Qualitative features of semantic fluency performance in mesial and lateral frontal patients

Abstract

Semantic verbal fluency is widely used in clinical and experimental studies. This task is highly sensitive to the presence of brain pathology and is frequently impaired after frontal lesions. Besides the total number of words generated, a qualitative analysis of their sequence can add valuable information about the impaired cognitive components. Thirty-four frontal patients and a group of matched controls were examined. Besides the number of words and subcategories retrieved by each group, we analysed two distinct aspects of the word sequence: the search strategy through a semantically organised store and the ability to switch from one subcategory to another. We checked whether the pattern of impairment changed according to the lesion site within the frontal lobe. Overall, patients produced fewer words than controls. However, only lateral frontal patients presented a reduced semantic relatedness between contiguously produced words and a specifically increased proportion of switches to different subcategories. The performance of lateral frontal patients was in line with the hypothesis of a search strategy impairment and cannot be attributed to a switching deficit. The performance of mesial frontal patients could be ascribed to a general deficit of activation.

Introduction

The semantic fluency task (Bousfield & Sedgewick, 1944; Gruenewald & Lockhead, 1980; Rosen & Engle, 1997) requires the subject to produce as many words as possible from a given category (e.g. “animals”) within a brief fixed time. Although semantic fluency is widely used in clinical and experimental studies, it is still debated how the produced words should be evaluated in order to optimise the information yielded by the task. The number of words produced per se is highly sensitive to the presence of brain pathology, and is useful for basic clinical purposes; however, the same measure does not give information about the specific cognitive steps of the task. For this, we should first consider how semantic fluency operates in normal subjects. Then we can attempt to identify which stage of the normal processing is impaired in a patient, also with reference to our knowledge of the functions of different parts of the brain. The following facts seem relevant for our analysis.

When requested to produce all the items of a category (e.g. “fruit”), healthy subjects tend to generate sequences of words from the same subcategory, e.g. “soft fruit”, “dried fruit”, etc. (Gruenewald & Lockhead, 1980; Wixted & Rohrer, 1994). After having retrieved typical and familiar items of one-subcategory subjects generally shift to another subcategory. We can presume that shifting to a different subcategory represents a more efficient strategy than continuing to retrieve less typical or familiar items from the former subcategory. Accordingly, a hypothetical sequence of produced fruit exemplars might be: apple–pear–orange–tangerine–lemon–apricot–peach–strawberry–blackberry–raspberry–chestnut–hazelnut–almond–peanut. In symbolic notation, each word represented by a letter of the alphabet standing for a specific subcategory, the sequence can be written as AABBBCCDDDEEEE. Summing up, normal behaviour seems based on two distinct aspects: knowledge of the semantic relations within a category as the basis for forming subcategories, and the tendency to abandon a subcategory at a given point of the probe, i.e. to “switch”.

Among neurological patients, cases of frontal damage are of special interest for this task. Frontal patients generate fewer words in a semantic fluency task (Baldo & Shimamura, 1998; Stuss et al., 1998, Troyer et al., 1998). From a recent meta-analysis, the sensitivity of semantic fluency to frontal lesions is comparable to that of letter fluency and is higher than that of the Wisconsin Card Sorting Test (Henry & Crawford, 2004), although different lesion sites, particularly the left temporal lobe, can also cause an impairment. Troyer, Moscovitch and Winocur (1997) suggested that semantic fluency implies an executive function devoted to strategic retrieval processes and monitoring (the “frontal lobe” component) while a different component would be responsible for automatic encoding and information retrieval (the “temporal lobe” component). Imaging data provide further evidence in favour of the crucial role of frontal areas in the neural network subserving semantic fluency (Frith, Friston, Liddle, & Frackowiak, 1991; Gurd et al., 2002, Warburton et al., 1996).

Given these findings, a consequent research task is to identify which of the cognitive functions normally working during semantic fluency can be ascribed to the frontal lobes, as their inactivation is presumably responsible for the impairment that frontal patients present in this task. The most likely candidates (Stuss et al., 1998, Stuss and Levine, 2002; Troyer et al., 1998) are: (i) the generation and implementation of an effective search strategy; (ii) the ability to switch between different subcategories; (iii) the ability to initiate a task and keep it activated; (iv) the ability to monitor the task (useful, for instance, for avoiding repetitions of already retrieved words). These aspects deserve a closer analysis.

Compliance with a strategy is essential to conduct an efficient search through a semantically organised store such as the semantic memory. The hallmark of a disorganised search through semantic memory is the production of a disorganised word sequence. An example of such a sequence for the category “fruit” may be: apple–orange–cherry–blackberry–pear–tangerine–banana–raspberry–lemon–apricot; in symbolic notation, the sequence is ABCCABDCBE. The same effect on sequence organization could depend on a semantic memory damage able to scramble the common associative and categorical links among the items; however, the literature does not report reliable cases of primary semantic deficits due to frontal dysfunction. Therefore, we should presume that, although the semantic store per se is spared in frontal patients (Sylvester & Shimamura, 2002), it cannot be acted upon by a suitable search strategy during the word retrieval process: we term this “loss of strategy”. In this case, the sequence of produced words would be no longer grouped in subcategorical clusters. As a consequence, we should observe a lower average relatedness between successive words and a corresponding higher frequency of relative switching between different subcategories.

A further type of deficit might affect the ability to abandon an old subcategory for a new one, i.e. to switch after the production of the main exemplars of a given subcategory in order to avoid excessive slowing down of the retrieval speed. An example of a sequence that may be produced by a patient suffering from switching deficit is: orange–tangerine–lemon–grapefruit–clementine–strawberry–blackberry–gooseberry–blueberry–raspberry, in symbolic notation AAAAABBBBB. In the case of an inability to abandon a perused subcategory, defined here as “loss of switching”, we would expect a lower switch rate and a higher semantic relatedness between successive words. Moreover, this class of patients is also expected to explore less subcategories than healthy participants.

Frontal patients could also suffer from an initiation and activation deficit and be simply slower while retrieving each item of the sequence. The sequence produced in this case would be similar to the normal one but the number of words would be lower due to the time constraint. Also the number of subcategories and absolute number of switches would be less, while the relative number of switches and average semantic relatedness between successive words would remain unaffected. In symbolic notation, a performance of this type might be AABBBCC (compare with the sequence for healthy controls: AABBBCCDDDEEEE).

Finally, a failed monitoring of already retrieved words or already used cues would increase the repetitions. The influence of this deficit on semantic relatedness or on the number of switches would depend on whether or not the repetitions are included in the computations.

The literature provides evidence that the cognitive functions cited above are related to the frontal lobes or even to more circumscribed parts of these structures. Frontal patients present deficits in the generation and implementation of strategies (Alexander, Stuss, & Fansabedian, 2003; Burgess & Shallice, 1996b; Fletcher, Shallice, & Dolan, 2000; Gershberg & Shimamura, 1995; Incisa della Rocchetta & Milner, 1993; Moscovitch, 1992, Owen et al., 1990; Stuss, Alexander, Palumbo, & Buckle, 1994), in switching between cognitive sets (Milner, 1964, Owen et al., 1991; Troyer et al., 1998, Warrington, 2000), in the initiation of responses (Burgess & Shallice, 1996b; Godefroy, Lhullier-Lamy, & Rousseaux, 2002) and in the monitoring and checking of an ongoing task (Henson, Shallice, & Dolan, 1999; Luu, Flaisch, & Tucker, 2000; Reverberi, Lavaroni, Gigli, Skrap, & Shallice, 2005).

Coming back to semantic fluency, what we need is a way to evaluate the status of the different cognitive components of the task starting from a raw series of words. In an influential work, Troyer and collaborators (1997) devised a principled solution based on a two-component model of memory search (Moscovitch, 1992), which assumes the existence of an executive component devoted to strategic retrieval processes and monitoring and another component responsible for automatic encoding and information retrieval. Accordingly, Troyer and collaborators introduced two indices for scoring semantic fluency, that they termed “switching” and “clustering”. Switching was defined as the absolute number of transitions between subcategories and clustering was defined as the average number of words of semantically related items in a sequence (“cluster”). In the study by Troyer and colleagues, frontal patients produced a lower number of words and, interestingly, a lower absolute number of switches between subcategories. However, in our opinion, the absolute count of the switches between different subcategories does not help to discriminate among the above-mentioned possible components of frontal impairment. In fact, all the factors that decrease the number of produced words can also affect the absolute switch index. In particular, a mere initiation deficit would decrease the absolute number of switches, as a potentially normal sequence would be cut only because of its slow production (see online supplementary material for an extended discussion of these points). Strong criticism based on the numerical implications of the proposal of Troyer and colleagues was made by Mayr (2002), while Reverberi, Capitani and Laiacona (2004) pointed out the danger connected to a too-loose definition of subcategories and, consequently, of switches.

Considering the different criticisms put forward regarding the “clustering” and “switching” indices, Reverberi and colleagues (2004) suggested a different approach that avoids counting switches but introduces instead an index for describing the structure of the word sequence. Studying the category “fruit”, the authors preliminarily collected semantic proximity ratings for each pair of the 32 items most frequently produced in a semantic fluency task. Using these data, they computed the average semantic proximity between the pairs of successively produced words, a quantitative index that reflects the degree of semantic organization of a sequence.

The average semantic proximity seems a promising index for rating the semantic structure of a series of words¹ and the approach for analysing semantic fluency proposed by Reverberi et al. (2004) closely fits the aims of the present study. Moreover, proximity data also grants a sound and empirically validated way to identify the taxonomical tree of the explored category, providing a principled definition of the subcategories. Indeed, only through a stringent definition of the subcategories and a list of their components the number of switches can be reliably quantified. Since the absolute number of switches is ambiguous, it seems wiser to calculate the relative number of switches with respect to the sequence length.

The aim of the present report is to analyse the cognitive dimensions that characterise the semantic fluency impairment of frontal patients. We studied a sample of patients affected by frontal lesions by means of some original indices derived from previous studies on healthy participants carried out by our group. As the literature suggests that the mesial aspects of the frontal lobe are related to general activation while the lateral aspects are crucial for strategy compliance, we aimed to verify whether the type of performance of our patients differed according to the locus of the frontal lesion.