Syntactic and morphosyntactic processing in stroke–induced and primary progressive aphasia

2.1. Method

2.2. Results

Percent correct production (SPPT) and comprehension (SCT) of each sentence type for the PPA and StrAph participant groups are presented in Tables 5 and 6, respectively and in Fig. 2. A 2 (group) × 2 (modality) × 2 (canonicity) ANOVA for the PPA participants revealed significant main effects of group (F (1, 23) = 5.87, p = 0.024), with PPA-G showing poorer performance than PPA-L, modality (F (1, 23) = 9.55, p = 0.005), with production more difficult than comprehension, and sentence canonicity (F (1, 23) = 39.61, p < 0.001), with noncanonical sentences more difficult than canonical ones. A significant two-way interaction effect also was found between modality and canonicity (F (1, 23) = 8.13, p = 0.009), indicating that the difference between canonical and noncanonical sentences was greater for production compared to comprehension. Additionally, there was a significant interaction between group and canonicity (F (1, 23) = 10.73, p = 0.003), with the canonicity effect greater for the PPA-G than for the PPA-L group. There was also a significant group × canonicity × modality interaction (F (1, 23) = 9.328, p = 0.006). Post-hoc analysis using the Mann-Whitney test revealed that the PPA-L group performed significantly better than the PPA-G group in production of noncanonical sentences (Z = −2.19, p = 0.028). Further, the difference between groups for comprehension of noncanonical sentences approached significance (Z = −1.72, p = 0.085). However, the two groups did not differ in production or comprehension of canonical forms (Z = −.0784, p = 0.433 and Z = −1.28, p = 0.281, respectively).

For the StrAph data, a 2 × 2 × 2 ANOVA revealed a main effect of group (F (1, 35)= 5.75, p = 0.021), with StrAg showing poorer performance than StrAn, as well as a main effect of modality (F (1, 35) = 14.01, p = 0.001) and canonicity (F (1, 35) = 54.28, p < 0.001). Significant interactions were also found between modality and canonicity (F (1, 35) = 10.99,p = 0.002), and between canonicity and group (F (1, 35) = 5.4, p = 0.026). Although we did not find a three-way group × canonicity × modality interaction, follow-up comparisons between the StrAg and StrAn groups, using the Mann-Whitney test, were performed. These analyses revealed that the StrAg group’s performance was significantly poorer than that of the StrAn group for production of noncanonical sentences (Z = −2.78, p = 0.005) and the difference in comprehension of noncanonical sentences approached significance (Z = −1.91, p = 0.056). However, the two groups did not differ significantly in the comprehension (Z = −1.53, p = 0.125) or production (Z = −1.1, p = 0.272) of canonical sentences.

Statistical comparison between the two agrammatic groups, namely PPA-G and StrAg, yielded no main effect of aphasia type (F (1, 31) = 0.078, p = 0.782) (see Fig. 2). There was a significant effect of modality (F (1, 31) = 16.03, p < 0.001), with production more impaired than comprehension, and of canonicity (F (1, 31) = 49.85, p < 0.001), with noncanonical sentences more difficult than canonical ones. There was also a significant two-way interaction between modality and canonicity (F (1, 31) = 13.35, p = 0.001), with the canonicity effect larger for production than for comprehension.

2.3. Discussion

This experiment revealed considerable similarities between PPA and StrAph. In both groups, individuals with agrammatic deficits (namely, PPA-G and StrAg) were significantly more impaired in production of noncanonical sentences than participants with the logopenic variant of PPA or stroke-induced anomic aphasia. Likewise, there was a trend toward impaired comprehension of noncanonical sentences in the two agrammatic groups, compared to the other two groups. Interestingly, the deficits did not extend to canonical sentences, on which production and comprehension for the agrammatic participants was comparable to that of the logopenic and anomic groups. The selective deficit in noncanonical sentences for agrammatic speakers indicates a syntactically related impairment. It is important to point out that deficits in comprehension of noncanonical sentence comprehension may be related to general impairments in verbal working memory (e.g., [10]; but see [51] for arguments against this view). However, we find that a working memory account does not completely explain the deficit patterns found here. Notably, comprehension of subject-relative constructions, which are canonical and similar in length to object relative constructions, was superior to object-relative structures (85.4% correct for subject relatives, 74.5% correct for object relatives for PPA-G; 90% for subject relatives compared to 60.9% for object relatives for StrAph).

Another interesting result of the experiment is that the PPA-G and the StrAg groups performed very similarly, with no significant differences found for any tasks or conditions. This indicates that PPA-G individuals indeed exhibit noncanonical sentence deficits in line with those seen in stroke-induced agrammatic speakers.

3.1. Participants

Participants included 7 PPA-G, 10 PPA-L, 20 StrAg and 9 StrAn individuals, recruited from the subject pools of the Aphasia and Neurolinguistics Research Laboratory in Evanston, IL and the Cognitive Neurology and Alzheimer’s Disease Center in Chicago, IL. The participants met the same selection criteria as in Experiment 1. All 17 PPA participants, and 25 of the 29 StrAph participants, were right handed. The groups were matched for education (Kruskal Wallis, χ²(3) = 3.36, p = 0.339) and age (Kruskal-Wallis, χ²(3) = 7.64, p = 0.054) (see Table 1).

The PPA participants recruited for the study showed no evidence of stroke or other neurological disorder and all presented a history of progressive language deficits with relatively spared performance in other cognitive domains. As in Experiment 1, the MMSE, WMS, Facial Recognition and Trail Making Tests were administered to rule out possible memory, visual perceptual, and attention deficits (see scores in Table 2). All participants scored 24 or higher on the MMSE.

As in Experiment 1, PPA variants were determined based on scores in the PPVT and the NAT. Both the PPA-G and the PPA-L groups showed relatively spared word comprehension, with no significant difference between groups (PPA-G PPVT mean: 95.2; PPA-L PPVT mean: 92.8, Mann-Whitney, Z = −0.453, p = 0.650). Further, on the NAT, the two groups did not differ significantly on canonical sentences (PPA-G mean: 85.72, PPA-L mean: 95.33, Mann-Whitney, Z = −0.959, p = 0.338). However, the PPA-G group was significantly more impaired than the PPA-L group on noncanonical sentences (PPA-G mean: 43.81, PPA-L mean: 90.0, Mann-Whitney, Z = −3.45, p = 0.001).

We also administered the WAB to all PPA participants with mean AQs of 84.2 (range: 75.3–95.2) and 91.5 (range: 83.2–97.2), for the PPA-G and PPA-L participants, respectively, which again differed significantly between groups (Mann-Whitney, Z = − 2.0, p = 0.045) due to significant differences in WAB fluency scores (Mann-Whitney, Z = − 2.39, p = 0.017). However, the groups did not differ significantly on the auditory comprehension (Mann-Whitney, Z = − 0.86, p = 0.389), repetition (Mann-Whitney, Z = −1.468, p = 0.142) or naming (Mann-Whitney, Z = − 0.245, p = 0.806) subtests of the WAB.

The stroke aphasic participants all suffered a left hemisphere stroke, with the exception of one, who presented with a right hemisphere lesion and crossed aphasia, with an average of 5.6 years post onset. Aphasia type was determined as in Experiment 1. All participants presented with mild-to-moderate aphasia (WAB-AQ mean: StrAg: 78.3, range 65.5–87.6; StrAn: 87.4, range: 69.4–93.3), with WAB fluency scores ranging from 4 to 5 (mean = 4.5) for the former group and from 6 to 9 (mean = 8.0) for the latter. As in Experiment 1, statistical analyses comparing the two groups indicated significant differences in WAB-AQ (Mann-Whitney, Z = −3.39, p = 0.001) and fluency scores (Mann-Whitney, Z = −4.44, p < 0.001), but the two groups did not differ significantly on auditory comprehension (Mann-Whitney, Z = − 0.33, p = 0.74), repetition (Mann-Whitney, Z = − 1.53, p = 0.125), or naming (Mann-Whitney, Z = − 1.56, p = 0.119) subtests of the WAB. (See Table 3 for language test scores for the PPA and StrAp groups.)

The two agrammatic groups, PPA-G and StrAg, did not differ on their WAB-AQs (Mann-Whitney, Z = 0.1772, p = 0.081). Similarly, there was no difference in WAB-AQ between the PPA-L and the StrAn group (Mann-Whitney, Z = 1.103, p = 0.278).

3.2. Materials and procedure

Ten two-argument (transitive) verbs were selected, including five with regular (e.g., tickle) and five with irregular (e.g., eat) past-tense morphology, with each elicited in six different forms: infinitive (e.g., to eat), progressive (e.g., is eating), present singular (e.g., eats), present plural (e.g., eat), past regular (for regular verbs, e.g., tickled), and past irregular (for irregular verbs, e.g., ate), resulting in a total of 50 experimental trials (see Table 7 for a set of sample stimuli). Note that the first two verb forms (i.e., the infinitive and progressive) are nonfinite – that is, uninflected for tense (in the progressive, tense is marked on the auxiliary verb, and the lexical verb is a participle), and the latter four are finite forms – inflected for tense. Black-and-white line drawings were prepared for all verbs, depicting an agent and a theme (e.g., a man eating a hamburger) to elicit all verb forms, except for the present plural, for which two agents and two themes were illustrated (e.g., two men eating hamburgers).

A sentence completion task, using temporal adverbs such as everyday (to elicit present tense singular and plural forms) and yesterday (to elicit past tense forms), was used to obligate production of each target verb form. Together with each stimulus picture, a sentence with a missing verb was presented and read aloud and participants were asked to complete the sentence by providing the correct verb form. To control for verb retrieval difficulties, the verb stem of each target form also was provided on each stimulus template (see Fig. 3). Responses were counted as correct if the target verb form was produced within 10 seconds of stimulus presentation. For any responses with self-corrections, the last attempt was scored.

3.3. Results

Table 8 presents percent correct responses in each of the nonfinite and finite conditions for the PPA and StrAph participant groups, with these data presented in graphic form for the PPA and StrAph groups in Fig. 4. Analysis of the PPA data using a 2 (group) × 2 (tensed/nontensed) ANOVA revealed a main effect of group (F (1,15) = 6.68, p = 0.021), with PPA-G more impaired than PPA-L, and a main effect of tense (F (1,15) = 30.8, p < 0.001), with tensed verbs more difficult than nontensed ones. There was also an interaction between group and tense (F (1,15) = 5.28, p = 0.036), with the tense effect greater for the PPA-G group than for the PPA-L group. Post-hoc comparisons showed significantly greater impairment for the PPA-G compared to the PPA-L participants on finite (Mann-Whitney, Z = −2.01, p = 0.044), but not on nonfinite (Mann-Whitney, Z = −1.44, p = 0.149), verb forms.

Similar analyses of the StrAph data also revealed a significant main effect for group (F (1, 27) = 5.56, p = 0.026), with StrAg more impaired than StrAn, and a main effect of tense (F (1,27) = 36.89, p < 0.001). In addition, as in the PPA group, post-hoc comparisons revealed a significantly greater impairment for finite, but not nonfinite, forms for StrAg compared to the StrAn group (Mann-Whitney finite: Z = −2.05, p = 0.04; nonfinite Z = −1.37, p = 0.17).

A 2 (group) × 2 (tensed/nontensed) ANOVA for the two agrammatic groups (PPA-G and StrAg) revealed a main effect of tense (F (1, 25) = 34.0, p < 0.001), but no main effect of aphasia type (F (1, 25) = 1.47, p = 0.237) or interaction between tense and aphasia type (F (1, 25) = 0.586, p = 0.451) (see Fig. 4).

3.4. Discussion

As in the previous experiment, these results point to interesting symmetries between StrAph and PPA. In both cases participants with agrammatism scored lower than those with anomic or logopenic impairments on finite verb form production, though not on nonfinite verb form production. In addition, as in Experiment 1, no differences were found between the two agrammatic groups (namely StrAg and PPA-G).

These findings corroborate previous research showing that stroke agrammatic individuals perform significantly more poorly than individuals with stroke-induced anomic aphasia in the production of inflected verb forms (e.g. [5]). The data also provide important new information distinguishing between agrammatic and logopenic variants of PPA. While individuals with PPA-G show impairments in production of verb tense (mirroring those seen in individuals with agrammatism resulting from stroke in this and previous studies), individuals with PPA-L do not exhibit such deficits.

4.1. Method

4.2. Results

Results on the six outcome measures specified above for each participant group are shown in Table 9. A MANOVA with the six output measures as dependent variables revealed an effect of participant group on all the measures (p’s < 0.001), except for noun to verb ratio (p = 0.072). A post-hoc Bonferroni test was further undertaken to assess between group differences. With regard to measures of fluency, for both WPM and MLU significant differences between the control group and all patient groups were found (p’s < 0.01 and p’s ≤ 0.001, respectively). In addition, the PPA-L group produced significantly more WPM compared to the PPA-G group (p = 0.006) and the StrAg (p = 0.002), although the two groups did not differ significantly for MLU (p = 0.827). No significant differences were noted between the two agrammatic groups for WPM (p = 1) or MLU (p = 0.758) (see Fig. 5).

Turning next to the proportion of grammatical sentences no significant difference was found between the control and PPA-L groups (p = 0.309), however, both groups differed significantly from the PPA-G group (p’s < 0.001). The StrAg group likewise displayed a lower proportion of grammatical sentence than controls (p < 0.001). In addition, the two agrammatic groups did not differ significantly from each other (p = 1). A similar pattern was noted for the proportion of correctly inflected verbs. Although significant differences were found between both agrammatic groups and the control group (PPA-G vs. control, p = 0.005, StrAg vs. control, p = 0.013), as well as between the PPA-L and PPA-G group (p = 0.018), the PPA-L and control groups did not differ from one another (p = 1). Further, no significant difference was found between the two agrammatic groups (p = 1). (see Fig. 6).

For noun-to-verb ratio (N:V ratio), mean ratio for the PPA-G group was higher than that for the PPA-L group, indicating that the agrammatic participants used more nouns than verbs in their narratives. However, this difference did not reach significance (p = 0.245), due to the large variability within the agrammatic group. As with previous measures, the two agrammatic groups did not differ on N:V ratio (p = 1). With regard to open- and closed-class word production, a significant difference was found between the StrAg group and the control group, as well as between the StrAg and PPA-G groups, for open-class to closed-class ratios (p’s < 0.01). However, PPA-G, PPA-L and control participants did not differ significantly from one another (p’s = 1). (see Fig. 7).

4.3. Discussion

Results of the two fluency measures in the experiment, WPM and MLU, indicate that both PPA groups as well as the stroke-induced agrammatic patient groups were less fluent than healthy controls in their narrative production. The WPM data further showed that the agrammatic PPA participants were significantly less fluent than the PPA-L group, in line with previous research (e.g. [2,5,68]). However, MLU did not differentiate the two patient groups. In support of Thompson et al. [57], these data indicate that use of the term non-fluent to classify individuals with PPA does not differentiate between agrammatic and logopenic PPA variants. Indeed, both PPA groups demonstrate nonfluent speech compared to normal, however, the two differ in syntactic and morphosyntactic abilities. Importantly, there were no differences in fluency between the two agrammatic groups (i.e. StrAg and PPA-G).

Two of our measures, namely the proportion of grammatical sentences and proportion of correctly inflected verbs, clearly indicate that the agrammatic groups are impaired with regard to (morpho)syntactic processing, showing deficits in forming syntactically correct sentences and in using grammatical morphemes to inflect verbs. These findings are consistent with previous ones, suggesting syntactic and morphosyntactic deficits in agrammatic PPA [2,55,57,68]. In contrast, performance of the logopenic PPA participants (i.e., PPA-L) was less impaired than that found in the agrammatic PPA group. These findings suggest relatively unimpaired (morpho)syntactic production ability in this patient group. However, it is worth noting that the proportion of grammatical sentences produced by PPA-L individuals was smaller than that of controls but this difference did not reach significance (i.e., 80.47% for PPA-L; 92.52 for controls). Thompson et al. [57] and Wilson et al. [68] reported similar patterns, with the PPA-L participants producing significantly fewer grammatical sentences compared to the unimpaired, control speakers, although, again, the PPA-L group produced significantly more grammatical sentences than the PPA-G group. In addition, in the present study, the PPA-L group showed very high levels of correct verb inflection (i.e., a mean of 97%), a pattern also noted in Thompson et al. [57]. Nevertheless, the mild deficit in grammatical sentence production found in PPA-L warrants further investigation to determine the robustness of this effect. For example, grammatical accuracy might be disrupted not because of syntactic processing deficits but rather because of more general word-finding difficulty, which might lead to rephrasing and resultant agrammatic output.

Results of the noun to verb ratio measure again indicate a similar deficit in the two agrammatic groups. The agrammatic participants (both PPA-G and StrAg) showed a trend towards producing more nouns and fewer verbs, compared to healthy speakers, suggesting a grammatical deficit involving the latter category (the same trend can be seen in the PPA-G group in Thompson et al. [57]). Graham et al. [30], however, did not find evidence of verb production deficits in their nonfluent PPA patients, possibly because their patient cohort included both agrammatic and logopenic variants.

A somewhat surprising result emerged from the open-class to closed-class word production ratio measure. Whereas for StrAg participants, the ratio was higher than that of controls, suggesting, as predicted, a limited use of grammatical, closed-class morphemes, the PPA-G group patterned with the PPA-L and control groups, rather than with the StrAg group, suggesting no impairment for this variable. Notably, Graham et al. [57] likewise did not find a reduced proportion of function words in this group. Based on these results, Graham and colleagues suggested that agrammatism (or progressive nonfluent aphasia) in PPA is not akin to a true agrammatism as seen secondary to stroke aphasia. However, this conclusion is problematic in light of the other agrammatic features displayed by speakers with PPA-G, e.g. reduced proportion of grammatical sentences and correct verb inflection. Notably, Thompson et al. [57] also found significantly reduced argument structure production in their cohort of PPA-G participants, compared to other PPA groups, which also is a pervasive pattern in agrammatism resulting from stroke [5,16,37,42,62]. However, we note that further research comparing open- to closed-class word production in progressive and stroke-induced aphasia is needed to determine whether this result reflects a genuine difference between PPA-G and StrAg and, if so, why. Indeed, Wilson et al. [68] found a significantly reduced ratio of closed class words in the agrammatic variant of PPA (albeit this was driven by only a few patients). Thompson et al. [55] also found reduced closed class word production in their three agrammatic participants. In addition, this pattern was seen in some of the PPA-G participants in Thompson et al. [57], although significant differences were not found between groups.

4.4. General discussion

In the present study, we examined various measures of syntactic and morphosyntactic performance in several participant groups, including individuals with agrammatic and logopenic variants of primary progressive aphasia (PPA) and those with agrammatic and anomic aphasia resulting from stroke. We examined comprehension and production of canonical and noncanonical sentence structures using the Northwestern Assessment of Verbs and Sentences – Sentence Comprehension Test and Sentence Production Priming Test, and tested production of tensed and nontensed verb forms using the Northwestern Assessment of Verb Inflection. We further examined narrative samples, focusing on measures of speech fluency as well as properties of grammatical speech production. The results showed that agrammatic PPA and stroke-induced aphasic individuals pattern alike on several (morpho)syntactic measures. Although the two groups differed in overall production of open- and closed-class words in narrative production, they showed similar deficits in production (and, to a lesser degree, comprehension) of noncanonical sentences, difficulties producing tensed verb forms, and their speech included fewer grammatical sentences and correctly inflected verbs than that of healthy speakers. Our results thus corroborate the findings of Thompson et al. [57] (as well as those of [47,55]), showing that despite the differences between PPA-G and StrAg individuals in disease etiology and mechanism, these two patient groups present with similar language deficit patterns.

In contrast, we found that logopenic PPA speakers did not display the syntactic and morphosyntactic deficits characteristic of the agrammatic groups, and performed significantly better on all tested measures, with the exception of open-to-closed class words in narratives (although other studies show closed-class word production deficits in PPA [55,68]. Notably, syntactic and morphosyntactic performance of the logopenic PPA group was comparable to that of speakers with stroke-induced anomic aphasia in Experiments 1 and 2, and to healthy speakers in Experiment 3. In general, the behavioral pattern of logopenic PPA suggests relatively intact syntactic knowledge, much like that of StrAn speakers.

It can be noted that the poorer performance of the agrammatic groups on pre-experimentallanguage tasks may have predicted these patterns, since participants were categorized as agrammatic based on their ability to produce sentences (i.e., on the Northwestern Anagram Test for PPA participants and in spontaneous speech samples for StrAph). We note, however, that the agrammatic groups in our experiments also were significantly less fluent than the anomic/logopenic groups, based on their WAB fluency scores (the mean fluency score for PPA-G participants across the experiments was 5.516, for StrAg – 4.502, for PPA-L – 8.714, and for StrAn – 7.74). Although fluency and grammatical ability often coincide with one another in StrAg, they do not in PPA-G [57]. We also point out that the NAT only grossly addresses sentence production ability in that this test requires participants to arrange written words to form sentences and, hence, off-line strategies can be used to generate sentences. Further, the spontaneous speech samples used to classify the StrAg participants were derived from the WAB picture description task, which often results in short samples that do not adequately illustrate patient abilities (see Thompson et al. [57], for discussion of the limitations of picture description for evaluating spontaneous speech) and the samples were not thoroughly analyzed. The results for Experiments 1 and 2 were derived from constrained production tasks, and Experiment 1 also examined syntactic comprehension. In addition, Experiment 3, which tested narrative production ability, used a story-telling task and a sophisticated analysis system to quantify linguistic ability.

The systematic comparison of (morpho)syntactic impairments associated with PPA and aphasia resulting from stroke reported in this paper thus revealed symmetry between these neurolinguistic disorders, such that the deficit patterns presented by individuals with the agrammatic variant of PPA parallel, to a large degree, those of the agrammatism associated with stroke; whereas the (morpho)syntactic characteristics of the logopenic variant of PPA are similar to those of anomic, stroke-induced, aphasia. These findings indicate that use of linguistically specific and sophisticated tests designed to systematically examine grammatical ability can and should be used to study the language deficit patterns seen in PPA. We further suggest that such tests be used clinically to differentiate between the different subtypes of PPA. This practice will not only contribute to our understanding of PPA and differences in deficits across PPA variants, it also will be informative with regard to how, or if, the language patterns associated with PPA differ from or are similar to those found in stroke aphasia. This latter issue, importantly, may provide insights into how language is processed in the brain. In particular, longitudinal studies examining the neural correlates of progressive (morpho)syntactic deterioration in PPA, compared to static (or improving) neurobehavioral correlates of such processing in stroke-induced aphasia may lead to important discoveries about how the brain computes syntax and morphosyntax.