Does phonological working memory impairment affect sentence comprehension? A study of conduction aphasia

Abstract

Background: The nature of the relation between phonological working memory and sentence comprehension is still an open question. This question has theoretical implications with respect to the existence of various working memory resources and their involvement in sentence processing. It also bears clinical implications for the language impairment of patients with phonological working memory limitation, such as individuals with conduction aphasia.

Aims: This study explored whether limited phonological working memory impairs sentence comprehension in conduction aphasia.

Methods & Procedures: The participants were 12 Hebrew-speaking individuals with conduction aphasia who, according to 10 recall and recognition span tasks, had limited phonological short-term memory in comparison to 296 control participants. Experiments 1 and 2 tested their comprehension of relative clauses, which require semantic-syntactic reactivation, using sentence–picture matching and plausibility judgement tasks. Experiments 3 and 4 tested phonological reactivation, using two tasks: a paraphrasing task for sentences containing an ambiguous word in which disambiguation requires re-accessing the word form of the ambiguous word, and rhyme judgement within sentences. In each task the distance between a word and its reactivation was manipulated by adding words/syllables, intervening arguments, or intervening embeddings.

Outcomes & Results: Although their phonological short-term memory, and hence their phonological working memory, was very impaired, the individuals with conduction aphasia comprehended relative clauses well, even in sentences with a long distance between the antecedent and the gap. They failed to understand sentences that required phonological reactivation when the phonological distance was long.

Conclusions: The theoretical implication of this study is that phonological working memory is not involved when only semantic-syntactic reactivation is required. Phonological working memory does support comprehension in very specific conditions: when phonological reactivation is required after a long phonological distance. The clinical implication of these results is that because most of the sentences in daily language input can be understood without phonological reactivation, individuals with phonological working memory impairment, such as individuals with conduction aphasia, are expected to understand sentences well, as long as they understand the meaning of the sentences and do not attempt to repeat them or encode them phonologically.

Keywords:

• Aphasia
• Conduction aphasia
• Working memory
• Syntax
• Sentence comprehension
• Hebrew

Acknowledgments

This research was supported by a research grant from the National Institute for Psychobiology in Israel (Friedmann 2004-5-2b), and by the Israel Science Foundation (grant no. 1296/06, Friedmann).

Notes

¹For the three individuals who were tested 2–3 months post onset, all tests were administered within a short time, and each session included re-administration of the span test. No change in spans was detected for any of them within the testing period.

²In fact, even if the participants decide not to attend to the padding material, their inattention should not matter. Baddeley (1997) and Salamé and Baddeley (1987, 1989) describe the unattended speech effect, according to which immediate recall is impaired even when the verbal material that is heard when trying to remember a sequence is irrelevant and unattended (e.g., when it is in a foreign language or includes nonwords) (Colle & Welsh, 1976; Salamé & Baddeley, 1987).

³We did not collect response times, partly because comparing response times between healthy individuals and aphasics is problematic, as brain-damaged individuals typically show slower-than-normal response times (Tartaglione, Bino, Manzino, Apadavecchia, & Favale, 1986). Additionally, to give our participants the greatest chance to demonstrate their abilities, we preferred not to impose any time limits on them.

⁴This effect of age on comprehension of object relatives, which require syntactic-semantic activation, is in line with previous findings (e.g., Zurif, Swinney, Prather, Wingfield, & Brownell, 1995). Indeed, the participants in the oldest age group still performed 92.5% correct and above in the most demanding conditions, but the decrease in performance may be due to the shallower semantic encoding. The finding that, with age, word spans decrease but nonword spans stay unchanged indicates that phonological encoding remains the same in elderly individuals, whereas lexical-semantic encoding deteriorates (see Gvion & Friedmann, this issue). This may be the source for the older participants' relative difficulty with relative clause comprehension that whereas limited phonological encoding does not cause impaired comprehension of object relatives, decreased ability to encode semantically does. A study of object relative comprehension in Hebrew-speaking adults aged 71–82 years (Gvion, Shaham-Zimmerman, Tvik, & Friedmann, in press) indicated that, as in the current study, the elderly participants showed decreased word spans but their nonword spans were similar to those of the young controls. Crucially, their comprehension of object relatives in which the two NPs in the relative clause were semantically similar was significantly poorer (81% correct) than that of the younger participants (95%). It thus seems that an inability to encode a representation of NPs that would be semantically rich enough to allow for a distinction between them, led to the elderly participants' difficulty in comprehension. In a way, the relation between memory and comprehension in elderly people is a mirror image of the one we find in individuals with conduction aphasia: whereas individuals with conduction aphasia are impaired in phonological encoding and show no deficit in comprehending relative clauses, elderly individuals show deterioration in semantic encoding and poorer comprehension of object relatives.

⁵For TG we have data from only one set of sentences, five sentences per condition, and he made errors in two sentences from two conditions (GAD 2, 9), so his performance in these conditions was also not significantly better than chance.

Figure 2. Individuals with conduction aphasia: comprehension of subject relatives with the different distances between the antecedent and the gap (% correct).

Figure 3. Individuals with conduction aphasia: comprehension of object relatives with the different distances between the antecedent and the gap (% correct).

⁶Some of the participants with aphasia found the test very frustrating, so we had to stop the test at a certain point without completing it. We also had to discard some sentences from the data analysis because it was impossible to judge from the participants' responses whether they understood the sentence or not. This is why the analysis of the results is based on a slightly smaller number than the number of participants times the number of sentences. On these bases, 36 of the 528 short-distance ambiguous sentences and 37 of the 528 long-distance sentences with ambiguous words were discarded from the analysis.

Figure 5. Individuals with conduction aphasia: comprehension of sentences with an ambiguous word that gets disambiguated after a short or long distance: Experiment 3 (% correct).

⁷Because some of the participants had deficits that also involved the phonological output buffer, we analysed the correlation of sentence comprehension performance only with the recognition spans.

⁸Two other domains that might require a syntactic working memory are building the layers in the syntactic tree and maintaining and comparing competing syntactic representations. The syntactic tree consists of three hierarchically ordered layers: from top to bottom, the C(omplementiser) P(hrase), the I(nflection) P(hrase), and the V(erb) P(hrase) (Chomsky, 1986, 1995; Rizzi, 1997). One can assume that each layer adds load to the syntactic WM, and that limited syntactic WM might allow for the projection of only one or two of these layers. And indeed, according to some accounts, individuals with agrammatic aphasia are impaired in constructing the top syntactic layers (Friedmann, 2001, 2002, 2005, 2006; Friedmann & Grodzinsky, 1997, 2000), and studies of brain activation during sentence comprehension draw a similar picture: several areas of the brain show significantly more activation when hearing a sentence that includes more syntactic layers (Shetreet, Friedmann, & Hadar, 2009). Another task that might be performed by a syntactic WM is maintaining two competing syntactic representations in order to compare and choose between them. Such a mechanism is suggested for the processing of various syntactic structures (see, for example, Grodzinsky & Reinhart, 1993, for pronominal binding; Reinhart, 2004, for focus acquisition and stress shift; and Fox, 1995, 2000; and see Vosse & Kempen, 2000, for holding tentative attachments between nodes of lexical frames until a single syntactic tree is built). Researchers investigating brain tissue reach similar conclusions. For example, Thompson-Schill (2005) and Thompson-Schill, D'Esposito, Aguirre, and Farah (1997) have suggested on the basis of fMRI data that Broca's area has a role in selecting among competing alternatives from semantic memory.

⁹For comparison, the individuals with conduction aphasia who participated in Experiment 1 performed significantly better than the five individuals with agrammatism reported in Gvion (2007) on the 80 subject relatives and 80 object relatives (U = 67.5, p = .02, and U = 82, p = .0004, respectively), and also significantly better than 14 individuals with agrammatism (7 agrammatic aphasics reported in Friedmann & Shapiro, 2003, and 7 reported in Gvion, 2007) in the sentences with GAD 2 (U = 186.5, p = .05, for subject relatives; U = 240.5, p <.0001, for object relatives). Similarly, in Experiment 2 their comprehension of object relatives with double embedding was between 80% and 97% correct, whereas the three individuals with agrammatism reported in Gvion (2007) attained only 57–63% correct, not significantly above chance.