Semantic priming for coordinate distant concepts in Alzheimer's disease patients

Abstract

Semantic priming paradigms have been used to investigate semantic knowledge in patients with Alzheimer's disease (AD). While priming effects produced by prime–target pairs with associative relatedness reflect processes at both lexical and semantic levels, priming effects produced by words that are semantically related but not associated should reflect only semantic activation processes. This study was aimed at further investigating automatic semantic priming effects in AD patients when semantically related concepts with little to no lexical association are used. Twenty patients with mild to moderate AD and 20 matched controls (NCs) performed a lexical decision task on 30 concept pairs (15 in the living and 15 in the non-living domain) in an automatic semantic priming paradigm. In order to investigate the relationship between priming alteration and semantic damage, we chose concepts from a database. This allowed us to quantify semantic indexes relative to the structural representation at the feature level.

No priming was found in NCs or mild AD patients, probably because feature similarity was insufficient in the concept pairs used. Similar to the hyperpriming observed in previous studies, the appearance of priming in the moderate AD group suggests early semantic damage in which attribute knowledge is partially affected. Furthermore, the finding that priming was predicted by the level of sharing (in the semantic system) of features common to the two concepts in the pairs indicates that the level of redundancy of attribute information is the main factor responsible for resiliency to neurological damage in AD.

Introduction

In patients with Alzheimer's disease (AD), difficulty in performing tasks that tap semantic knowledge (such as naming, verbal fluency or object recognition) occurs early and increases during the course of the illness. This suggests early and progressive impairment of semantic memory in these patients (Nebes, 1989, Salmon et al., 1999). It should be noted, however, that poor scores on these kinds of tasks do not necessarily imply a loss of semantic knowledge. In fact, besides more basic lexical and/or perceptual deficits, poor performance on these tests might also be caused by reduced attentional and/or executive resources, which are needed to implement effective retrieval strategies. One way to obviate the confounding effect of attentional and/or executive deficits in the assessment of semantic knowledge in patients with AD is to use priming paradigms that permit assessing the structural integrity of semantic memory automatically. This minimizes the need for the intervention of non-semantic cognitive processes.

The semantic priming effect is revealed by subjects’ response times (in word pronunciation or lexical decision tasks), which are faster to target words following related primes (e.g. chair–table) than to target words following unrelated primes (e.g. car–table) or neutral primes (e.g. XXX–table). Faster response times to related than unrelated or neutral prime–target pairs are generally believed to reflect the spreading of activation from the prime to concepts that share a semantic relationship with it (Collins & Loftus, 1975). According to Neely, 1977, Neely, 1991, the priming effect is automatic, that is, it occurs rapidly and is not affected by preparatory strategies or expectations when the interval between the presentation of the prime and the target, known as stimulus onset asynchrony (SOA), is brief (<400 ms) and the percentage of related prime–target pairs is low. Instead, intentional (i.e. strategy-based) processes may contribute to semantic priming when longer SOAs or a higher proportion of related pairs are used.

Contrasting results were reported in studies that used semantic priming paradigms to investigate the status of semantic knowledge in AD patients. Indeed, when the paradigms depended on automatic priming effects, some studies documented less than normal priming in individuals with AD (e.g. Silveri, Monteleone, Burani, & Tabossi, 1996), others reported equivalent priming in AD patients and normal controls (e.g. Ober et al., 1991, Perri et al., 2003), and yet others described increased priming (hyperpriming) in AD patients with respect to matched healthy individuals (e.g. Balota et al., 1999, Giffard et al., 2001). Procedural differences in the priming experiments used as well as heterogeneity of the dementia samples recruited might explain the discrepant results. In particular, the type of semantic relationship between the prime and the target, that is, superordinate, coordinate, or attribute, and the dementia severity of AD patients seem able to influence the priming effects observed and even to interact with them. In fact, while normal priming is usually observed in AD patients in the semantic relationship based on the superordinate category (Glosser et al., 1998, Ober et al., 1991, Rogers and Friedman, 2008), hypopriming is more frequently reported in experiments in which the target is an attribute of the prime (Giffard et al., 2002, Rogers and Friedman, 2008); and, finally, hypo- or hyperpriming is observed when the prime and the target belong to the same category (coordinate condition) (Giffard et al., 2001, Giffard et al., 2002, Glosser et al., 1998, Rogers and Friedman, 2008). Moreover, it has been suggested that the type of priming alteration in AD patients, that is, hypo- or hyperpriming, is related to severity of the semantic memory degradation and interacts with the kind of relationship existing between prime and target. Indeed, in a longitudinal study of 24 AD patients followed up for 18 months Giffard et al. (2002) showed that the pattern of the priming effect changed over the course of the disease. In particular, these authors observed that while the priming effect in the attribute condition declined linearly, in the coordinate condition it first increased abnormally and then decreased. The authors interpreted these data with reference to a connectionist model of semantic memory. In this kind of model, concepts are represented as a distributed and interconnected network of units, each representing a particular semantic attribute of the concepts. As each individual featural unit may be shared by different concepts (e.g. tiger and lion), the semantic similarity between two concepts is directly related to the number of features shared by the two concepts (e.g. has a tail) and inversely related to the number of distinctive features that pertain to only one of the two concepts (e.g. has stripes and has a mane). In this view, the priming effect occurring in the coordinate condition of a semantic priming paradigm should be related to the degree of featural overlap between the prime and the target words, with larger priming effects observed when the prime and the target concepts share many featural units and each is characterized by a few distinctive features and progressively less priming observed as the relative weight of shared and distinctive features is reversed. When interpreting the qualitative pattern of semantic impairment in AD patients, authors who adopt the connectionist models of semantic memory frequently make the further assumption of a different resiliency of featural units to pathology-induced degradation depending on a variety of characteristics of individual features. In particular, it has been claimed that features shared by many concepts should be those most resistant to the damage of AD pathology and that distinctive features specific for only one or very few concepts represent the most vulnerable part of semantic knowledge (Devlin, Gonnerman, Andersen, & Seidenberg, 1998). According to Giffard et al., 2001, Giffard et al., 2002, early in the course of AD the loss of distinctive features, such as “has stripes” or “has a mane”, of two similar concepts, such as tiger and lion, makes the concepts progressively more similar if not identical, because the residual representation is characterized only by shared features such as “has four legs”, “is wild”, “has fur”. This is why pairs consisting of coordinate concepts, such as tiger–lion, produce hyperpriming at this stage of semantic degradation, whereas pairs consisting of a concept and its main attribute, such as tiger–stripe, produce hypopriming. As semantic memory deteriorates further, affecting shared as well as distinctive features, the two concepts become less close and the priming effect decreases also in the coordinate condition.

One limit of Giffard et al., 2001, Giffard et al., 2002 straightforward interpretation of their data is that in their experiments they used strongly associated pairs in a lexical decision task. Indeed, although the priming effect elicited by related word pairs is generally referred to as semantic, the use of associated pairs (in which relatedness between prime and target words derives not only from a semantic base but also from the temporal contiguity and co-occurrence of words in speech or text) makes the observed priming will likely reflect processes operating at both lexical and semantic levels (Hutchison, 2003, Shelton and Martin, 1992). In line with this view are the results of studies by Glosser et al. (1998) and Rogers and Friedman (2008). These authors investigated automatic priming for coordinate but not associated word-pairs in groups of AD patients with disease severity comparable to that of the patients investigated by Giffard et al., 2001, Giffard et al., 2002. Reinforcing the suspicion that lexical phenomena may have contributed to the observation of normal as well as hyperpriming in Giffard et al., 2001, Giffard et al., 2002, both studies documented less than normal priming effects in AD patients. This finding is particularly suggestive considering that in Rogers and Friedman's study (2008) AD patients obtained a normal priming effect when tested on semantically related word-pairs that were also lexically associated.

Shenaut and Ober (1996) proposed an alternative interpretation of the hyperpriming phenomenon in patients with AD. They suggested that impaired attentional processes, rather than semantic system degradation, are responsible for the exaggerated priming effects in these patients. Indeed, these authors point out that in AD hyperpriming has been most frequently observed with paradigms encouraging intentional strategies of response, such as expectancy mechanisms and post-lexical semantic matching processes. Expectancy is a pre-lexical mechanism that drives the subject to anticipate possible related targets at the appearance of the prime, which facilitates the processing of expected targets and inhibits recognition of unrelated ones. Conversely, postlexical semantic matching processes may occur in lexical decision paradigms after the appearance of a target related to the prime. In fact, if subjects notice the semantic relationship between prime and target, they are biased toward a “word” response (no relationship exists if the target is a non-word). On the contrary, if subjects notice the absence of a semantic relationship between prime and target, they have to inhibit the non-word response. Thus, the lexical decision is favored when the target follows a related prime but is slowed down in the unrelated condition. Accordingly, the resulting priming is not due to the automatic pre-activation of the target by the prime, but to a matching process that intervenes after the appearance of the target word. According to Shenaut and Ober (1996), attention based priming paradigms prompt AD patients (as well as healthy subjects) to divide their attention among expectancy generation, semantic memory search, and lexical decision processes. However, these multiple cognitive operations overwhelm their limited attentional resources, especially on trials in which the target is unrelated to the prime, thus resulting in the amplification of the inhibitory component of the priming effect and producing hyperpriming.

The present study was aimed at further investigating automatic semantic priming in patients with mild to moderate AD. Our primary goal was to verify whether Giffard et al.’s observation of a hyperpriming phenomenon in AD patients for coordinate pairs (Giffard et al., 2002) would also be elicited when concept pairs were semantically related but did not have a strong lexical association. For this purpose, in our experiment we used word pairs related at a coordinate semantic level that had little or no lexical association.

A secondary goal of this study was to examine the relationship between semantic priming and semantic structure modification in AD. For this purpose, the prime–target pairs were formed by concepts selected from a database derived from a normative study on the semantic representation of 64 concepts in the living and non-living domains. In that study, normative data were drawn from the results of a feature-listing task administered to Italian-speaking normal young volunteers (Zannino, Perri, Pasqualetti, Caltagirone, & Carlesimo, 2006). This allowed quantifying various semantic indexes relative to the structural representation of the concepts at the featural level. On this basis, it was possible to quantify the level of semantic relatedness of each prime–target pair in terms of between-concept semantic distance (a measure based on the number of features shared and unshared by the two concepts) and to obtain a measure of a variety of semantic indexes believed to constitute markers of vulnerability or resiliency of semantic information to AD pathology. In fact, it has been hypothesized that in addition to the level of feature sharing, the resiliency of semantic knowledge depends strictly on the level of feature correlation (two features are correlated if the occurrence of one increases the probability that the other will occur in the semantic representation of the same concept). Highly correlated features should be very resistant to damage because of the collateral activation they receive from each other (Devlin et al., 1998, Moss et al., 1998). Also feature dominance (an index that refers to the frequency with which a particular feature is named by people in describing a given concept on a feature-listing task) has been proposed to represent a gradient of resilience against neurological damage (Zannino et al., 2006). In fact, features that are accessed more frequently (because they are more salient for the concept) may also be the ones that can still be activated in a pathological condition. Finally, particular attention has been paid to the distinction between living and non-living categories. In fact, it was thought that a different distribution of shared and distinctive features for living and non-living categories as well as a different distribution in categories of the feature correlation level were at the base of the different susceptibility of the two categories to neurological damage in AD (Devlin et al., 1998, Gonnerman et al., 1997, Moss et al., 1998, Tyler et al., 2000). In this regard, it was recently documented that in early AD patients semantic priming may differ depending on whether the words in the pairs belong to categories in living or non-living domains (Hernandez, Costa, Juncadella, Sebastian-Galles, & Renè, 2008).