Effect of speech rate and complexity on sentence comprehension in Alzheimer’s disease

ABSTRACT

Previous research has not shown a benefit of slowed speech on the comprehension of sentences by persons with Alzheimer’s disease (AD). The objective of this study was to replicate and extend the findings from previous research by employing a novel speech rate manipulation that inserted strategic pauses at phrase and clause boundaries in sentences. Fourteen participants with AD were instructed to match auditorily presented sentences to one of several pictures that corresponded to the correct meaning of each sentence. The sentences varied in their speech rate and grammatical complexity. The results show that participants’ comprehension did not significantly benefit from the altered speech rate, though participants did demonstrate better comprehension of simpler than more complex sentences. The findings extend previous research by showing that even when employing a more natural method of slowing the speech signal it did not benefit AD participants’ comprehension. The results also contribute to evidence-based clinical recommendations concerning speech modifications to facilitate verbal comprehension in Alzheimer’s disease.

KEYWORDS:

• Alzheimer’s disease
• Speech rate
• Sentence comprehension

Introduction

Alzheimer’s disease (AD) is a progressive brain disorder that affects memory, language, reasoning, and judgement. These impairments together alter a person’s ability to communicate during everyday activities, implicating both comprehension and production of speech. Difficulties in understanding spoken language can lead to problematic behaviors, caregiver burnout, and increase the risk of institutionalization (Burgio, Allen-Burge, Stevens, Davis, & Marson, 2018; Cummings, Miller, Hill, & Neshkes, 1987; Feast et al., 2016; Mirdrikvand, Panahi, & Hoseyni Ramaghani, 2016; Steeman, Abraham, & Godderis, 1997). Care partners often struggle to support their loved one’s communication needs, and have reported that communication challenges are a major source of stress (Orange, 1991; Richter, Roberto, & Bottenberg, 1995; Savundranayagam, Hummert, & Montgomery, 2005; Small, Geldart, & Gutman, 2000; Williamson & Schulz, 1993; Wilson, Rochon, Mihailidis, & Leonard, 2012; Wray, 2020; Young, Manthorp, Howells, & Tullo, 2011).

Recognizing the palpable impact that AD has on communication and wellbeing, a variety of strategies have been proposed for care partners to improve communication. Two common recommendations are that care partners should modify their speech by speaking slowly and using simple sentences (Alzheimer Society of Canada, 2022; Alzheimer’s Association, 2022; Ostuni & Santo-Pietro, 1991; Ripich, Ziol, Fritsch, & Durand, 2000; Small, Gutman, Makela, & Hillhouse, 2003; Wilson et al., 2012; Young et al., 2011). Although a number of studies have found that persons with AD perform better when presented with shorter and/or simpler sentences compared to longer and/or more complex sentences (Bayles, 2003; Bickel, Pantel, Eysenbach, & Schröder, 2000; Choi & Jeon, 2019; Gilardone, Longo, & Papagno, 2023; Grober & Bang, 1995; Kempler, Almor, Tyler, Andersen, & MacDonald, 1998; Liu, Wang, Wang, & Sun, 2019; MacDonald, Almor, Henderson, Kempler, & Andersen, 2001; Markova, Horvathova, Kralova, & Csefalvay, 2017; Rochon, Waters, & Caplan, 1994; Small, Kemper, & Lyons, 1997), there has been no empirical validation that slowed speech facilitates comprehension for individuals with AD. In fact, several studies have found that a slowed rate of speech may actually compromise their message comprehension (Kemper & Harden, 1999; Pashek & DiVenere, 2006; Small, Andersen, & Kempler, 1997; Small, Kemper, et al., 1997; Tomoeda, Bayles, Boone, Kaszniak, & Slauson, 1990).

Previous research has manipulated speech rate by slowing the speech signal across the entire sentence (e.g., systematically inserting silent gaps at every n^th portion of the signal, such as after each word in the sentence). This method of slowing may actually increase cognitive processing demands due to the unnatural prosody that results from the gaps between each word as well as the listener having to hold and integrate discrete segments of information for a longer period of time (i.e., across the entire sentence). In support of the latter explanation, Small, Andersen, et al. (1997) found that significant deficits in working memory were related to lack of benefit from such slowed speech by persons with AD.

An alternative more natural method of slowing the speech rate is to provide more time for the listener to process information at strategic points in the sentence. Specifically, phrase and clause boundaries are strategic junctures where listeners interpret chunks of related information. For example, in the sentence ‘The man who the policeman handcuffed scolded the youngster’, there are two clauses—a main clause: the man … scolded the youngster, and an embedded clause: the policeman handcuffed [the man]. By inserting a pause at the junctures surrounding the embedded clause (i.e., before who, and after handcuffed), the listener is first cued to the beginning of the embedded clause and then provided with additional time to compute the meaning of the embedded clause before continuing on to process the remaining part of the sentence. Findings from research on healthy younger and older adults, as well as individuals with aphasia, indicate that they use pauses at such strategic junctures to facilitate their comprehension (Blumstein, Katz, Goodglass, Shrier, & Dworetsky, 1985; Lasky, Weidner, & Johnson, 1976; Wingfield, Kemtes, & Miller, 2001; Wingfield, Lahar, & Stine, 1989).

To our knowledge, this alternative method of slowing speech has not been tested with the AD population. The present study, therefore, fills a gap in research on the effect of speech rate manipulation by creating pauses at critical grammatical junctures (i.e., phrase or clause boundaries). This study also replicates previous research by examining the impact of sentence complexity, and its relationship to speech rate. We predicted that persons with AD would show greater comprehension of sentences with simpler syntactic structure and of sentences that have pauses inserted at strategic processing boundaries. In addition, we hypothesized that because more complex sentences create greater processing demands, the effect of pauses will be larger for more complex sentences.

Materials and methods

Participants

Fourteen participants were recruited through the Clinic for Alzheimer’s Disease and Related Disorders (CARD) at the University of British Columbia, as well as through advertisements posted in the Alzheimer Society of British Columbia Newsletter. Each participant was diagnosed with probable AD or possible AD by a physician according to acceptable research criteria and clinical practice (American Psychiatric Association, 2013; McKhann, Drachman, et al., 2011; McKhann, Knopman, et al., 2011; Minati, Edginton, Bruzzone, & Giaccone, 2009). Participants were administered the Mini-Mental State Examination (MMSE; Folstein, Folstein, & McHugh, 1975) and the Modified Mini-Mental State (3MS; Teng & Chui, 1987) to assess cognitive status. Only participants who scored in the mild-to-moderate range were included in this study (Mild: N = 12; Moderate: N = 2). Participants were not clinically depressed in scoring 14 or less out of 30 on the Geriatric Depression Scale (Yesavage et al., 1982).

Nine of the participants were male, and 5 were female. Participants were 57–91 years of age (M = 78), and resided in the community with a full-time care partner. Participants had sufficient functional vision to read newspaper-size fonts and hearing to perceive the audio material in test and training activities.

Measures

Sentence picture matching (SPM) task

Participants were instructed to listen to sentences presented one at a time and point to one of four pictures that accurately depicted the meaning of the heard sentence (see Figure 1). The sentences were taken from Small, Kemper, et al. (1997) and implemented using the App ‘Keynote’, where scanned-in pictures and recorded audio files were paired for presentation to participants on an iPad. Prior to administering the experiment, participants were given a word comprehension test to ensure that they could hear and recognize all of the individual picture elements from the sentence stimuli. Each participant completed the word comprehension pre-test activity and two sentence-picture matching task practice trials prior to starting the test SPM trials to ensure that the words and sentences were sufficiently audible and that each participant understood and was able to follow the procedure for the experiment.

Figure 1 . Example of Sentence Picture Matching Task Stimulus: ‘The child across from the mother served the father.’

Grammatical complexity

Three types of sentences were used in examining the effects of grammatical complexity: (a) Active, (b) Passive, and (c) Subject-Object (see Table 1 for examples). These sentence types were selected because they were likely to show robust differences in complexity effects.

Table 1 . Sentence type examples.

Sentence Type	Sentence Example
Active	The child across from the mother served the father.
Passive	The child was served by the father.
Subject-Object	The mother who the child bumped into served the father.

That is, Active sentences are considered to be the simplest sentence type, followed by Passive sentences, which have a reversed mapping of thematic roles, and finally, Subject-Object sentences, which are the most complex because of the presence of an embedded clause as well as the reversed thematic role assignment in the embedded clause. Ten sentences were created for each of the three sentence types, for a total of 30 sentences. For each sentence, one target picture and three foils (or distractor pictures) were presented in quadrants (see Figure 1).

Speech rate

Each of the 30 sentences, and 2 practice trials, were recorded at both a normal speech rate (180 WPM), and a slowed speech rate (110 WPM). Stimuli were recorded at 32 bit rate on Pocket WavePad in a sound booth using an Optimus unidirectional headset microphone. Both the slowed and normal speech rates were recorded by an adult female using her natural voice. There was no mechanical manipulation of speech to control the duration of the sentence, but rather an extended natural pause was placed at strategic boundaries. For Active and Passive sentences, the pause occurred at a phrase boundary (e.g., Active: The child [pause] across from the mother [pause] served the father; Passive: The child [pause] was served [pause] by the father), and for Subject-Object sentences at phrase and clause boundaries (e.g., The mother [pause] who the child bumped into [pause] served the father). Pauses at these junctures resembled naturally occurring pauses in speech (Blumstein et al., 1985; Lasky et al., 1976; Wingfield et al., 1989; Wingfield et al., 2001).

The target duration of the sentences at the normal speech rate (180 WPM) was predetermined, calculated by the following Words per Minute equation: (number of words) x 60/180 = time in seconds. The target duration of the sentences at a slowed speech rate (110 WPM) was predetermined, calculated by the following Words per Minute equation: (number of words) x 60/110 = time in seconds. For example, a sentence with 10 words had an overall target duration of 3.33 s at the normal rate, and a target duration of 5.45 s at the slow rate, with a +/− 0.1 s margin of error. The 40% reduction for the slow rate was selected based on findings from previous research indicating that smaller speech reductions (15% or 25%) have little or no effect on performance (Small, Kemper, et al., 1997; Tomoeda et al., 1990).

Each spoken sentence was recorded at the normal rate, and then again with pauses in the slow rate condition. When the sentence length achieved the target duration (+/− 0.1 s) it was reviewed by two researchers for clarity, prosody, and naturalness. The goal was to maintain consistent pace and style of speaking pre- and post-pause across conditions, with the difference of speech rate being predominantly in the duration of the pauses. The pauses were implemented in a manner to maintain natural prosody throughout the sentence.

Procedures

Word comprehension pre-test

In order to ensure word comprehension abilities would not interfere with participants’ sentence comprehension, their comprehension of each lexical item (i.e., character or object) appearing in the SPM pictures was tested before administering the SPM test trials.

Sentence picture matching practice trials

Participants were instructed to listen to two sentences (Active, Subject-Object), one at a time, and point to one of the four pictures that best depicted the meaning of each sentence. Sentences were repeated as necessary, and volume was adjusted to a comfortable listening level.

Sentence picture matching test trials

The SPM task was administered in two blocks of 30 sentences, for a total of 60 sentences. The order of blocks and speech rate was counterbalanced across participants. Each of the 30 sentences was presented twice, once at a normal speech rate condition (Block A) and once at a slower speech condition (Block B). Participants received one block of sentences in one session, and the other block of sentences in a second session spaced at least one week apart. Although it is unlikely that persons with AD would experience priming as a result of repeated presentation of the stimuli (Salmon, Shimamura, Butters, & Smith, 1988), presenting each block in a different session served to minimize this possibility. Participants were given only one opportunity to listen to each sentence (i.e., no repetitions allowed). If there was an administration error (e.g., the stimulus was played before the participant had indicated readiness, or there was an environmental distraction), the trial was repeated. After the participant heard the sentence, they were asked to point to one of four pictures that best depicted the sentence they heard.

All study activities occurred in the participants’ homes. Study protocols were approved by the university’s Behavioural Research Ethics Board (#H12-01806), and all participants provided written informed consent.

Data analysis

We conducted descriptive and inferential statistics using SPSS (version 26) statistical software. A 2 X 3 Repeated Measures ANOVA was used to examine effects of speech rate and sentence type. Follow-up paired samples t-tests were conducted and effect sizes calculated to assess differences between conditions.

Results

Comprehension accuracy by speech rate and sentence type across participants is presented in Table 2. Results from a repeated measures analysis of variance indicated there was no main effect of speech rate on sentence comprehension F (1,12) = 0.03, p = 0.86, and no significant interaction between speech rate and sentence type F (2,24) = 1.10, p = 0.35. However, there was a statistically significant effect of grammatical complexity on sentence comprehension, F (2,24) = 17.22, p < 0.01, ηp 2 = 0.59.

Table 2 . Comprehension accuracy by condition.

	Normal Speech Rate			Slow Speech Rate			Comprehension by Sentence Type^a
P#	Active	Passive	Subject-Object	Active	Passive	Subject-Object	Active	Passive	Subject-Object
1	10	9	8	10	9	8	10	9	8
2	3	4	4	5	7	4	4	5.5	4
3	4	3	4	4	2	2	4	2.5	3
4	3	6	2	6	5	2	4.5	5.5	2
5	5	7	2	5	5	6	5	6	4
6	7	9	4	.^b	.	.	7	9	4
7	7	9	5	9	9	4	8	9	4.5
8	10	10	9	10	10	6	10	10	7.5
9	9	9	7	10	9	8	9.5	9	7.5
10	7	10	5	5	9	5	6	9.5	5
11	5	7	4	4	2	4	4.5	4.5	4
12	8	10	6	9	5	8	8.5	7.5	7
13	6	5	3	6	8	1	6	6.5	2
14	7	6	5	8	7	5	7.5	6.5	5
Means (out of 10)	6.5	7.42	4.86	7.0	6.69	4.85	6.75	7.14	4.82

^a Collapsing across speech rate.

^b Due to scheduling issues, the data for the slow rate condition could not be collected for participant 6.

Follow-up paired samples t-tests revealed a significant difference between Active (M = 6.75, SD 2.21) and Subject-Object (M = 4.82, SD = 1.99) sentence types (t(13) = 6.23, p < 0.01, d = 0.92), and between Passive (M = 7.14, SD = 2.21) and Subject-Object (M = 4.82, SD = 1.99) sentence types (t(13) = 4.87, p < 0.01, d = 1.10), with large effect sizes. There was no statistically significant difference between Active and Passive sentence types (p > 0.05). These results show that when participants were presented with the most complex sentence type they responded correctly less than 50% of the time, whereas for simpler sentence types (Active, Passive), they were able to respond correctly on at least two-thirds of the trials.

Discussion

The primary objective of this study was to explore the impact of a new manipulation of speech rate, and a replication of grammatical complexity, on sentence comprehension by persons with Alzheimer’s Disease (AD). The findings provided only partial support for the predicted effects of these variables.

Sentence complexity

The hypothesis that persons with AD would show greater comprehension of sentences that have simpler syntactic structure (i.e., Active and Passive single clause) than sentences that have more complex syntactic structure (i.e., Subject-Object two clause) was confirmed. These results align with previous research that reported a decline in sentence comprehension in AD as sentence complexity increases (Bickel et al., 2000 Choi & Jeon, 2019; Grober & Bang, 1995; Kempler et al., 1998; Liu et al., 2019; MacDonald et al., 2001; Markova et al., 2017; Rochon et al., 1994; Small, Kemper, et al., 1997). In the embedded Subject-Object sentences (e.g., ‘The woman who_i the girl followed e_i shouted at the boy’), the relativized Subject noun phrase (The woman) is associated with an object gap (e_i) in the embedded clause, which requires the listener to hold this noun phrase (The woman) until it can be assigned a thematic role by the embedded verb (followed). The latter noncanonical assignment of a thematic role, along with retaining the Subject noun phrase in memory until the main clause verb (shouted) is encountered, places demands on working memory, which is compromised in AD.

Although Active sentences are generally simpler to process than Passive sentences due to their canonical assignment of thematic roles, in the present study we matched the length of Subject-Object and Active Sentences by including an additional prepositional phrase. This addition required that listeners assign an extra thematic role (location) to the noun in the prepositional phrase (The woman beside the boy shouted at the girl). Thus, while some previous research has reported worse comprehension for Passive than Active sentences, our findings of no significant differences between these two sentence types can be attributed to the extra processing demands in our Active sentences. The finding that comprehension was significantly better for Active than Subject-Object sentences indicates that grammatical complexity, not sentence length, influenced performance.

Speech rate

The second and third hypotheses that persons with AD would demonstrate improved comprehension at a slowed speech rate, especially for the more complex sentence type, were not confirmed. No differences in sentences comprehension were observed between the normal and slowed speech rate conditions for all sentence types. In other words, the extended pause appeared to neither benefit nor hinder sentence comprehension. These findings replicate several previous studies that reported null findings for various speech rate manipulations, while nevertheless showing robust effects of other variables (e.g., sentence complexity) despite having relatively small participant samples (Pashek & DiVenere, 2006; Small, Andersen, et al., 1997; Small, Kemper, et al., 1997; Tomoeda et al., 1990). Thus, the converging findings across studies suggest that the lack of an effect of speech rate was not a result of low power. Moreover, a post-hoc power analysis indicated that our study was sufficiently powered (>80% confidence) to detect a medium effect size of .30–.50 with a sample of 14 participants.

Previous studies have investigated sentence comprehension in people with dementia using time expansion methods that may have distorted speech at the syllable and word level through mechanically stretching the stimuli. Our study ensured that a natural speech signal remained intact through only manipulating speech rate by extending pauses at phrase and clause boundaries. While a study by Holland and Fletcher (2000), using a comparable speech rate manipulation method, found improved recall and recognition of auditorily presented sentences for healthy older adults, this strategy did not benefit people with dementia in our study.

The rationale for extending pauses at phrase and clause boundaries was to provide participants with AD more time for their working memory to process the verbal input (Wingfield, Tun, Koh, & Rosen, 1999). The presence of a pause has been theorized to provide a perceptual cue for segmentation of the speech signal. Phrase boundaries can be used to break sentences into chunks, thereby making them easier to process. However, the benefits of having more time to process information, even at strategic boundaries, may be cancelled out by the extra demands placed on maintaining and integrating information in working memory. This interpretation has found support in several previous studies (Pashek & DiVenere, 2006; Small, Andersen, et al., 1997; Small, Kemper, et al., 1997; Tomoeda et al., 1990).

Limitations

Similar to many previous studies that have investigated speech rate and sentence comprehension in AD, our study included a relatively small sample size. Although our sample had sufficient power to detect an effect of each experimental variable, it was not necessarily representative of the larger population of persons with AD. Thus, future research with a larger sample size would allow for greater generalization. Another limitation of our study design was that participants’ hearing acuity was not formally assessed (e.g., through audiometric screening). We did ensure that each participant could hear the stimuli; however, it is possible that any participants who had undetected/mild hearing loss had to expend more listening effort to achieve successful word recognition, which would have increased the demands on their language processing system, particularly as sentence complexity increased.

Conclusion

The findings of this study highlight the importance of basing recommendations for speech modifications such as slower speech rate on empirical evidence rather than assumptions and intuitions. Similar to other methods of slowing speech rate, inserting extended pauses at strategic points in the sentence did not have an overall beneficial effect on comprehension. On the other hand, we corroborated previous research that found people with AD experience more difficulty understanding complex sentences. Thus, there is converging evidence supporting the clinical recommendation that care partners use simpler sentences when speaking to someone with AD. However, as with all speech accommodations, one should weigh whether the immediate conversational context and goals warrant reducing demands, or whether it might be an opportunity to help individuals maintain their capacity for engaging in more complex communication.

Acknowledgments

The authors are grateful to the participants who volunteered their time and energy to this study. We would also like to thank the research personnel who contributed to materials preparation, data collection, and/or data analysis.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.