An analysis of error patterns in children's backward digit recall in noise.

The purpose of the study was to determine whether perceptual masking or cognitive processing accounts for a decline in working memory performance in the presence of competing speech. The types and patterns of errors made on the backward digit span in quiet and multitalker babble at -5 dB signal-to-noise ratio (SNR) were analyzed. The errors were classified into two categories: item (if digits that were not presented in a list were repeated) and order (if correct digits were repeated but in an incorrect order). Fifty five children with normal hearing were included. All the children were aged between 7 years and 10 years. Repeated measures of analysis of variance (RM-ANOVA) revealed the main effects for error type and digit span length. In terms of listening condition interaction, it was found that the order errors occurred more frequently than item errors in the degraded listening condition compared to quiet. In addition, children had more difficulty recalling the correct order of intermediate items, supporting strong primacy and recency effects. Decline in children's working memory performance was not primarily related to perceptual difficulties alone. The majority of errors was related to the maintenance of sequential order information, which suggests that reduced performance in competing speech may result from increased cognitive processing demands in noise.

Introduction

Listeners with and without hearing loss have trouble with working memory tasks in the presence of background noise.[123] Reduced working memory performance amid noise may result from perceptual (auditory) difficulties at the encoding stage, increased (cognitive) processing load, or a combination of the two. In fact, Rabbitt[45] describes noise as an extrinsic factor that interferes with a listener's ease of perception and processing. Background noise challenges a listener's ability to perceive and process speech and consequently takes away cognitive resources that are available to a listener for usage on working memory storage and retrieval. This has potential negative consequences for learning in schoolgoing children who encounter degraded acoustic environments on a daily basis (e.g., classroom lectures, small group activities, and multiple-talker discussions).[6] If we can identify the relative contributions of perceptual and cognitive processing factors, we may be able to help improve comprehension abilities under adverse listening conditions. Therefore, research to reveal the individual contributions of perceptual and processing factors involved in listening amid noise for schoolgoing children is warranted.

Working memory

Working memory involves the active manipulation and maintenance of information over a short period of time.[7] It positively relates to a number of high-level cognitive abilities such as language comprehension,[8] reading ability,[9] arithmetic skills,[10] fluid intelligence,[11] learning, and scholastic achievement.[12] Working memory is limited in its capacity[13] and given this limit, extrinsic (e.g., background noise) and intrinsic factors (e.g., hearing loss) contribute to degraded speech, which consumes cognitive resources that can otherwise be used for processing information. Background noise, in particular, contributes to a substantial decline in working memory performance.[1214] For example, children with normal hearing demonstrated worse auditory working memory performance in competing speech at commonly experienced signal-to-noise ratios (SNRs) compared to quiet.[3] The presence of noise impeded working memory processes in real-time and resulted in the inaccurate recall of target items. This finding is consistent with Rabbitt's[5] view that perceptual processing is of the utmost priority under adverse listening conditions; cognitive processing resources are directed toward decoding the target signal in noise and supporting auditory processing.[15] This priority for perceptual processing draws resources away from cognitive processing and storing of other information for later recall, which manifests as effortful listening, greater number of task errors, and loss of information from temporary storage.[101516]

Perceptual abilities in noise

When compared to adults, children with normal hearing perform worse on speech-in-noise tasks.[1117] They require a more favorable SNR for successful communication, especially in the presence of competing speech.[12] For instance, children's word identification thresholds are worse for 12-talker babble than for filtered noise.[18] This difficulty with perceptually segregating a relevant word from competing background of speech is referred to as perceptual masking.[13] Children's increased susceptibility to perceptual masking explains their elevated, masked speech perception thresholds in competing speech.[1319] The effects of competing speech have not only been observed in tone or word identification but also in working memory tasks that involve immediate recall of digits, syllables, and final words of sentences.[1420] In one study, 8-11-year-old children were more affected by irrelevant competing speech as compared to tones.[14] The author suggested that competing speech interfered with the target stimuli because of their shared features (i.e., phonological code). This finding supports the view that children are less able than adults to separate irrelevant background speech and are thus, more susceptible to competing speech-induced disruption. It is unknown as to whether a child's immature attention or peripheral auditory abilities account for the difficulties in the presence of competing speech but the possibility of perceptual masking poses concern for any task administered in competing speech. Thus, it is important to examine the role that perceptual audibility plays in children's working memory in noise performance.

Cognitive processing in noise

Although irrelevant competing speech may lead to perceptual masking, it may also contribute to increased cognitive processing demands. In fact, several studies have reported that listening to speech in degraded acoustical conditions can result in a high cognitive load;[15] “cognitive load” refers to any factor that places excessive demand on the central attentional and mnemonic processes.[21] For example, when a listener's working memory capacity is mostly consumed with separating relevant speech from background noise, the listener has few cognitive resources available to inhibit distraction from irrelevant information. Thus, listening amid background noise has both perceptual and cognitive elements but can largely be a top-down and cognitive resource demanding process.[12223] However, the degree of top-down involvement depends on the difficulty of the listening conditions and the complexity of the task.[2425] An emerging cognitive hearing science model, the Ease of Language Understanding (ELU) model,[2627] describes the role of working memory as implicit (automatic) in quiet or conditions with low levels of distraction (e.g., filtered noise).[28] However, when the input is degraded due to the presence of competing speech, a mismatch arises and explicit working memory resources are called upon. In addition to a mismatch between long-term memory information, subvocal verbal rehearsal (the ability to repeat items in memory before recall), and serial scanning (the rate at which each item is recalled), it may contribute to increased processing demand.[1429] Subvocal rehearsal and serial scanning apply especially to digit span tasks of the working memory. In a group of pediatric cochlear implant users and their age-matched peers with normal hearing, the maintenance of a sequential order of information accounted for the majority of errors on the forward and backward digit span tasks.[30] In contrast, adult listeners committed a greater proportion of simple misidentifications or item errors on forward and backward digit span tasks in degraded listening conditions.[14] The authors explained that developmental factors (i.e., cognitive capacity and efficiency improve with age) and sensory deprivation (i.e., children with profound deafness were included) might account for the disparities in the findings.

In the current study, we investigated the types and patterns of errors made on the backward digit span task in children without sensory deprivation (i.e., hearing loss). The goal of this detailed error analysis was to determine whether perceptual masking or increased cognitive processing demands primarily account for the decline in working memory performance in the presence of noise. We expected that there would be a minimal contribution of perceptual masking (i.e., item errors) and instead, the majority of the errors will be related to increased cognitive processing demands (i.e., maintaining sequential order information while inhibiting competing speech).

Methods

Error analysis data were drawn from four previous experiments. Detailed descriptions of materials, procedures, and levels of performance were reported by Osman and Sullivan.[3]

Participants

A total of 55 children between the age of 7 years and 10 years [mean = 8.9 years, standard deviation (SD) = 1.2] were studied. All the children had hearing thresholds ≤20 dB(HL) bilaterally at octave frequencies between 0.25 kHz and 8 kHz, as determined by a hearing screening. The participants were typically developing monolingual American English speakers with no known neurological disorders, no history of recurrent otitis media, normal vision, and no reported cognitive difficulties or barriers to education as determined by parent interview.

Stimuli

The data of interest were the errors made on the backward digit recall task from the Working Memory Test Battery for Children (WMTB-C)[31] administered in quiet and degraded listening conditions. Data from the forward digit recall task were not included because it involved only temporary storage and recall of auditory items. Our motivation for an in-depth analysis of the backward digit recall task data was twofold:

The backward digit recall is a comprehensive measure of working memory: It involves both storage and processing aspects of the working memory and has little linguistic influence,

The listening conditions are ecologically valid: Multitalker babble at −5 dB SNR is representative of everyday acoustically degraded situations encountered by schoolgoing children.[32]

The four-talker babble consisted of one male and three female speakers and was supplied with the QuickSIN Speech-in-Noise test (version 1.3, Etymotic Research Elk Grove Village, IL USA).

Procedure

The participants were tested individually in a double-walled sound booth in the presence of two examiners. One examiner managed the experimental equipment and monitored the child's performance. The other examiner monitored the child's behavior during testing. The child was seated 1.5 m from the front speaker. The target stimulus to be remembered was presented at 0° azimuth from the front loudspeaker at a fixed level of 65 dBA, as measured at the location of the participant's head with a Quest SoundPro 3M sound level meter (Quest Technologies, Medley, FL, USA). In the backward digit recall task, the examiner presented sequences of digits via monitored live voice and the child had to recall each sequence in the reverse order. For example, the sequence 5-1-3 presented at a rate of 1 digit/s would be correctly recalled as 3-1-5. Practice trials were given with two-digit and three-digit sequences in order to ensure that each child understood the concept of “backwards.” Test trials began with two digits and increased by one digit in each block until the child was unable to recall four correct trials at a particular block. A block consisted of six trials. For each child, a correct trial was scored as 1 and an incorrect trial was scored as 0.

Classification of errors

In consistence with Burkholder et al.[14] the errors made during the backward digit span task were classified into the following two main categories: Item errors and order errors. Errors were considered item errors if a digit(s) that was not present in the list was repeated. Errors were classified as order errors if correct digits were repeated but in an incorrect order. To account for the different possible types of item and order errors, a template of error subtypes was generated [Table 1]. For example, if a participant hears 5-1-9, he/she is supposed to produce 9-1-5. However, if he/she produces 5-9-1 as his/her response, this is coded as an initial position order error. The reason behind this is that the initial input position is not produced third and the initial output position does not reflect the last input serial position. The template of errors was carefully constructed by one of the authors and reviewed by the other author to ensure that an error type could be identified for each to-be-remembered target stimulus [Table 1].

Table 1

Descriptions and examples of item and order-error subtypes

Error Type	Description	Example
Item Errors
Omissions	The child did not provide a response within the 10-second response window.	Stimuli: 5-1-9 Target Response: 9-1-5 Omission: ---
Incorrect items/ incorrect order	The child recalled digits not included in the target stimuli.	Stimuli: 5-1-9 Target Response: 9-1-5
		Incorrect items/incorrect order: 5-8-4
Order Errors
Initial position error	The child correctly recalled the digits presented, but made an error in the initial position.	Stimuli: 5-1-9 Target Response: 9-1-5 Initial position error: 5-9-1
Intermediate position error	The child correctly recalled the digits presented, but made an error in the intermediate (middle) position.	Stimuli: 5-1-9 Target Response: 9-1-5 Intermediate position error: 9-5-1
Final position error	The child correctly recalled the digits presented, but made an error in the final position.	Stimuli: 5-1-9 Target Response: 9-1-5 Final position error: 1-5-9
Reversal error	The child correctly recalled the digits presented, but not in reverse order.	Stimuli: 5-1-9 Target Response: 9-1-5 Reversal: 5-1-9

Incorrect responses were tallied and transferred to an error database (n = 869). Two trained research assistants judged independently the type of error made for each item in the database. Point-to-point agreement was 98% (range = 94-100%), indicating excellent scoring reliability. The number of incorrect trials divided by six total trials (a block) was tabulated in percentages.

Results

Statistical analyses were performed using NCSS software, version 8 (Kaysville UT, USA). To analyze the effects of listening condition (quiet condition, degraded condition), error type (item errors, order errors), and digit span (two, three, four, five), a repeated measures analysis of variance (RM-ANOVA) was performed. The dependent variable was the number of incorrect trials divided by total six trials (a block) presented to each child. In order to equate the different number of total item errors made in quiet (235) and in the degraded listening condition (243), the proportions of errors and standard deviations were also calculated for each error type [Table 2].

Table 2

Proportions of each error type in quiet and degraded listening condition

Error Type	Quiet Listening Condition (%)	Degraded Listening Condition (%)
Item Errors
Omissions	6.80	4.10
Incorrect items/incorrect order	17.0	3.70
Total	23.8 % (11.0)	7.8 % (5.0)
Order Errors
Initial position error	9.2	4.4
Intermediate position error	57.8	69.5
Final position error	3.4	3.79
Reversal error	5.8	13.7
Total	76.2 % (2.3)	92.2 % (2.7)

The results of the RM-ANOVA revealed two significant main effects and three interactions. There was a main effect of error type [F (1, 869) = 127.66, P < 0.001] whereby order errors occurred more frequently than item errors. An interaction was observed between the listening condition and error type [F (1,869) = 7.88, P < 0.005]; there was a substantial increase in the frequency of order errors in the degraded listening condition compared to the quiet condition [Figure 1].

Figure 1

Mean proportion of the item and order errors made on the backward digit span recall task in quiet and degraded listening conditions [+1 standard error (SE)]

There was the main effect of digit span [F (3, 869) = 44.55, P < 0.001]. As displayed in Figure 2, children committed the greatest proportion of errors in the fourth span (i.e., four digits per trial item). An interaction between error type and digit span [F (3, 869) = 28.33, P < 0.001] and an interaction between listening condition and digit span [F (3, 869) = 21.33, P < 0.001] were observed. Post hoc comparisons using the Tukey-Kramer multiple comparison tests indicated that item errors were mostly in later spans than in earlier spans in the quiet condition, suggesting that item errors were partly related to reaching working memory capacity limits [Figure 2a]. In the degraded listening condition, the patterns of item errors across the span were evenly distributed, suggesting that the errors were not capacity-related but rather a result of increased processing brought on by noise [Figure 2b]. A one-tailed chi-square test indicated that the profile of errors in quiet were significantly different from the profile of errors in the degraded listening condition [Figure 2; X2 (7, N = 200) = 57.3, P .003].

Figure 2

Frequency of item and order errors across digit span length in (a) quiet and (b) degraded listening condition

Finally, to assess the specific errors made in quiet and the degraded listening condition more accurately, a separate univariate ANOVA was run on the subtypes of item and order error data. This analysis revealed a significant main effect of error subtype [F (5, 478) = 24.59, P < 0.01]. Post hoc analyses indicated that intermediate position errors occurred more frequently than any other type of error, especially in the degraded listening condition, i.e., children had significantly more difficulty recalling the correct order of intermediate middle items than recalling the order of initial and final items, supporting strong primacy and recency effects.

Discussion

The purpose of the present study was to determine whether difficulties on the auditory backward digit span task were due to perceptual masking or cognitive processing demands. We evaluated the type and frequency of errors in two listening conditions in children with normal hearing. Overall, the results suggest that children's difficulty in recalling digits in the reverse order was primarily related to cognitive processing demands and not perceptual audibility.

Cognitive processing demands

Models of working memory suggest that a limited capacity of cognitive resources is available to a child for online processing and storage of temporary information.[7] In consistence with the previously reported capacities, the mean working memory capacity in this study was four; children made the most errors when they were asked to recall a sequence of four digits in the reverse order. The errors were infrequent in short spans but increased rapidly as digit span length increased. As children approached their working memory capacity, their patterns of errors provided insight into two kinds of challenges: Capacity limits due to an individual's recall limit and capacity limits brought on by an extrinsic distractor (i.e., background noise). In the quiet listening condition, there was a large increase of item errors across the digit span. Children committed omissions and repeated fewer digits of a sequence as they approached the limits of their individual recall capacity. In the degraded listening condition, item errors were minimal and evenly distributed across digit span length. Instead, there was a large increase of order errors across the digit span. The increased proportion of order errors across the span suggests that the capacity limits were exceeded because of the concurrent processing demands of recalling digits in the reverse order and simultaneously ignoring the background sounds. Listening in noise-exhausted capacity limits by consuming working memory resources contributed to more order errors.[333435] These findings reflect the limit to the quantity of information that can be held in working memory and the types of errors that arise when the limits are impeded by noise.

Order errors predominated in terms of type and frequency. A detailed examination of the order errors revealed bow-shaped serial position curves; this confirmed that order errors result from the loss of temporal order information during encoding or spoken recall. Most order errors were made with intermediate items, which is consistent with the order-error curves observed in previous studies.[36] This finding can be explained by the intermediate items’ susceptibility to a greater number of possible order confusions and reflects processing demand.[37] While initial items and final position items can be readily recalled, intermediate items require devoted attentional and processing resources for accurate recall. The fact that intermediate position errors occur more frequently in degraded listening condition further suggests that background noise competes and adds to the challenge of accurately recalling digits in the correct order. The demands of ignoring background sounds together with the demands of recalling temporal order information contributes to the majority of order errors observed in the degraded listening condition. Our results are consistent with Burkholder and Pisoni's[30] finding, where 8-10-year-old children with profound hearing loss and their age-matched typically developing peers made more order errors than item errors on the backward digit recall span task.

Perceptual masking

Children, with and without hearing loss, have trouble extracting and attending to target acoustic information in the presence of competing background sounds.[38] Schoolgoing children are particularly susceptible to perceptual masking whereby multitalker maskers interfere with target speech.[13] We previously found that children experienced a significant decrease in accuracy on the backward digit span task in a four-talker masker at an unfavorable SNR compared to a quiet listening condition;[3] however, children's performance on the backward digit span task was unaffected in a broadband signal matched to the noise of an unoccupied classroom at a favorable SNR (air conditioning on at +15 dB SNR).[28] Thus, the use of a four-talker masker and a degraded SNR (-5 dB SNR) warranted concern for the possibility of perceptual masking in this study. If perceptual masking was a problem, we would have observed a large proportion of item errors; children would have reported digits that were not presented in the original target list or would have made omissions. However, the proportion of item errors was low in quiet and in the degraded listening condition. This indicates that even though children received a degraded auditory signal in the four-talker masker, errors on the backward digit span were mainly due to the retention of sequential order information and were not related to reduced audibility.

Limitations

While our findings demonstrate processing-related errors in the degraded listening condition, it is unknown as to what specifically contributes to the increased cognitive load. It is possible that audibility affected the initial processing of the number sequence (i.e., the child heard 5-1-3 instead of 5-3-1) and the children did, in fact, recall it in the reverse order (i.e., produced 3-1-5) but it was scored as an order error. However, since the children are consistently able to produce an accurate response in the forward digit recall task in noise, we believe that this is the least likely case. We suspect that if audibility affected the initial encoding/processing of the digits, the pattern of order errors would have been more consistent across all spans rather than increase as the child approached his or her capacity. In the future, we suggest the use of independent measures, which will systematically differentiate between auditory/encoding and cognitive processing demands brought by noise. Moreover, given the span procedure where testing stopped when a child was unable to remember four of the six lists of a block, every child had a different list length in every condition. Therefore, in the future, a separate analysis for children with different abilities (spans) is recommended.

Conclusions

The error patterns observed in this study are supported by the load theory, which describes a passive perceptual mechanism and an active mechanism of cognitive control.[39] In this view, even if a target signal is well-perceived (i.e., minimal-to-no masking effects), there is an increased cognitive processing involvement inhibiting that irrelevant distractor. The large proportion of order errors and serial position effects reported in this study suggest that noise drains cognitive resources away from the primary task at hand (i.e., recall of digits in the reverse order). This has several clinical and educational implications. If decreased working memory performance in noise is a consequence of high cognitive processing demands, then auditory comprehension, language literacy skills, and other academic-related cognitive abilities might also be at risk.

Financial support and sponsorship

The University of Washington Royalty Research Fund (#65-6403) funded this research.

Conflicts of interest

There are no conflicts of interest.