Masked Sentence Research Articles

Gauging, enriching and applying geography knowledge in Pre-trained Language Models

To employ Pre-trained Language Models (PLMs) as knowledge containers in niche domains it is important to gauge the knowledge of these PLMs about facts in these domains. It is also an important pre-requisite to know how much enrichment effort is required to make them better. As part of this work, we aim to gauge and enrich small PLMs for knowledge of world geography. Firstly, we develop a moderately sized dataset of masked sentences covering 24 different fact types about world geography to estimate knowledge of PLMs on these facts. We hypothesize that for this niche domain, smaller PLMs may not be well equipped. Secondly, we enrich PLMs with this knowledge through fine-tuning and check if the knowledge in the dataset is infused sufficiently. We further hypothesize that linguistic variability in the manual templates used to embed the knowledge in masked sentences does not affect the knowledge infusion. Finally, we demonstrate the application of PLMs to tourism blog search and Wikidata KB augmentation. In both applications, we aim at showing the effectiveness of using PLMs to achieve competitive performance.

Two-Year Outcomes of Cochlear Implant Use for Children With Unilateral Hearing Loss: Benefits and Comparison to Children With Normal Hearing.

Children with severe-to-profound unilateral hearing loss, including cases of single-sided deafness (SSD), lack access to binaural cues that support spatial hearing, such as recognizing speech in complex multisource environments and sound source localization. Listening in a monaural condition negatively impacts communication, learning, and quality of life for children with SSD. Cochlear implant (CI) use may restore binaural hearing abilities and improve outcomes as compared to alternative treatments or no treatment. This study investigated performance over 24 months of CI use in young children with SSD as compared to the better hearing ear alone and to children with bilateral normal hearing (NH). Eighteen children with SSD who received a CI between the ages of 3.5 and 6.5 years as part of a prospective clinical trial completed assessments of word recognition in quiet, masked sentence recognition, and sound source localization at regular intervals out to 24-month postactivation. Eighteen peers with bilateral NH, matched by age at the group level, completed the same test battery. Performance at 24-month postactivation for the SSD group was compared to the performance of the NH group. Children with SSD have significantly poorer speech recognition in quiet, masked sentence recognition, and localization both with and without the use of the CI than their peers with NH. The SSD group experienced significant benefits with the CI+NH versus the NH ear alone on measures of isolated word recognition, masked sentence recognition, and localization. These benefits were realized within the first 3 months of use and were maintained through the 24-month postactivation interval. Young children with SSD who use a CI experience significant isolated word recognition and bilateral spatial hearing benefits, although their performance remains poorer than their peers with NH.

Multimodal Pre-Training Based on Graph Attention Network for Document Understanding

Document intelligence as a relatively new research topic supports many business applications. Its main task is to automatically read, understand, and analyze documents. However, due to the diversity of formats (invoices, reports, forms, etc.) and layouts in documents, it is difficult to make machines understand documents. In this paper, we present the GraphDoc, a multimodal graph attention-based model for various document understanding tasks. GraphDoc is pre-trained in a multimodal framework by utilizing text, layout, and image information simultaneously. In a document, a text block relies heavily on its surrounding contexts, accordingly we inject the graph structure into the attention mechanism to form a graph attention layer so that each input node can only attend to its neighborhoods. The input nodes of each graph attention layer are composed of textual, visual, and positional features from semantically meaningful regions in a document image. We do the multimodal feature fusion of each node by the gate fusion layer. The contextualization between each node is modeled by the graph attention layer. GraphDoc learns a generic representation from only 320k unlabeled documents via the Masked Sentence Modeling task. Extensive experimental results on the publicly available datasets show that GraphDoc achieves state-of-the-art performance, which demonstrates the effectiveness of our proposed method. The code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/ZZR8066/GraphDoc</uri> .

Capacity limits in sentence comprehension: Evidence from dual-task judgements and event-related potentials

There is an ongoing controversy over whether readers can access the meaning of multiple words, simultaneously. To date, different experimental methods have generated seemingly contradictory evidence in support of serial or parallel processing accounts. For example, dual-task studies suggest that readers can process a maximum of one word at a time (White, Palmer & Boynton, 2018), while ERP studies have demonstrated neural priming effects that are more consistent with parallel activation (Wen, Snell & Grainger, 2019). To help reconcile these views, I measured neural responses and behavioral accuracy in a dual-task sentence comprehension paradigm. Participants saw masked sentences and two-word phrases and had to judge whether or not they were grammatical. Grammatically correct sentences (This girl is neat) produced smaller N400 responses compared to scrambled sentences (Those girl is fled): an N400 sentence superiority effect. Critically, participants' grammaticality judgements on the same trials showed striking capacity limitations, with dual-task deficits closely matching the predictions of a serial, all-or-none processing account. Together, these findings suggest that the N400 sentence superiority effect is fully compatible with serial word recognition, and that readers are unable to process multiple sentence positions simultaneously.

Age-related deficits in dip-listening evident for isolated sentences but not for spoken stories

Fluctuating background sounds facilitate speech intelligibility by providing speech ‘glimpses’ (masking release). Older adults benefit less from glimpses, but masking release is typically investigated using isolated sentences. Recent work indicates that using engaging, continuous speech materials (e.g., spoken stories) may qualitatively alter speech-in-noise listening. Moreover, neural sensitivity to different amplitude envelope profiles (ramped, damped) changes with age, but whether this affects speech listening is unknown. In three online experiments, we investigate how masking release in younger and older adults differs for masked sentences and stories, and how speech intelligibility varies with masker amplitude profile. Intelligibility was generally greater for damped than ramped maskers. Masking release was reduced in older relative to younger adults for disconnected sentences, and stories with a randomized sentence order. Critically, when listening to stories with an engaging and coherent narrative, older adults demonstrated equal or greater masking release compared to younger adults. Older adults thus appear to benefit from ‘glimpses’ as much as, or more than, younger adults when the speech they are listening to follows a coherent topical thread. Our results highlight the importance of cognitive and motivational factors for speech understanding, and suggest that previous work may have underestimated speech-listening abilities in older adults.

A Word-Granular Adversarial Attacks Framework for Causal Event Extraction.

As a data augmentation method, masking word is commonly used in many natural language processing tasks. However, most mask methods are based on rules and are not related to downstream tasks. In this paper, we propose a novel masking word generator, named Actor-Critic Mask Model (ACMM), which can adaptively adjust the mask strategy according to the performance of downstream tasks. In order to demonstrate the effectiveness of the method, we conducted experiments on two causal event extraction datasets. Experiment results show that, compared with various rule-based masking methods, the masked sentences generated by our proposed method can significantly enhance the generalization of the model and improve the model performance.

Entity Slot Filling for Visual Captioning

To explore the specific visual aspects and the language consistency at the same time, this paper introduces a new image captioning task, dubbed <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">entity slot filling captioning (ESFCap)</i> . It is similar to the masked entity completion tasks in NLP, which are widely used to study language context and has been successfully employed to improve language understanding. Specifically, given a sentence with blank for describing an image, the ESFCap task aims to fill the blank with proper text content according to the visual information. The filled text should be grounded to correct visual entities and also in concordance with the sentence structure. To support the ESFCap research, we collect and release an entity slot filling captioning dataset, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Flickr30k-EnFi</i> , based on Flickr30k-Entities. The Flickr30k-EnFi dataset consists of 31,783 images and 565,750 masked sentences, as well as the text snippets for the masked slot. For tackling the ESFCap task, we propose a multi-modal fusion model equipped with a novel adaptive dynamic attention module, termed AdaMFN. The AdaMFN model effectively leverages both global and local information from vision and language. It is also able to adaptively focus on the key linguistic knowledge and visual regions to generate correct filling results. The experimental results and analysis demonstrate the effectiveness of our proposed model.

Effects of Presentation Level on Spatial Hearing With and Without Bone-Conduction Amplification in Congenital Unilateral Aural Atresia.

This study assessed the effect of ipsilateral bone-conduction amplification on spatial hearing abilities in subjects with congenital unilateral aural atresia (CUAA). Twelve patients with unilateral conductive hearing loss secondary to CUAA and normal hearing in the contralateral ear were tested. Most (75%) had limited experience with a bone-conduction hearing aid (BCHA). Performance was evaluated with and without a BCHA fitted acutely on a softband. Spatial hearing abilities were evaluated in two paradigms. Spatial release from masking was evaluated by comparing masked sentence recognition with a target and two speech maskers either colocated at 0 degree or with the maskers separated at +90 degrees and -90 degrees. Sound source localization was evaluated in a 180 degrees arc of loudspeakers on the horizontal plane. Performance was evaluated at 50 and 75 dB SPL, and results were compared for patients tested with and without a BCHA. Group level results indicate similar spatial release from masking in the aided and unaided conditions at both presentation levels. Localization at 50 dB SPL was similar aided and unaided, but at 75 dB SPL the root mean square error was lower unaided than aided (17.2 degrees vs 41.3 degrees; p = 0.010). Use of a BCHA in patients with CUAA may interfere with auditory cues required for sound source localization when the signal level is intense enough to overcome the patient's conductive hearing loss. These findings have potential clinical implications in fitting of BCHAs to support optimal spatial hearing in patients with CUAA.

Effectiveness of Place-based Mapping in Electric-Acoustic Stimulation Devices.

The default mapping procedure for electric-acoustic stimulation (EAS) devices uses the cochlear implant recipient's unaided detection thresholds in the implanted ear to derive the acoustic settings and assign the lowest frequency filter of electric stimulation. Individual differences for speech recognition with EAS may be due to discrepancies between the electric frequency filters of individual electrode contacts and the cochlear place of stimulation, known as a frequency-to-place mismatch. Frequency-to-place mismatch of greater than 1/2 octave has been demonstrated in up to 60% of EAS users. Aligning the electric frequency filters via a place-based mapping procedure using postoperative imaging may improve speech recognition with EAS. Masked sentence recognition was evaluated for normal-hearing subjects (n = 17) listening with vocoder simulations of EAS, using a place-based map and a default map. Simulation parameters were based on audiometric and imaging data from a representative 24-mm electrode array recipient and EAS user. The place-based map aligned electric frequency filters with the cochlear place frequency, which introduced a gap between the simulated acoustic and electric output. The default map settings were derived from the clinical programming software and provided the full speech frequency range. Masked sentence recognition was significantly better for simulated EAS with the place-based map as compared with the default map. The simulated EAS place-based map supported better performance than the simulated EAS default map. This indicates that individualizing maps may improve performance in EAS users by helping them achieve better asymptotic performance earlier and mitigate the need for acclimatization.

Cognitive and Linguistic Contributions to Masked Speech Recognition in Children.

Purpose The goal of this study was to examine the effects of cognitive and linguistic skills on masked speech recognition for children with normal hearing in three different masking conditions: (a) speech-shaped noise (SSN), (b) amplitude-modulated SSN (AMSSN), and (c) two-talker speech (TTS). We hypothesized that children with better working memory and language skills would have better masked speech recognition than peers with poorer skills in these areas. Selective attention was predicted to affect performance in the TTS masker due to increased cognitive demands from informational masking. Method A group of 60 children in two age groups (5- to 6-year-olds and 9- to 10-year-olds) with normal hearing completed sentence recognition in SSN, AMSSN, and TTS masker conditions. Speech recognition thresholds for 50% correct were measured. Children also completed standardized measures of language, memory, and executive function. Results Children's speech recognition was poorer in the TTS relative to the SSN and AMSSN maskers. Older children had lower speech recognition thresholds than younger children for all masker conditions. Greater language abilities were associated with better sentence recognition for the younger children in all masker conditions, but there was no effect of language for older children. Better working memory and selective attention skills were associated with better masked sentence recognition for both age groups, but only in the TTS masker condition. Conclusions The decreasing influence of vocabulary on masked speech recognition for older children supports the idea that this relationship depends on an interaction between the language level of the stimuli and the listener's vocabulary. Increased cognitive demands associated with perceptually isolating the target talker and two competing masker talkers with a TTS masker may result in the recruitment of working memory and selective attention skills, effects that were not observed in SSN or AMSSN maskers. Future research should evaluate these effects across a broader range of stimuli or with children who have hearing loss.

Masked Sentence Model Based on BERT for Move Recognition in Medical Scientific Abstracts

Abstract Purpose Move recognition in scientific abstracts is an NLP task of classifying sentences of the abstracts into different types of language units. To improve the performance of move recognition in scientific abstracts, a novel model of move recognition is proposed that outperforms the BERT-based method. Design/methodology/approach Prevalent models based on BERT for sentence classification often classify sentences without considering the context of the sentences. In this paper, inspired by the BERT masked language model (MLM), we propose a novel model called the masked sentence model that integrates the content and contextual information of the sentences in move recognition. Experiments are conducted on the benchmark dataset PubMed 20K RCT in three steps. Then, we compare our model with HSLN-RNN, BERT-based and SciBERT using the same dataset. Findings Compared with the BERT-based and SciBERT models, the F1 score of our model outperforms them by 4.96% and 4.34%, respectively, which shows the feasibility and effectiveness of the novel model and the result of our model comes closest to the state-of-the-art results of HSLN-RNN at present. Research limitations The sequential features of move labels are not considered, which might be one of the reasons why HSLN-RNN has better performance. Our model is restricted to dealing with biomedical English literature because we use a dataset from PubMed, which is a typical biomedical database, to fine-tune our model. Practical implications The proposed model is better and simpler in identifying move structures in scientific abstracts and is worthy of text classification experiments for capturing contextual features of sentences. Originality/value T he study proposes a masked sentence model based on BERT that considers the contextual features of the sentences in abstracts in a new way. The performance of this classification model is significantly improved by rebuilding the input layer without changing the structure of neural networks.

Sentence perception in noise by hearing-aid users: Relations with syllable-constituent perception and the use of context

Masked sentence perception by hearing-aid users is strongly correlated with three variables: (1) the ability to hear phonetic details as estimated by the identification of syllable constituents in quiet or in noise; (2) the ability to use situational context that is extrinsic to the speech signal; and (3) the ability to use inherent context provided by the speech signal itself. These conclusions are supported by the performance of 57 hearing-aid users in the identification of 109 syllable constituents presented in a background of 12-talker babble and the identification of words in naturally spoken sentences presented in the same babble. A mathematical model is offered that allows calculation of an individual listener's sentence scores from estimates of context utilization and the ability to identify syllable constituents. When the identification accuracy of syllable constituents is greater than about 55%, individual differences in context utilization play a minor role in determining the sentence scores. As syllable constituent scores fall below 55%, individual differences in context utilization play an increasingly greater role in determining sentence scores. When a listener's syllable constituent score is above about 71% in quiet, the listeners score in quiet will above about 55% in noise. [Watson and Miller are shareholders in Communication Disorders Technology, Inc., and may profit from sales of the software used in this study.] Masked sentence perception by hearing-aid users is strongly correlated with three variables: (1) the ability to hear phonetic details as estimated by the identification of syllable constituents in quiet or in noise; (2) the ability to use situational context that is extrinsic to the speech signal; and (3) the ability to use inherent context provided by the speech signal itself. These conclusions are supported by the performance of 57 hearing-aid users in the identification of 109 syllable constituents presented in a background of 12-talker babble and the identification of words in naturally spoken sentences presented in the same babble. A mathematical model is offered that allows calculation of an individual listener's sentence scores from estimates of context utilization and the ability to identify syllable constituents. When the identification accuracy of syllable constituents is greater than about 55%, individual differences in context utilization play a minor role in determining the sentence scores. A...

Development of speech recognition in noise or two-talker speech: Context effects related to response alternatives and sentence meaning

Speech recognition in normal-hearing adults is affected by pragmatic restrictions on the target content. For example, masked sentence recognition is better when target speech is composed of semantically meaningful compared to anomalous sentences. Similarly, masked word recognition is better when assessed in a close-set than an open-set task, even after accounting for changes in chance performance; this effect is most pronounced when the response alternatives in the closed-set task are acoustically distinct. In both the cases, restricting the set of plausible responses reduces the fidelity of acoustic cues necessary to perform the task. Although young school-age children benefit from this type of context, the magnitude of benefit relative to that observed for adults depends on the masker type. The benefit associated with increasingly restricted response alternatives is similar for young children and adults when the masker is noise, but young children derive little or no benefit when the masker is two-talker speech. Children and adults also differ with respect to the benefit of semantic context for sentences presented in noise or two-talker speech. Potential factors responsible for these developmental effects will be discussed, including maturation of auditory stream segregation, working memory, and acoustic/phonetic templates supporting word recognition.

In a Concurrent Memory and Auditory Perception Task, the Pupil Dilation Response Is More Sensitive to Memory Load Than to Auditory Stimulus Characteristics.

Objectives:Speech understanding may be cognitively demanding, but it can be enhanced when semantically related text cues precede auditory sentences. The present study aimed to determine whether (a) providing text cues reduces pupil dilation, a measure of cognitive load, during listening to sentences, (b) repeating the sentences aloud affects recall accuracy and pupil dilation during recall of cue words, and (c) semantic relatedness between cues and sentences affects recall accuracy and pupil dilation during recall of cue words.Design:Sentence repetition following text cues and recall of the text cues were tested. Twenty-six participants (mean age, 22 years) with normal hearing listened to masked sentences. On each trial, a set of four-word cues was presented visually as text preceding the auditory presentation of a sentence whose meaning was either related or unrelated to the cues. On each trial, participants first read the cue words, then listened to a sentence. Following this they spoke aloud either the cue words or the sentence, according to instruction, and finally on all trials orally recalled the cues. Peak pupil dilation was measured throughout listening and recall on each trial. Additionally, participants completed a test measuring the ability to perceive degraded verbal text information and three working memory tests (a reading span test, a size-comparison span test, and a test of memory updating).Results:Cue words that were semantically related to the sentence facilitated sentence repetition but did not reduce pupil dilation. Recall was poorer and there were more intrusion errors when the cue words were related to the sentences. Recall was also poorer when sentences were repeated aloud. Both behavioral effects were associated with greater pupil dilation. Larger reading span capacity and smaller size-comparison span were associated with larger peak pupil dilation during listening. Furthermore, larger reading span and greater memory updating ability were both associated with better cue recall overall.Conclusions:Although sentence-related word cues facilitate sentence repetition, our results indicate that they do not reduce cognitive load during listening in noise with a concurrent memory load. As expected, higher working memory capacity was associated with better recall of the cues. Unexpectedly, however, semantic relatedness with the sentence reduced word cue recall accuracy and increased intrusion errors, suggesting an effect of semantic confusion. Further, speaking the sentence aloud also reduced word cue recall accuracy, probably due to articulatory suppression. Importantly, imposing a memory load during listening to sentences resulted in the absence of formerly established strong effects of speech intelligibility on the pupil dilation response. This nullified intelligibility effect demonstrates that the pupil dilation response to a cognitive (memory) task can completely overshadow the effect of perceptual factors on the pupil dilation response. This highlights the importance of taking cognitive task load into account during auditory testing.

Masked Sentence Recognition in Children, Young Adults, and Older Adults: Age-Dependent Effects of Semantic Context and Masker Type.

Masked speech recognition in normal-hearing listeners depends in part on masker type and semantic context of the target. Children and older adults are more susceptible to masking than young adults, particularly when the masker is speech. Semantic context has been shown to facilitate noise-masked sentence recognition in all age groups, but it is not known whether age affects a listener's ability to use context with a speech masker. The purpose of the present study was to evaluate the effect of masker type and semantic context of the target as a function of listener age. Listeners were children (5 to 16 years), young adults (19 to 30 years), and older adults (67 to 81 years), all with normal or near-normal hearing. Maskers were either speech-shaped noise or two-talker speech, and targets were either semantically correct (high context) sentences or semantically anomalous (low context) sentences. As predicted, speech reception thresholds were lower for young adults than either children or older adults. Age effects were larger for the two-talker masker than the speech-shaped noise masker, and the effect of masker type was larger in children than older adults. Performance tended to be better for targets with high than low semantic context, but this benefit depended on age group and masker type. In contrast to adults, children benefitted less from context in the two-talker speech masker than the speech-shaped noise masker. Context effects were small compared with differences across age and masker type. Different effects of masker type and target context are observed at different points across the lifespan. While the two-talker masker is particularly challenging for children and older adults, the speech masker may limit the use of semantic context in children but not adults.

Neural capacity limits during unconscious semantic processing.

A growing body of neuroimaging data suggests that direct measurements of brain activity can reveal subliminal effects that remain invisible with behavior measures alone. We examined whether sentence comprehension processes could be triggered by a sequence of masked words. On each trial, participants viewed a rapid sequence of masked or unmasked words, including a subject noun, three adverbs and followed by a visible target verb. To probe the capacity limits of unconscious processing, we measured event-related potentials associated with the semantic congruency between the noun and the verb, while varying the subject position in each sentence. Unmasked sentences produced significant behavioral effects of congruency, paralleled by robust N400 effects, independently of subject-verb distance. By contrast, masked sentences produced no behavioral effect and elicited N400 effects only when subjects and verbs were separated by 0 or 1 word. The present results suggest that semantic integration of multiple words can occur unconsciously only if the distance between the words to be integrated does not exceed two words. Although the possibility remains that even longer sequence of invisible words may produce similar neural effects in different experimental settings, our ERP data show that only conscious perception gives access to a buffer that enables robust sentence-level processing independently of temporal distance.

Reading Behind the Lines: The Factors Affecting the Text Reception Threshold in Hearing Aid Users.

The visual Text Reception Threshold (TRT) test (Zekveld et al., 2007) has been designed to assess modality-general factors relevant for speech perception in noise. In the last decade, the test has been adopted in audiology labs worldwide. The 1st aim of this study was to examine which factors best predict interindividual differences in the TRT. Second, we aimed to assess the relationships between the TRT and the speech reception thresholds (SRTs) estimated in various conditions. First, we reviewed studies reporting relationships between the TRT and the auditory and/or cognitive factors and formulated specific hypotheses regarding the TRT predictors. These hypotheses were tested using a prediction model applied to a rich data set of 180 hearing aid users. In separate association models, we tested the relationships between the TRT and the various SRTs and subjective hearing difficulties, while taking into account potential confounding variables. The results of the prediction model indicate that the TRT is predicted by the ability to fill in missing words in incomplete sentences, by lexical access speed, and by working memory capacity. Furthermore, in line with previous studies, a moderate association between higher age, poorer pure-tone hearing acuity, and poorer TRTs was observed. Better TRTs were associated with better SRTs for the correct perception of 50% of Hagerman matrix sentences in a 4-talker babble, as well as with better subjective ratings of speech perception. Age and pure-tone hearing thresholds significantly confounded these associations. The associations of the TRT with SRTs estimated in other conditions and with subjective qualities of hearing were not statistically significant when adjusting for age and pure-tone average. We conclude that the abilities tapped into by the TRT test include processes relevant for speeded lexical decision making when completing partly masked sentences and that these processes require working memory capacity. Furthermore, the TRT is associated with the SRT of hearing aid users as estimated in a challenging condition that includes informational masking and with experienced difficulties with speech perception in daily-life conditions. The current results underline the value of using the TRT test in studies involving speech perception and aid in the interpretation of findings acquired using the test.

The Influence of Linguistic Proficiency on Masked Text Recognition Performance in Adults With and Without Congenital Hearing Impairment.

The authors first examined the influence of moderate to severe congenital hearing impairment (CHI) on the correctness of samples of elicited spoken language. Then, the authors used this measure as an indicator of linguistic proficiency and examined its effect on performance in language reception, independent of bottom-up auditory processing. In groups of adults with normal hearing (NH, n = 22), acquired hearing impairment (AHI, n = 22), and moderate to severe CHI (n = 21), the authors assessed linguistic proficiency by analyzing the morphosyntactic correctness of their spoken language production. Language reception skills were examined with a task for masked sentence recognition in the visual domain (text), at a readability level of 50%, using grammatically correct sentences and sentences with distorted morphosyntactic cues. The actual performance on the tasks was compared between groups. Adults with CHI made more morphosyntactic errors in spoken language production than adults with NH, while no differences were observed between the AHI and NH group. This outcome pattern sustained when comparisons were restricted to subgroups of AHI and CHI adults, matched for current auditory speech reception abilities. The data yielded no differences between groups in performance in masked text recognition of grammatically correct sentences in a test condition in which subjects could fully take advantage of their linguistic knowledge. Also, no difference between groups was found in the sensitivity to morphosyntactic distortions when processing short masked sentences, presented visually. These data showed that problems with the correct use of specific morphosyntactic knowledge in spoken language production are a long-term effect of moderate to severe CHI, independent of current auditory processing abilities. However, moderate to severe CHI generally does not impede performance in masked language reception in the visual modality, as measured in this study with short, degraded sentences. Aspects of linguistic proficiency that are affected by CHI thus do not seem to play a role in masked sentence recognition in the visual modality.

Modulation masking and masked speech perception in normal-hearing school-age children and adults

It has recently been suggested that adults’ ability to recognize speech in noise is limited by the inherent modulation of that noise—a form of modulation masking. This study evaluated whether immature masked speech perception in school-age children could be due to greater susceptibility to modulation masking. To gauge modulation sensitivity, the first experiment measured sinusoidal modulation detection for rates of 10–300 Hz carried by a 5000-Hz pure tone. There were large individual differences, but little evidence of a child/adult difference. This adult-like modulation detection for a tonal carrier contrasts with published findings of adult/child differences in modulation detection for a noise-band carrier, suggesting that children may be more susceptible to modulation masking than adults. The second experiment evaluated masked sentence recognition for speech that was filtered into 28 adjacent equivalent rectangular bands (100–7800 Hz), with alternate bands presented to opposite ears. Maskers were composed of either noise bands or tones, one centered on each speech band. These stimuli have been argued to characterize effects of modulation masking. Young children tended to perform more poorly than adults overall. Masker effects will be discussed in terms of possible developmental differences in energetic and modulation masking.

Musical training, individual differences and the cocktail party problem.

Are musicians better able to understand speech in noise than non-musicians? Recent findings have produced contradictory results. Here we addressed this question by asking musicians and non-musicians to understand target sentences masked by other sentences presented from different spatial locations, the classical ‘cocktail party problem’ in speech science. We found that musicians obtained a substantial benefit in this situation, with thresholds ~6 dB better than non-musicians. Large individual differences in performance were noted particularly for the non-musically trained group. Furthermore, in different conditions we manipulated the spatial location and intelligibility of the masking sentences, thus changing the amount of ‘informational masking’ (IM) while keeping the amount of ‘energetic masking’ (EM) relatively constant. When the maskers were unintelligible and spatially separated from the target (low in IM), musicians and non-musicians performed comparably. These results suggest that the characteristics of speech maskers and the amount of IM can influence the magnitude of the differences found between musicians and non-musicians in multiple-talker “cocktail party” environments. Furthermore, considering the task in terms of the EM-IM distinction provides a conceptual framework for future behavioral and neuroscientific studies which explore the underlying sensory and cognitive mechanisms contributing to enhanced “speech-in-noise” perception by musicians.