This Article Abstract Full Text (PDF) Services Download to citation manager PubMed PubMed Citation

Sentence Production by Young and Older Adults in Controlled Contexts

Gerontology Center, University of Kansas, Lawrence.

Address correspondence to Susan Kemper, Gerontology Center, 1000 Sunnyside Avenue, Room 3090, University of Kansas, Lawrence, KS 66045. E-mail: Skemper{at}KU.edu

	Abstract

TOP Abstract Methods Results Discussion References

 
In this experiment we compared young and older adults' abilities to produce complex sentences under controlled conditions. We asked participants to memorize sentence stems differing in syntactical complexity and then to produce a complete sentence using the stem. The length, complexity, and content of young adults' responses varied with the syntactical complexity of the stems, whereas older adults' responses did not. These results suggest that working memory processing limitations impose a "ceiling" on older adults' production of complex sentences, limiting their length, complexity, and content.

MOST studies of aging and language production have compared oral or written language samples elicited from young and older adults (Kemper, Thompson, & Marquis, 2001). In contrast, controlled tasks have been widely used to assess psycholinguistic constraints on production (Bock, 1996). These tasks can be used to examine how aging affects production. For example, Kemper, Herman, and Lian (2003) found that older adults took longer, made more errors, and produced less complex sentences than young adults, and older adults' production difficulties increased with the number of words they were given to incorporate in their sentences. One limitation of the previous study is that participants rarely produced the sorts of complex, multiclause constructions that have been the focus of naturalistic studies of speech. We designed the present task to probe participants' abilities to produce sentences involving complex constructions under controlled conditions.

In our task we modified the sentence generation task used by Kemper and colleagues (2003) and combined it with a task used by Ferreira (1991) to induce participants to generate complex sentences. Ferreira required participants to memorize a sentence and then repeat it when cued. The latency to respond was a function of the syntactic complexity of the sentence. We revised this procedure by requiring participants to memorize sentence fragments or stems, not complete sentences; we also revised it to directly examine sentence planning processes by requiring the participants to complete the stems. The stems varied; some were main clauses ending in that, what, or who and participants were required to complete the embedded clause; others began with that, what, or whom clauses and participants were required to supply a main clause. On some trials, additional words were presented after the stem and participants were instructed to use these words in their responses. We analyzed memorization times, response latencies, and response errors; in addition, we scored sentence completions for length, propositional content, and grammatical complexity.

	METHODS

TOP Abstract Methods Results Discussion References

 
Participants
Thirty-four young adults, 18 to 28 years of age, and 39 older adults, 70 to 80 years of age, participated in the task. The young adults were recruited by solicitations on campus. The older adults were recruited from a registry of previous research participants. All participants were paid a modest honorarium. Participants were required to produce at least four valid responses out of six trials in each experimental condition. We excluded data from 9 young adults and 9 older adults for excessive errors. The final group of participants consisted of 24 young adults (M = 19.7, SD = 1.58) and 24 older adults (M = 72.6, SD = 2.31). We screened all participants for hearing acuity, and we excluded those with clinically significant hearing loss from participation. A hearing loss was defined as (a) a hearing loss of greater than 40 dB at 500, 1000, 2000, or 4000 Hz according to pure tone audiometrics or (b) self-report of six or more problems on the Hearing Handicap Inventory (Ventry & Weinstein, 1982). There was 1 young adult and there were 6 older adults whom we excluded for hearing loss. We screened all older adults for possible dementia with the Short Portable Mental Status Questionnaire (Pfeiffer, 1975). We administered the Digits Forward, Digits Backward, and Digit Symbol tests (Wechsler, 1958), the Daneman and Carpenter (1980) reading span test, the Shipley (1940) vocabulary test, and two versions of a Stroop test to all participants. The Stroop tests required participants to name the color of blocks of Xs printed in colored inks or to name the color of color words printed in contrasting colored inks, such as RED printed in blue ink; we computed an interference score. Table 1 summarizes these comparisons of young and older adults. An alpha level of =.05 was set for these and all subsequent t and F tests.

In addition, a set of 27 nouns referring to human characters, such as dentist, butcher, and waiter, and 27 nouns referring to locations, such as kitchen, office, and store, were also selected; 0, 1 (a character or a location), or 2 (a character and a location) were presented after the stem, and the participants were instructed to use these words in their responses. All were common nouns (10 or more occurrences per million words).

Procedure
We used EPRIME (Schneider, Eschman, & Zuccolotto, 2002) to present the stimuli and collect responses. We counterbalanced the stems for presentation such that each participant was tested on 18 sentence-initial and 18 sentence-final stems, but on only one item from each pair of stems. Stems were presented with zero, one, or two additional words to be incorporated in the sentence. The additional words were randomly selected for presentation. Each trial consisted of a fixation point presented for 2 s followed by the presentation of a stem. The participant was instructed to memorize the stem, and told that he or she would be required to produce a sentence beginning with the stem. When the participant had memorized the stem, the participant pressed a response key. The time to memorize the stem was automatically recorded. When the participant pressed the response key, zero, one, or two additional words would appear along with a cue to respond. If no additional words were presented, the response cue was presented alone. The participants were instructed to "produce a sentence, as quickly as possible, using the stem and any additional words presented on the computer screen." They were further reminded that their sentence should begin with the stem, and they were instructed to "use all of the words presented" and encouraged to "add other words to make a complete, grammatical sentence." The words remained on the screen until the participant spoke into a microphone. As soon as a response box connected to the microphone detected a vocal response, the words were removed from the computer screen. Response latencies were recorded from the onset of the response cue to the onset of the participant's response. The participant's response was audiorecorded and later transcribed.

Coding
We initially classified each response as a response error or a valid response. We subcategorized response errors as (a) nonfluent responses with lexical or nonlexical fillers, or false starts; (b) responses that were anomalous or meaningless; (c) sentence fragments; and (d) memory errors including incorrect or partial recall of the stem or stimulus words. Multiple errors could occur on a single trial.

We coded valid responses for propositional density (PDensity, computed as the number of propositions expressed in the sentence divided by the number of additional words used in the sentence times 10), grammatical complexity in terms of developmental level (DLevel) and developmental sentence scoring (DSS), and length in words (the number of additional words used in the sentence), using procedures described by Kemper and colleagues (2003). PDensity is a measure of the amount of information contained in a response, relative to the number of words in the response; DLevel and DSS are measures of grammatical complexity. DLevel assesses the overall complexity of the response whereas DSS assesses the complexity of pronouns, main and embedded verbs, and other sentence constituents. Coded examples are given in Table 2. We assessed intercoder reliability for each level of coding. Reliability averaged better than 90% for all levels of coding.

	RESULTS

TOP Abstract Methods Results Discussion References

PDensity
The overall main effect of age group was significant for PDensity, F(1, 28) = 157.57, p <.01, ² =.65; PDensity differed as a function of the number of additional words, F(2, 27) = 8.55, p <.01, ² =.39, and the Additional words x Age group interaction was significant, F(2, 27) = 5.51, p =.01, ² =.21. Young adults produced more propositions as the number of additional words increased, increasing from PDensity = 4.2 (SD =.2) with zero additional words to PDensity = 6.4 (SD =.3) with two additional words; older adults limited their sentences to an average of PDensity = 4.4 (SD =.3) regardless of the number of additional words. Further, PDensity varied with locus of embedding, F(1, 28) = 6.41, p =.02, ² =.19, and the Locus of embedding x Age group interaction was significant, F(1, 28) = 4.91, p =.02, ² =.27. Young adults produced more propositions for right-branching stems (PDensity = 6.5, SD =.2) than for left-branching stems (PDensity = 4.3, SD =.2), whereas older adults averaged PDensity = 4.4 (SD =.3) regardless of the locus of embedding.

DLevel
The overall main effect of age group was not significant for DLevel, F(1, 28) < 1.0. DLevel did not differ overall as a function of the number of additional words, F(1, 28) < 1.0; however, the Age group x Additional words interaction was significant, F(2, 27) = 4.79, p <.02, ² =.26. For young adults, DLevel scores increased from an average of 4.7 (SD =.2) for zero additional words to an average of 5.2 (SD =.2) for two additional words. DLevel scores for older adults did not vary with the number of additional words (M = 3.7, SD =.3). DLevel scores also varied with locus of embedding, F(1, 27) = 758.336, p <.01, ² =.96. As expected, DLevel scores for left-branching stems were systematically higher (M = 5.0, SD =.1) than those for right-branching stems (M = 2.1, SD =.28), reflecting the scoring system.

DSS
The overall main effect of age group was significant for DSS, F(1, 28) = 21.54, p <.01, ² =.44. DSS varied as a function of the number of additional words, F(2, 27) = 9.64, p <.01, ² =.42, and the Additional words x Age group interaction was significant, F(2, 27) = 10.70, p <.01, ² =.44. DSS scores for young adults increased monotonically with the number of additional words from 7.3 (SD =.6) points per sentence with zero additional words to 13.5 (SD =.8) points per sentence with two additional words. DSS scores for older adults did not vary with the number of additional words, averaging 7.2 (SD =.6) points per sentence. In addition, DSS scores varied with the locus of embedding, F(1, 28) = 10.27, p <.01, ² =.27, and the Locus of embedding x Age group interaction was significant, F(1, 28) = 19.89, p <.01, ² =.59. DSS scores for young adults were higher for right-branching stems (M = 11.4, SD =.5) than for left-branching stems (M = 9.3, SD =.5), whereas DSS scores for older adults did not vary with the locus of embedding (M = 7.2, SD =.6).

Memorization time
The latency to memorize the stem did not vary with age group, locus of embedding, or the number of additional words. Participants required an average of 3.2 s (SD =.2) to memorize the stems.

Response latency
The latency to produce a fluent sentence using the stem and all additional words did vary with age group, F(1, 28) = 25.04, p <.01, ² =.48. Response latency increased with the number of additional words, F(2, 27) = 12.67, p <.01, ² =.49. Young and older adults' latencies increased with the number of additional words; 0 = 3.2 s (SD =.3), 1 = 3.6 (SD =.2), 2 = 4.4 (SD =.3). What was more interesting was that the Age group x Locus of embedding interaction was significant, F(1, 28) = 7.73, p <.01, ² =.48. Older adults (M = 11.38, SD =.49) responded more slowly than young adults (M = 3.8 s, SD =.2), and older adults responded more slowly to left-branching stems (M = 4.4, SD =.3) than to right-branching stems (M = 13.2, SD =.3). Young adults' response latencies did not vary with locus of embedding.

Response Errors
Young adults made one or more errors on 10% of the trials, whereas older adults made one or more errors on 26% of the trials.

Nonfluent responses
Both main effects for age group and locus of embedding as well as the Age group x Locus of embedding interaction, F(1, 28) = 15.39, p <.01, ² =.36, were significant. Older adults (M = 16.0%, SD = 4.5) produced more nonfluent responses than young adults (M = 2.0%, SD = 1.5), and nonfluent responses by older adults were more common in response to left-branching stems (M = 20.3, SD = 5.5) than in response to right-branching stems (M = 11.4, SD =.5).

Sentence fragments
The main effect of age group was significant, F(1, 28) = 4.44, p =.05, ² =.14, as was the Age group x Locus of embedding interaction, F(1, 28) = 52.56, p <.01, ² =.65. Older adults produced more sentence fragments (M = 15.5%, SD = 4.1) than young adults (M = 13.0%, SD = 2.4). More sentence fragments occurred in response to left-branching stems (M = 20.1%, SD = 7.9) than in response to right-branching stems (M = 8.5%, SD = 3.0).

Anomalous sentences
The main effects for locus of embedding and age group were significant, as was the Age group x Locus of embedding interaction, F(1, 28) = 10.72, p <.01, ² =.28. More anomalous sentences were produced in response to left-branching stems (M = 10.4%, SD = 5.2) than in response to right-branching stems (M = 4.2%, SD = 3.3) by older adults; young adults produced few anomalous sentences in response to either type of stem (M = 1.1%, SD = 1.4).

Memory errors
The main effect of locus of embedding was significant, F(1, 28) = 18.59, p <.01, ² =.39. More memory errors were committed in response to left-branching stems (M = 8.6%, SD = 2.8) than in response to right-branching stems (M = 3.4%, SD = 3.7).

	DISCUSSION

TOP Abstract Methods Results Discussion References

 
The length, grammatical complexity, and propositional content of the young adults' valid responses varied with locus of embedding; hence syntactic complexity. The responses were longer, more informative, and more complex when the young adults were given right-branching stems to complete than when they were given left-branching stems to complete, particularly when they were given two or three additional words to use in their response. This finding is consistent with theoretical arguments and experimental demonstrations of the asymmetries between left- and right-branching constructions (Gibson, 1998). Left-branching constructions impose a greater burden on working memory during production because they require the speaker to anticipant and plan for the main clause while producing the embedded clause.

The length, content, and complexity of older adults' responses did not vary with locus of embedding. Indeed, the older adults' responses appeared to be limited to 6.6 additional words, PDensity = 4.4, DLevel = 3.7, and 7.2 DSS points regardless of the type of stem or the number of additional words. Their response latencies were slower for left-branching than right-branching stems and increased when they were given 2 or 3 additional words to incorporate into their response. Whereas young adults made few errors, older adults produced many nonfluent responses, sentence fragments, and anomalous sentences; further, older adults' errors were more frequent in response to left-branching stems than in response to right-branching stems. This pattern of results suggests that the increased memory load imposed by the left-branching stems and by the additional words impaired older adults' ability to produce valid responses.

The present study, like that of Kemper and colleagues (2003), suggests that there is a "ceiling" on older adult's speech at DLevel = 3.7 and PDensity = 4.4 because the complexity and content of older adults' sentences do not rise above these averages when they are asked to complete sentence stems, even left-branching stems. These results also confirm earlier findings based on the analysis of oral and written language samples (Kemper et al., 2001) in showing that older adults produce less complex sentences than young adults. Processing limitations, arising from reduced working memory, inhibitory deficits, or slowed processing speed, may impose this ceiling by limiting older adults' abilities to construct complex, informative sentences, even during controlled production tasks.

	Acknowledgments

 
This research was supported by Grant RO1AG09952 from the National Institute on Aging. We thank Jennifer Nartowitz for her assistance.

	Footnotes

 
Decision Editor: Margie E. Lachman, PhD

Received for publication October 9, 2003. Accepted for publication April 15, 2004.

	References

TOP Abstract Methods Results Discussion References

 

• Bock, K. (1996). Language production: Methods and methodologies. Psychonomic Bulletin & Review, 3,395-421.
• Daneman, M., Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Ability, 19,450-466.
• Ferreira, F. (1991). Effects of length and syntactic complexity on initiation times for prepared utterances. Journal of Memory and Language, 30,210-233.
• Francis, W. N., Kucera, H. (1982). Frequency analysis of English usage: Lexicon and grammar. Boston: Houghton Mifflin.
• Gibson, E. (1998). Syntactic complexity: Locality of syntactic dependencies. Cognition, 68,1-76.[Medline]
• Kemper, S., Herman, R. E., Lian, C. (2003). Age differences in sentence production. Journal of Gerontology: Psychological Sciences, 58B,P260-P269.
• Kemper, S., Thompson, M., Marquis, J. (2001). Longitudinal change in language production: Effects of aging and dementia on grammatical complexity and propositional content. Psychology and Aging, 16,600-614.[Medline]
• Pfeiffer, E. (1975). A short portable mental status questionnaire for the assessment of organic brain deficit in elderly patients. Journal of the American Geriatrics Society, 23,433-441.[Medline]
• Schneider, W., Eschman, A., Zuccolotto, A. (2002). E-prime user's Guide. Pittsburgh, PA: Psychology Software Tools Inc.
• Shipley, W. C. (1940). A self-administered scale for measuring intellectual impairment and deterioration. Journal of Psychology, 9,371-377.
• Ventry, I. M., Weinstein, B. E. (1982). The Hearing Handicap Inventory for the Elderly: A new tool. Ear and Hearing, 3,128-134.[Medline]
• Wechsler, D. (1958). The measurement and appraisal of adult intelligence. Baltimore: Williams & Wilkins.

This Article Abstract Full Text (PDF) Services Download to citation manager PubMed PubMed Citation