiple nodes in a network, each of these being activated in a various degree (e.g., Plaut, 1995). Some authors explored then the "semantic richness" of the words as a variable influencing the processing of the concepts (e.g., Kounios, et al., 2009; Grondin, Lupker and McRae, 2009).

Semantic richness and number-of-features
Semantic richness is function of two central properties of word meaning : ambiguity and concreteness. While concreteness has already been much described in this article, it has been proposed that ambiguous words have a richer semantic representation than non ambiguous words because of their multiple meanings leading to a speed advantage of processing during lexical decision and naming (e.g., Hino and Lupker, 1996, cited in Grondin et al., 2009). However, this effect disappears in a categorization task (Hino, Pexaman and Lupker, 2006, cited in Grondin et al., 2009). But, more importantly, semantic richness depends mainly on the number of features (NoF) a concept has (Grondin et al., 2009). So, the higher the NoF is, the better the performance of the subjects (while performing a paced-reading task, as in Pexman, Holyk and Monfils, 2003, cited in Grondin et al., 2009). Yet, the "ease of predication" hypothesis (e.g., Jones, 2002) postulates that concrete concepts have a higher NoF than the abstract ones leading to better overall performance in semantic processing.

Context availability hypothesis
Another explanation for the concreteness effect may be described in terms of "context availability" of the informations needed to process the concepts (Schwanenflugel et al., 1983). According to this theory, abstract concepts are not processed as good as the concrete ones because they usually do not have as much contextual information available to be processed as the concrete ones do. The concrete words would indeed have been more familiar and therefore associated with more propositions in long-term memory. This account has received support from several studies showing that, when presented with as much context information, the abstract concepts are processed as good as the concrete ones in a lexical decision task (e.g., Schwanenflugel et al., 1983; Van Hell et al., 1998).
However, the experiment of Tolentino et al. (2009) is, for example, not entirely consistent with the context availability hypothesis that postulates quantitative differences in a single system because of the interhemispheric differences of activation they reported (see also the dual coding hypothesis arguments).

Other models
Up to present, the models that have been summarized in the previous sections seem to be the major ones. Nevertheless, this should not obliterate other pertinent accounts such as the Sensory/Functional Theory (SFT) (e.g., Warrington & Shallice, 1984) where living things would be much represented by sensory/perceptual attributes, whilst non-living things would be more represented by functional attributes. Another possible account is the Hierarchical Interactive Theory (HIT) of Humphreys and Forde (2001, cited in Macoir, 2009), similar to the SFT but "adopting an amodal theory of semantic memory with separate visual (pictorial and written words) and auditory (environmental sounds and spoken words) inputs" (Macoir, 2009, p. 529).

The abstract concepts question
It has been mainly argued, throughout this article, that abstract concepts do not rely on perceptual features. However, according to Barsalou and colleagues (Barsalou, 1999; Barsalou and Wiemer-Hastings, 2005), abstract concepts may not be represented only by amodal symbols. For example, Barsalou (1999, 2008) proposes that, when a abstract concept has been perceived, subjects create a representation of it by applying three mechanisms : framing, introspection and selectivity. First, the individual will identify a sequence of events indicating the frame in which the abstract concept has been encountered. Then, he will focus on the physical informations of the event as well on the introspective and proprioceptive informations. Finally, the individual will identify the central element of this concept amongst the amount of features that characterizes it. Barsalou (1999) asserts that, once the concept has been perfectly integrated, it will be subsequently simulated directly by the sensori-motor systems.
However, even if this account allows some perceptual simulation of the abstract concepts, Dove (2009) signals that the introspection mechanism will, in turn, evoke other abstract concepts that will also be subjects to a processing as heavy as the first one. Because introspection would be a perceptual simulation associated with judgments formation, these judgments are also likely to contain abstract concepts. Thus, amodal mechanisms would be necessary to short-cut this endless loop.

Discussion
It would be difficult to assert, considering the amount of evidences I described, that there is no such thing as a concreteness effect. Behavioral, electrophysiological, neuroimaging studies and brain-damaged patients present a great source of information regarding the semantic network organization.
However, in spite of all these evidences of different performance and/or neural activity between abstract and concrete concepts, researches showed scattered results. It is then difficult to determine the reason of these different patterns. It is effectively plausible that concrete concepts activate more features (modal and amodal) than abstract concepts. Therefore, the reasoning of dual coding theory, that asserts that concrete concepts will also activate the RH, may sound well-founded due to the imagery specialization of the RH. However, as Kounios et al. (1994) remind it, an ERP activation observed in the RH may in fact originate from the LH. Moreover, embodied cognition theories would argue that concrete (and sometimes abstract) concepts do not only activate visual properties, but also simulate other sensori-motor and introspective representations (e.g., Ghiot and Tettamanti, 2010). It is our opinion that all these different patterns of results come from design differences leading then to different cognitives strategies. As Solomon and Barsalou (2004) showed it, when subjects are asked to perform a superficial semantic task, they may use only a word association strategy (bypassing then the perceptual simulation). More specifically, when performing a property verification task, subjects had to access conceptual knowledge when presenting an associated false property of the initial word (e.g., owl - tree).

Conclusions
From all these researches, it currently seems that the perceptual and the great number of features of the concrete words compared to the abstract ones seem to be the major parameters involved in the (reverse) concreteness effect. However, while network models are able to simulate quite precisely this effect, one drawback of the conceptually embodied models is still the (currently) non-existing computational implementation.
Also, it would be interesting to clarify the influence of some parameters such as the age of acquisition of the words and the level of education of the participants.
Finally, it would be pertinent to investigate to what extend the phenomenon I describe throughout this article would be useful for the improvement of diagnoses, and especially of rehabilitation strategies for brain-damaged people.



For the exact references, just ask me and I'll send them to you.

No comments:

Post a Comment