An Experimental Paradigm for Measuring the Effects of Ageing on Sentence Processing

Summary

Here, we present a protocol to examine the age-related decline in sentence processing using a maze task that allowed us to localize the processing difficulty at each word of the sentences during reading.

Abstract

Previous studies have found that older adults have greater difficulties in processing syntactically complex sentences than younger adults. However, the exact regions where the difficulties arise have not been fully identified. In this study, a maze task was implemented to investigate how older adults and younger adults differentially processed two types of sentences with different levels of syntactic complexity, namely subject relative clauses and object relative clauses. Participants were asked to choose between two alternatives at each segment of the sentences. The task required participants to engage in a strictly incremental mode of processing. The reading times for each segment were recorded, allowing the quantification of the difficulty of sentence reading. The task allowed us to identify the exact locations of the processing difficulty and thus facilitate a more accurate assessment of the age-related decline in sentence processing. The results indicate that the effect of ageing was found mainly at the head nouns, but not at other regions of the sentences, a finding which suggests that the maze task is an effective method to identify the exact location of the ageing effect on sentence processing. The implications of this experimental paradigm for investigating the effect of ageing on sentence processing are discussed.

Introduction

Sentence processing is an essential process for human beings to comprehend sentences in natural languages. It requires integrating incoming words into the existing sentential structures and establishes relationships between words in the sentence in a way which conforms to the grammatical constraints. As sentence processing is a resource-demanding process, older adults tend to encounter difficulties due to the age-related decline in working memory. Previous studies have found that older adults have great difficulties in comprehending and processing sentences with complex syntactic structures such as relative clauses (RCs): "The reporter whom the senator attacked admitted error"¹. Most previous studies focused on RCs to investigate the effect of ageing on sentence processing. Compared with younger adults, older adults have lower accuracy and efficiency in RC comprehension. However, despite the extensive studies on RC processing by older adults, it is still not clear which parts of RCs are more difficult to process for older adults as they read sentences word by word. Zurif et al. (1995) found that older adults had greater difficulty at the RC verbs ("attacked") as they were less efficient to assign the patient roles to the objects of the verbs ("reporter")². However, some studies suggest that older adults did not perform differently from the younger controls at any particular region. Rather the difficulties occurred when participants finished reading the entire sentences³. Still, several other studies found that older adults performed less efficiently than younger adults at the RC subjects ("the senator") and verbs ("attacked")^4,5. In other words, the precise locations of the processing difficulties for older adults have not yet been clearly identified.

Sentence processing is an incremental process that requires sufficient cognitive resources. According to the Dependency Locality Theory (DLT)^6,7, the processing difficulties or costs in sentence processing are determined by the distance between the incoming word and the dependent word to which it is attached and thus the costs may vary at different regions in the sentence. The processing costs increase when the distance between the word and its dependent is longer. Therefore, due to the decline of working memory in older adults, they may encounter greater difficulties than younger adults at the words that are located far away from their dependents and are cognitively more costly, such as the verbs ('attacked') of object RCs. Although the DLT was supported by prior work^8,9, it has also been challenged by a number of studies¹⁰. The study of the precise locations of the processing difficulties in older and younger adults can test the validity of the DLT and help us to understand the relationships between sentence processing and cognitive resources. It can also provide a fine-grained picture of how ageing might influence online sentence processing.

Previous studies on the effect of ageing on sentence processing during reading have mostly adopted self-paced reading^5,11, self-paced listening⁴ and cross-modal priming². As an online measure, cross-modal priming tasks can be used to probe the activation of syntactic or lexical information in sentence processing. However, this task can only reveal the processing performance at a certain word or phrase and thus it does not allow us to measure participants' performance at different segments of the sentence to identify where their processing difficulties are mainly located. Actually, so far there has been only one study which used the cross-modal priming task to examine the age-related decline in sentence processing. Zurif et al. (1995) discovered that older adults encountered greater difficulty in processing English object RCs at the verb regions when the distance between gaps and fillers was longer². The study suggested that there was an age-related decline in RC processing as a result of the lack of working memory resources². However, due to methodological constraints, the study failed to provide information regarding the processing difficulty at other regions of RCs.

As one of the most frequently used approaches in the research into the effect of ageing on sentence processing, self-paced reading (SPR) tasks require participants to read sentences word by word by pressing a specific button. Participants indicate with a press of the button that they have finished reading the word on the screen and then the next word appear, replacing the previous one. The participants repeat this procedure until all words of the sentence are presented, at which point they are required to answer a comprehension question related to the meaning of the sentences they have just read. Reading times (RTs) are recorded and used to measure the processing difficulties for each sentence segment and responses to the sentence-final questions are used to measure the accuracy of sentence comprehension. SPR tasks differ significantly from natural reading because only one word is allowed to be displayed at a time and regressive eye movements are not possible¹². The unnatural aspect of SPR tasks has the advantage of reducing the number of strategies used by participants and thus the interference of processing strategies on participants' performance can be minimized. Therefore, SPR tasks are helpful for researchers to identify the processing difficulties related to processing every word in the sentences. The major problem with SPR tasks is that they cannot ensure accurate measurement of RTs as participants may delay button-pushing to memorize the sentences or speed up button-pushing to buffer each region for later construction in the sentences¹².

Apart from SPR tasks, eye-tracking has also been frequently used in studies on sentence processing^13,14,15. In eye-tracking experiments, full sentences are presented to participants and they are asked to read them naturally. The duration and location of eye fixations and regressions are recorded. The advantage of eye-tracking is that it allows for natural reading and puts little constraints on how participants read sentences¹². However, the naturalness of eye-tracking also makes it difficult for researchers to assess the exact processing costs because participants might adopt a wide variety of reading strategies such as skimming or word-by-word reading¹². Therefore, we are not clear whether the fixation location and duration reflect processing difficulties or the application of certain reading strategies.

In this study, we adopted a maze task^16,17, which can solve these problems with SPR tasks, eyetracking and cross-modal priming tasks. A maze task is a more updated reading paradigm and an adaptation of self-paced reading tasks. The task is an approach adopted in psycholinguistic experiments to record RTs as participants read sentences word by word. In the tasks, participants are required to read a sentence by making a choice between two alternatives for each segment of the sentence¹⁶. The procedure of a maze task is demonstrated in Figure 1.

Figure 1: A sample display of the sequential frames in a maze sentence. The figure shows that there are two alternative words in each frame, only one of which can continue the sentence. Please click here to view a larger version of this figure.

If the first word of the sentence is the, which is followed by the pair cat and on, the participants are expected to choose cat by pressing a button to indicate that the word on the left is the grammatical continuation of the sentence. When the next pair of words (to and ate) is presented, the participants are expected to choose ate because it can serve as the verb of the sentence. This procedure is repeated until they arrive at the end of the sentence. In case they make a mistake, the current trial will be terminated, and the next trial starts. An online demonstration of the maze task provided by Kenneth Forster can be found at http://www.u.arizona.edu/~kforster/MAZE.

The maze task is a highly constrained experimental paradigm in which participants are not permitted to look ahead or back to the previous section of the sentence^18,19. Each word must be integrated with the existing structure before participants can proceed to read the following word. This method requires participants to process the sentences very carefully and accurately and it reduces the processing strategies available to participants. It allows us to make an objective measurement of RTs in online sentence comprehension. Moreover, the maze task is more sensitive to the processing costs associated with integrating words into sentences^12,16, as it places unusual constraints on the processing strategies available to the participants. In this sense, the maze task is particularly suitable for the studies of sentence processing by older adults. Due to their richer language experience, older adults tend to adopt various compensation strategies such as prediction to make their performance comparable with younger adults⁴. These strategies make the age differences in sentence processing less noticeable. In a maze task, if participants make a response prematurely, they will be forced to respond randomly at the next frame and thus, they are most likely to make an error. Therefore, the task does not allow participants to make predictions and the effect measured by the task is unanticipated in nature. By reducing the number of sentence processing strategies available to older adults^12,16, the maze task allows for more objective assessment of the processing difficulties of older adults during sentence reading.

The maze task was utilized in our study to pinpoint the processing difficulties for older adults in sentence comprehension and to localize the effect of ageing. As older adults generally have slower speed of processing and longer reading times than younger adults, most previous studies evaluated the effect of ageing by measuring the extent to which the effect of syntactic complexity or processing cost is modulated by age, a common practice in the research into the effect of ageing on sentence processing. Most studies chose to use object relative clauses (ORCs) and subject relative clauses (SRCs) as the experimental stimuli as they represent syntactically more complex structures and less complex structures, as exemplified below. Therefore, in this study, we also examined the effect of ageing by evaluating the influence of age on the processing difference between SRCs and ORCs.

a. Subject relative clause, SRC: The girl who chased the boy was wearing a hat.
b. Object relative clause, ORC: The girl whom the boy chased was wearing a hat.

According to Gibson's Dependency Locality Theory^6,7, sentence (b) is more difficult to process than sentence (a) as there are higher storage costs and integration costs involved in (b) to assign patient role to the head noun girl. By comparing participants' performance in processing SRCs and ORCs, we are able to find out how they process sentences with different cognitive burdens. However, previous studies are divided with regard to their interpretation of the effect of syntactic complexity. Some studies maintained that if the effect of sentence complexity is exacerbated in older adults, it suggests that older adults are more sensitive to processing costs than younger adults^3,20. Therefore, the amplified effect of sentence complexity is regarded as the evidence for age-related decline in sentence processing. Other studies, however, proposed that an effect of sentence complexity indicated that older adults are capable of allocating cognitive recourses to sentence processing and thus it should be viewed as a sign for the preservation of sentence processing ability in older adults²¹. The two lines of research are divergent in their interpretation of the effect of sentence complexity mainly because the experimental methods used in these studies such as SPR or cross-modal priming cannot tell us whether older adults actually use their working memory resources to integrate the words into the previous sections of the sentence or they cannot integrate the words due to the lack of working memory recourses. The maze task can provide a perfect solution to this problem because participants are forced to integrate the words into the previous structures in this task. Therefore, it is safe to say that the RTs obtained from the maze task represent the actual processing costs or processing difficulty in sentence processing. The maze task provides an effective means to explore sentence processing by older adults. Qiao, Shen and Forster (2002) used the maze task to examine the processing of Mandarin RCs and discovered that this technique was highly sensitive to sentence complexity effect in the processing of Mandarin subject- and object-extracted RCs¹⁹. The study indicates that the maze task is applicable to the research which is intended to localize the experimental effect in Mandarin sentence processing or lexical processing.

Protocol

The experiment was approved by the Ethics Committee of Beijing Foreign Studies University and it complied with the guideline for experiments with human subjects. All participants in the experiment provided written informed consent.

1. Stimuli Construction

1. Construct experimental stimuli on the basis of the specific experimental questions. This protocol is intended to explore the effect of ageing on online sentence processing during reading, and as such, the visual stimuli used in the experiment are two types of RCs, namely SRCs and ORCs, which are the structures most frequently used in the studies of sentence processing by older adults.
2. Prepare at least 24 experimental sentences with twelve sentences in each condition and at least 24 fillers that are of various length and structures. The number of fillers can be increased to up to three times of the experimental sentences. The experimental sentences should be counterbalanced across the two conditions.
   NOTE: As our study focused on the processing of Mandarin sentences, all the sentences used in the experiment were in Mandarin Chinese. However, the protocol reported here may be also applicable to other languages.
3. Plausibility rating: Invite at least 30 participants to rate the semantic plausibility of the experimental sentences using a five-point Likert scale with scores from ''1'' (the least plausible) to ''5'' (the most plausible). Make sure the group is similar to the participants in age, gender ratio and education. Make sure they do not take part in the subsequent experiment. Ensure the sentences used in the study are all semantically plausible (>3) and semantic plausibility does not differ significantly between different experimental conditions.
   NOTE: In our study, 15 older adults and 15 younger adults were invited to participate in the plausibility rating (M_age = 43.6, SD = 24.1; Female 16, Male 14).
4. For each word in the experimental sentences, provide an ungrammatical alternative. In this way, prepare a pair of words for each segment of the sentences, with one of them being the grammatical continuation of the sentence and the other is not. Match the two alternatives in lexical frequency. Participants are expected to choose between the grammatically correct word and the grammatically incorrect one (See Figure 1 for sample sentences in English). An experimental sentence used in this study is shown in Figure 2.
5. Randomization and design. Randomize the sentences and the positions of the two alternatives in the sentences before presenting them to participants. Find the Selection tab of the Property Page in the stimulus presentation software (see Table of Materials) and set the selection method to Random.
   1. Divide the experiment into blocks with each lasting for no more than 10 min.
      NOTE: In the study, the experiment consisted of two blocks with 24 sentences in each block. Each block lasted for about 5 min.
   2. Allow participants to have a break between two blocks. Breaks for about 5 min can be helpful for younger adults. Allow longer breaks for older adults to ensure they can recover from fatigue.

Figure 2: Example of two experimental sentences. This figure shows that participants were presented two Chinese words in each frame and were required to choose the one which can grammatically continue the sentence. The words on the frame are Chinese characters with the English equivalents provided in parentheses. Please click here to view a larger version of this figure.

2. Participant Screening and Preparation for the Experiment

1. Recruit participants who are native speakers of the language tested. Ensure participants have normal or corrected-to-normal vision. Invite at least 30 younger adults and 30 older adults. Match the two age groups in years of education and gender ratio.
2. Ask participants whether they have had neurological, psychiatric or communication disorders before. Exclude the ones who report any history of these diseases.
3. To ensure older participants recruited are cognitively healthy, have them complete the Global Deterioration Scale²² before the experiment as a way of screening. Exclude the participants who score above level two.
4. Inform the participants that they should be free from exhaustion, hunger, illness or other conditions that make them uncomfortable on the day of the experiment.
5. Invite the participants individually or in groups to the laboratory room.

3. Procedure

1. Take participants to the laboratory room and instruct them to sit down at the computer workplaces.
2. Ask the participants to read and sign the written informed consent forms. Exclude the participants who cannot provide informed consent for any reason.
3. Provide verbal instructions to the participants to ensure they are clear about the requirements of the task. Ask participants to choose between the two words or phrases on the computer screen to form a grammatical continuation of the sentences. Inform them that their responses will be timed, so they are expected to respond as quickly and as accurately as possible.
4. Provide the written instruction to the participants and ask them to read it carefully. Answer any questions participants may have about the instructions. Present the following message on paper or on computer screen in participants' native language:
   "In this experiment, you will read a number of Chinese sentences. However, the sentences will be presented one word at a time. On each screen, two words will be presented and you need to choose the word which can grammatically continue the sentence as quickly as possible. If you choose the word on your left hand side, please press the F button on the keyboard and if you choose the word on the right, please press J. Please respond as quickly as possible. If you fail to respond in ten seconds, the next trial begins. If you have no further questions, press any key to start practice."
5. Ask the participants to complete a small number of practice questions to ensure they understand how to perform the task. Provide participants with feedbacks after they have made their choices regarding which of the two phrases is the grammatical continuation of the sentence. To minimize the disturbance to the natural reading process, only provide feedback when incorrect answers are detected. The feedbacks are short phrases ("incorrect") in the participants' native language, which are visually presented on the computer screen.
6. When participants complete the practice questions, ask them to make a choice as to whether they would like to redo the practice. If they press the button p on the keyboard, the practice session starts all over again and if they press the space bar, the practice session is over and the experimental session begins. Allow them to redo the practice as many times as possible until they are familiar with the procedures of the task. Provide them assistance whenever they encounter any difficulties during the practice session.
7. Allow the participants to perform the required experimental task when they are clear about the experimental procedures.
8. Present the items on the computer screen as black letters on a white background.
9. Have participants read the items and for each segment, choose between the two words placed side by side on the screen with a press of the button on the keyboard (F button for the word on the left and J button for the word on the right).
10. If participants choose the correct answer, present the next pair of words. When they choose a wrong answer, have the sentence terminate and have the next one appear.
11. Administer the digital span task from the Wechsler Memory Scale²³ to assess participants' working memory span. Test participants individually.
12. Ask participants to listen to a sequence of digits (1-9) that are presented auditorily at the rate of one digit per second.
13. Ask participants to repeat the digit sequence in the same order.
14. Start from short strings and increase the string length one digit at a time. Stop the test when participants make errors at two trials at a given length.
15. Record participants' responses on the record form. Use a digital recording device to record participants' responses.
16. Ask participants to complete the questionnaires about their background information, such as age, gender and years of education. Prepare participants' payment during the questionnaire.
17. Provide participants monetary compensation for their participation.
    NOTE: In our experiment, each participant was paid 20 RMB (about USD$3) for their participation. The amount of payment may vary in different experiments. Other approaches can also be used to motivate participants such as rewards.

4. Data Analysis

1. Obtain the RTs (ms) for each word in the sentences from the output files of the presentation software.
2. Exclude the RTs for the trials that are responded incorrectly from data analysis. In order to reduce the influence of outliners on the results, exclude the RTs above three standard deviations from the means for each participant in each condition.
3. Analyze the RTs for the regions of interest in the sentences, using repeated measures ANOVA with RTs as the dependent variable and age group (old, young), type of sentence (SRC, ORC) and region (1-6) as independent variables.

Representative Results

This study is intended to examine the effect of ageing on online sentence processing using a maze task. The RTs for each segment in the sentences are used to indicate the processing difficulty. In this study, we explored how a group of older adults (M_age = 65.2, SD = 3.04, Range = 59 -74) and younger adults (M_age = 19.1, SD = 1.04, Range = 18-23) differentially processed Mandarin SRCs and ORCs. The two age groups were matched in education (M_old = 12.8, SD = 3.21; M_young = 13.2, SD = 0.75; p = .620) and gender ratio (Old: 14 female; Young: 15 female; χ² = .067, p = .796). Specifically, the study aimed to explore the overall differences between younger and older adults in processing Mandarin SRCs and ORCs as well as the exact locations of such differences. It was hypothesized that due to the decline in working memory, older adults are more sensitive than younger adults to the experimental manipulation of syntactic complexity and the processing costs associated with integrating words into the previous sentential segments. Therefore, a stronger effect of syntactic complexity or sentence type may be found in older adults than in younger adults at the regions which require more processing resources such as RC head nouns.

The regions of interest in the experimental sentences include the RC noun (zuojia 'writer'), the RC verb (aimu 'love'), the RC marker (de), the head noun (yanyuan 'actor') and main clause verb (xihuan 'like'). The mean RTs for each region are presented in Table 1 and graphically summarized in Figure 3.

Group	Type	Region 1		Region 2		Region 3		Region 4		Region 5		Region 6		Region 7
		Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD
Old	SRC	1589	1044	2768	1197	2119	961	1467	826	2038	1282	2035	1123	1986	982
	ORC	1354	834	2322	941	2070	870	1436	694	1828	878	2078	1024	1937	938
Young	SRC	789	434	1558	677	1020	437	659	223	812	341	848	331	928	428
	ORC	765	519	1155	494	1071	388	713	245	816	333	972	464	972	465

Table 1: Mean reading times by group and region. The table presents the mean reading times and the standard deviations for each age group in each region.

Figure 3: Mean reading times (RTs) by group and region. The figure displays the average reading times for older adults and younger adults for each region in the sentence. The error bars represent the standard deviation of the data. Please click here to view a larger version of this figure.

Analysis was made to examine the effect of randomization order on the participants' performance and found that the effect of randomization was not significant. ANOVA was performed with group (old, young), type (SRC, ORC) and region (1-6) as predictors and RTs as the dependent variable. Group is the between-subjects variable. Type and region are the within-subjects variables. The results revealed a significant main effect of group (F₁ (1, 58) = 171.25, p < 0.001, η²_p = 0.71, F₂ (1, 23) = 273.13, p < 0.001, η²_p = 0.52), a significant main effect of type (F₁ (1, 58) = 14.9, p < 0.001, η²_p = 0.10, F₂ (1, 23) = 12.78, p < 0.001, η²_p = 0.03), a significant interaction between group and region (F₁ (1, 58) = 100.15, p < .001, η²_p = 0.64, F₂ (1, 23) = 118.12, p < .001, η²_p = 0.44), and a significant interaction between type and region (F₁ (1, 58) =18.43, p < 0.001, η²_p = 0.11, F₂ (1, 23) = 28.43, p < 0.001, η²_p = 0.02). The three-way interaction between group, type and region was significant (F₁ (1, 58) = 5.13, p < 0.05, η²_p = 0.05, F₂ (1, 23) = 2.71, p < 0.05, η²_p = 0.01).

To facilitate the interpretation of these results, we analyzed the data for each word using 2 (sentence type) x 2 (group) mixed ANOVAs by items and participants. In the SRC verb (aimu 'love')/ORC noun (zuojia 'writer'), there was a significant main effect of group (F₁ (1, 58) = 122.93, p < 0.001, η²_p = 0.10, F₂ (1, 23) = 337.06, p < 0.001, η²_p = 0.53), a significant main effect of RC type (F₁ (1, 58) = 34.82, p < 0.001, η²_p = 0.03, F₂ (1, 23) = 49.98, p < 0.001, η²_p = 0.07). Older adults had longer RTs than younger adults and the RTs for SRCs were longer than those for ORCs.

In the SRC noun (zuojia 'writer')/ORC verb (aimu 'love'), there was a significant main effect of group (F₁ (1, 58) = 174.98, p < 0.001, η²_p = 0.15, F₂ (1, 23) = 377.25, p < 0.001, η²_p = 0.43). The effect of sentence type was significant by items (F₂ (1, 23) = 43.98, p < 0.01, η²_p = 0.02). Older adults read this segment significantly more slowly than younger adults. No other significant interaction effect was found.

In the RC marker de, there was a significant main effect of group (F₁ (1, 58) = 177.66, p < 0.001, η²_p = 0.15, F₂ (1, 23) = 489.25, p < 0.001, η²_p = 0.45). Younger adults read the RC marker faster than older adults.

At the head nouns (yanyuan 'actor'), there was a main significant effect of group (F₁ (1, 58) = 371.07, p < 0.001, η²_p = 0.32, F₂ (1, 23) = 53.21, p < 0.001, η²_p = 0.04), a significant main effect of RC type (F₁ (1, 58) = 13.28, p < 0.001, η²_p = 0.02, F₂ (1, 23) = 346.30, p < 0.001, η²_p = 0.34), and a significant interaction effect between group and RC type (F₁ (1, 58) = 5.14, p < 0.05, η²_p = 0.01, F₂ (1, 23) = 4.25, p < 0.05, η²_p = 0.01). Pair-wise comparison shows that for older adults, the SRCs were more difficult to process than ORCs; for younger adults, however, there was no significant difference between SRCs and ORCs. This finding suggests that older adults were more sensitive to the experimental manipulation of sentence complexity or processing costs and they had greater difficulties than younger adults in processing syntactically more complex SRCs.

At the main verb (xihuan 'like'), a significant main effect of group was found (F₁ (1, 58) = 174.99, p < 0.001, η²_p = 0.15, F₂ (1, 23) = 124.02, p < 0.001, η²_p = 0.12). Younger adults performed faster than older adults.

To further explore whether the observed effect at head nouns was relevant to working memory, we performed a correlation analysis between working memory span and RT differences between SRCs and ORCs. The results revealed a significant negative relation (r = -0.41, p < 0.05), which suggests that participants with higher working memory span were less affected by syntactic complexity. As younger adults (M = 16.40, SD = 2.78) had larger working memory span than older adults (M = 12.17, SD = 2), they tended to be less affected by the manipulation of syntactic complexity.

Discussion

In this study, a maze task was used to find the exact location for the effect of ageing on sentence processing. The study examined how older adults and younger adults differentially processed Chinese SRCs and ORCs. In the task, participants were instructed to read a sentence by choosing between two alternatives, only one of which was the grammatical continuation of the sentence. They were required to make the choice as quickly and as accurately as possible. If participants chose the correct alternatives in the sequence, the selected words constituted a grammatical sentence. If they chose the wrong alternative, the current trial was terminated and the next trial began. The results suggest that the age-related difference was found only in the head nouns (yanyuan, 'actor') of the sentences. It was at the head nouns that the older adults encountered greater difficulties than younger adults in sentence processing. This might be because the head nouns are the areas where the parser starts to establish connections between the gap and the filler and integrate them to complete the assignment of semantic roles. According to the DLT, the processing costs of RCs are determined by the distance between the filler and the gap to which it is attached. In Chinese SRCs, the distance between the head noun and the gap is longer than that in ORCs, so it takes more processing resources to associate the head noun with the gap. Due to the working memory decline in older adults, the increased processing costs at the head nouns in SRCs make it difficult for them to process efficiently. In contrast, younger adults had sufficient working memory resources to process the sentence even if the processing costs increased at the head nouns. Therefore, they tended to be less disrupted by the increase in processing costs at head nouns than older adults. The finding indicates that older adults are more disrupted than younger adults by the increased processing costs during sentence reading. The findings support the DLT. It should be noted that as Chinese RCs are different from postnominal RCs in that they are prenominal, the finding that there was greater processing difference between SRCs and ORCs in older adults cannot be generalized to the languages with postnominal RCs such as English because the processing costs at the head nouns do not differ between SRCs and ORCs in postnominal structures. According to the DLT^6,7, it is at the RC verbs that the processing difference between English SRCs and ORCs is maximal. However, as most previous studies on the processing of postnominal RCs by older adults used SPR tasks, they are not very helpful to reveal the exact location of the processing difficulty. Therefore, further investigation is needed to explore where the processing difficulties are located during the processing of postnominal RCs.

These findings show that the protocol is successful in identifying the exact location of the ageing effect on sentence processing. Compared with SPR tasks or cross-modal priming tasks, the maze task allows for a more fine-grained analysis of the ageing effect on sentence processing as this approach is more sensitive to the differences in processing costs or difficulties in different types of sentences. By requiring the participants to choose the word that can grammatically continue the sentence between two alternative words, the task forces them to integrate the words with the existing structure before proceeding to the next segment. Unlike SPR or cross-modal priming tasks that allow older adults to adopt various processing strategies, which might be potentially disrupting for the identification of their processing difficulties, the maze task reduces the availability of processing strategies and thus contributes to a more objective assessment of processing difficulties. Therefore, the reading times measured by the maze task can accurately reflect the processing difficulties encountered by older adults. The maze task has been applied in prior studies on the processing of both postnominal RCs^12,16 and prenominal RCs¹⁹ and was found to be applicable in both situations. Therefore, the protocol offers an effective means to investigate the effects of ageing on sentence comprehension in languages with prenominal RCs as well as those with postnominal RCs.

The major limitation with this method is that it is highly unnatural, so it is not applicable to mimic the natural process of reading. For the studies that intend to find out how older adults read texts naturally, eye-tracking might be a better choice. However, the maze task is a suitable approach for investigating sentence processing under the highly constrained experimental conditions¹⁸.

Materials

Name	Company	Catalog Number	Comments
Computers	N/A	N/A	Used to present stimuli and record subjects' responses.
E-prime	PST	2.0.8.22	Stimulus presentation software
The Digital Working Memory Span Test	N/A	N/A	Used to assess subjects' working memory span. From Wechsler (1987).
The Global Deterioration Scale (GDS)	N/A	N/A	Used to assess subjects' general cognitive status. From Reisberg, Ferris, de Leon and Crook (1988)