Be concrete to be comprehended: Consistent imageability effects in semantic dementia for nouns, verbs, synonyms and associates

Abstract

There are two contrasting views on the nature of comprehension impairment in semantic dementia: (a) that it stems from degradation of a pan-modal “hub” that represents core conceptual knowledge or (b) that it results from degradation of modality-specific visual feature knowledge. These theories make divergent predictions regarding comprehension of concrete versus abstract words in the disorder. The visual hypothesis predicts that concrete words should be particularly impaired because they depend heavily on visual information. In contrast, the pan-modal hub hypothesis holds that all types of knowledge are affected but predicts less severe impairment of concrete words because they have richer and more detailed semantic representations than abstract words. We investigated concreteness effects in the comprehension of six SD patients. Across nouns, verbs, synonymous and associative relationships, a clear and consistent pattern emerged: concrete words were always comprehended more successfully than abstract words. These findings extend those of previous studies and suggest that conceptual impairment in SD is not confined to concepts that rely on visual information. Instead, all types of knowledge are affected by the progressive deterioration of modality-invariant representations (required for coherent pan-modal concepts). Concrete words succumb less quickly by virtue of their richer and more detailed semantic representations.

Introduction

This study is concerned with the nature of the semantic impairment in semantic dementia (SD), a neurodegenerative condition in which atrophy centred on the inferolateral, anterior temporal lobe accompanies a progressive and eventually profound breakdown in semantic knowledge (Snowden et al., 1989; Hodges and Patterson, 2007). Because SD patients present with a central and highly selective semantic deficit, the disorder is an important source of evidence for theories of semantic knowledge (Rogers et al., 2004; Patterson et al., 2007; Jefferies and Lambon Ralph, 2006; Simmons and Martin, 2009). A key issue in interpreting the nature of the disorder is to determine whether any particular domains of knowledge are disproportionately impaired. Though SD patients have an indisputable deficit in understanding concrete concepts (e.g., manmade objects and living things; Luzzi et al., 2007; Bozeat et al., 2000, 2002; Garrard and Carroll, 2006), the status of abstract word knowledge (e.g., advantage or chance) is more contentious. Comprehension deficits following brain damage typically affect abstract words to a greater extent than concrete words (Coltheart, 1980; Hoffman et al., 2010, 2011; Katz and Goodglass, 1990; Franklin, 1989), in line with more efficient processing of concrete words in healthy subjects (Kroll and Merves, 1986; James, 1975; Degroot, 1989). However, there are a number of reports of SD patients who show a reversal of this standard effect, i.e., better comprehension of abstract relative to concrete words (Cipolotti and Warrington, 1995; Warrington, 1975; Macoir, 2009; Breedin et al., 1994; Papagno et al., 2009; Reilly et al., 2007). Throughout the paper, we refer to this as an A > C pattern of comprehension.

These unusual “reversals” of the typical concreteness effect have been interpreted in two ways. Some researchers take them to be an important signature of the underlying semantic impairment in SD. On this view, SD patients suffer from damage to the visual association cortex in the ventral temporal lobes, which is particularly important for the representation of concrete words (Yi et al., 2007; Breedin et al., 1994; Macoir, 2009; Bonner et al., 2009). Because the meanings of abstract words depend strongly on verbal associations and are less dependent on visual-perceptual information (Paivio, 1986), their semantic representations are thought to be relatively spared. The semantic representations of abstract words might depend more heavily on superior temporal regions specialised for verbal comprehension, which are less severely affected in SD (Galton et al., 2001).

Other researchers maintain that A > C effects are not informative about the nature of semantic impairment in SD because they only occur in a small subset of SD cases (Jefferies et al., 2009; Hoffman and Lambon Ralph, 2011). Most reports of A > C effects are from single-case studies and until recently there had been no attempt to determine whether these cases were representative of the typical pattern of impairment in the disorder. The inferolateral aspects of the anterior temporal lobes are the focus of atrophy and hypometabolism in SD patients (Galton et al., 2001; Nestor et al., 2006). However, repetitive transcranial magnetic stimulation to this region in healthy subjects produces a greater (temporary) impairment for abstract over concrete words (i.e., a C > A pattern; Pobric et al., 2009). This runs counter to the idea that anterior temporal lobe atrophy gives rise to an A > C pattern of comprehension. Instead, these findings predict that SD patients should typically exhibit C > A effects, a view that is more in keeping with the “hub-and-spoke” semantic framework (Rogers et al., 2004; Pobric et al., 2010; Patterson et al., 2007). This theory states that semantic representations are underpinned by (a) a network of modality-specific regions of cortex distributed throughout the brain (“spokes”) and (b) a pan-modal “hub” that merges input from these regions to form modality-invariant, coherent conceptual representations (Lambon Ralph et al., 2010). Semantic impairment in SD is thought to arise from damage to the hub, affecting all types of semantic knowledge irrespective of whether they depend on visual or verbal information. On this view, abstract words are disproportionately impaired simply because they have intrinsically weaker semantic representations. Their weaker representation comes about because they are not associated with the rich perceptual experiences that characterise concrete words (Paivio, 1986; Plaut and Shallice, 1993; Jones, 1985). On this view, occasional reports of individual patients showing A > C effects represent a deviation from the typical pattern of semantic breakdown in SD.

A number of recent studies have attempted to adjudicate on this issue by assessing concrete and abstract knowledge in larger groups of SD patients. In support of the visual deficit hypothesis, Yi et al. (2007) tested 12 SD patients with a description-to-word matching task and found an A > C pattern for verbs, with nine patients showing an effect in this direction. However, there was no such effect for comprehension of nouns. In a follow-up study, Bonner et al. (2009) tested 11 patients with a synonym matching task with verbs and found a similar pattern at the group level. In contrast, Jefferies et al. (2009) gave a synonym judgement task consisting mostly of nouns to 11 SD patients and found a strong C > A effect, which was present at a statistically significant level in all 11 individuals. The differences between these studies could have arisen from differences in the patients themselves or variation in the experimental stimuli used to assess knowledge. We recently tested a single set of seven SD patients on all of the tasks from the above studies, along with some other concrete–abstract assessments (Hoffman and Lambon Ralph, 2011). Averaged across all tasks, there was a significant C > A effect, supporting the hub-and-spoke interpretation. There were also some informative differences between tasks. The Jefferies et al. task yielded a robust C > A effect, while no difference between concrete and abstract words emerged for the Yi et al. and Bonner et al. verb tasks. We attributed this variability to differences in the stimuli used in the various tasks. Tasks were more likely to reveal a C > A effect if they: (a) ensured that concrete and abstract words employed in the task were fully separated along the imageability scale and (b) ensured that stimuli were matched across conditions for word frequency. The two verb tasks did not meet these criteria. The selected concrete and abstract words were quite similar in imageability, resulting in lower sensitivity to concreteness effects than the Jefferies et al. task. Moreover, the abstract conditions featured higher frequency words than the concrete conditions. Lexical frequency has a powerful influence in SD, with patients invariably showing better comprehension of higher frequency/more familiar words (Hoffman et al., 2011; Jefferies et al., 2009; Lambon Ralph et al., 1998; Funnell, 1995). Consequently, these two methodological factors may have conspired to produce an apparent A > C effect for the Yi et al. and Bonner et al. stimuli.

We proposed that these differences in frequency and imageability (summarised in Table 1) can account for the inconsistent results across studies. However, another possibility is that the word class of the stimuli is a critical factor in determining the type of concreteness effects observed. Jefferies et al. (2009) employed mostly nouns in their test, while Bonner et al. (2009) used only verbs. In addition, Yi et al. (2007) found an A > C effect for verbs but no effect for nouns. Yi et al. claimed that concrete nouns are partially protected from semantic degradation because they belong to dense semantic neighbourhoods. Because concrete nouns tend to have many semantic neighbours with similar features, when a particular concept degrades it can be partially supported by remaining knowledge for its neighbours (Gonnerman et al., 1997). The same was not thought to be true of concrete verbs, which have fewer neighbours. Thus, while visual feature degradation predicts greater impairment for concrete words, this effect tends not be observed with nouns because concrete nouns are protected by virtue of their dense neighbourhoods. This is a critical point because it calls into question into the C > A effects reported by Jefferies et al. (2009) and Hoffman and Lambon Ralph (2011). These authors reported robust C > A effects in SD patients, but used nouns predominantly as stimuli. The visual feature hypothesis predicts that an A > C effect would emerge if the same patients were tested with verb stimuli.

In the present study, we directly compared concreteness effects for nouns and verbs in six SD patients, including five individuals who showed C > A effects for nouns in our previous study (Hoffman and Lambon Ralph, 2011). We paid particular attention to the key stimulus factors described above. We maximised the difference in imageability values between concrete and abstract words and we ensured that all of the words in each condition were carefully matched for lexical frequency. This allowed us to test whether A > C effects were present for verbs, as predicted by the visual deficit hypothesis, once the relevant stimulus factors were controlled for.

In a second experiment, we investigated another factor thought to influence concreteness effects. The Jefferies et al. test, in common with many other comprehension assessments, probed knowledge of synonymous relationships between words (e.g., frog and toad). This type of semantic relationship may not be well-suited to probing abstract word knowledge. While the semantic representations of concrete words are thought to be organised according to similarity in features, associative relationships (e.g., religion and prayer) are more important for abstract words (Crutch and Warrington, 2005; Crutch et al., 2009). Thus C > A effects may arise in synonym judgement tests because the format of the test is more compatible with the organisation of concrete concepts. This theory predicts that an A > C pattern could be observed in SD patients if their knowledge of associative relationships was probed. To test this prediction, we compared concrete and abstract words in an associative matching test. If the C > A effects in previous studies were due to the use of synonym judgements, one would expect a different pattern to emerge with this paradigm. We also included a second concrete condition in which probe and target shared perceptual similarity, to compare directly associative and similarity-based judgements for concrete words.