Revisiting which language declines more in Spanish-English bilinguals with Alzheimer's disease: Longitudinal decline patterns on the multilingual naming test.

Theories of bilingual language production predict that bilinguals with Alzheimer's disease (AD) should exhibit one of two decline patterns. Either parallel decline of both languages (if decline reflects damage to semantic representations that are accessed by both languages), or asymmetrical decline, with greater decline of the nondominant language (if decline reflects reduced ability to resolve competition from the dominant language with disease progression). Only two previous studies examined decline longitudinally with one showing parallel, and the other asymmetrical, decline. We examined decline over 2-7 years (3.9 on average) in Spanish-English bilinguals (N = 23). Logistic regression revealed a parallel decline pattern at one year from baseline, but an asymmetrical decline pattern over the longer decline period, with greater decline of the nondominant language (when calculating predicted probabilities of a correct response). The asymmetrical decline pattern was significantly greater for the nondominant language only when including item-difficulty in the model. Exploratory analyses across dominance groups looking at proportional decline relative to initial naming accuracy further suggested that decline of the nondominant language may be more precipitous if that language was acquired later in life, but the critical interaction needed to support this possibility was not statistically significant in a logistic regression analysis. These results suggest that accessibility of the nondominant language may initially be more resilient in early versus more advanced AD, and that AD affects shared semantic representations before executive control declines to a point where the ability to name pictures in single-language testing block is disrupted. Additional work is needed to determine if asymmetrical decline patterns are magnified by late age of acquisition of the nondominant language, and if more subtle impairments to executive control underlie impairments to language switching that occur in the earliest stages of AD (even preclinically).

A perceptual similarity space for speech based on self-supervised speech representations.

Speech recognition by both humans and machines frequently fails in non-optimal yet common situations. For example, word recognition error rates for second-language (L2) speech can be high, especially under conditions involving background noise. At the same time, both human and machine speech recognition sometimes shows remarkable robustness against signal- and noise-related degradation. Which acoustic features of speech explain this substantial variation in intelligibility? Current approaches align speech to text to extract a small set of pre-defined spectro-temporal properties from specific sounds in particular words. However, variation in these properties leaves much cross-talker variation in intelligibility unexplained. We examine an alternative approach utilizing a perceptual similarity space acquired using self-supervised learning. This approach encodes distinctions between speech samples without requiring pre-defined acoustic features or speech-to-text alignment. We show that L2 English speech samples are less tightly clustered in the space than L1 samples reflecting variability in English proficiency among L2 talkers. Critically, distances in this similarity space are perceptually meaningful: L1 English listeners have lower recognition accuracy for L2 speakers whose speech is more distant in the space from L1 speech. These results indicate that perceptual similarity may form the basis for an entirely new speech and language analysis approach.

The perception of code-switched speech in noise.

This study investigates heritage bilingual speakers' perception of naturalistic code-switched sentences (i.e., use of both languages in one sentence). Studies of single word perception suggest that code-switching is more difficult to perceive than single language speech. However, such difficulties may not extend to more naturalistic sentences, where predictability and other cues may serve to ameliorate such difficulties. Fifty-four Mexican-American Spanish heritage bilinguals transcribed sentences in noise in English, Spanish, and code-switched blocks. Participants were better at perceiving speech in single language blocks than code-switched blocks. The results indicate that increased language co-activation when perceiving code-switching results in significant processing costs.

Automatic recognition of second language speech-in-noise.

Measuring how well human listeners recognize speech under varying environmental conditions (speech intelligibility) is a challenge for theoretical, technological, and clinical approaches to speech communication. The current gold standard-human transcription-is time- and resource-intensive. Recent advances in automatic speech recognition (ASR) systems raise the possibility of automating intelligibility measurement. This study tested 4 state-of-the-art ASR systems with second language speech-in-noise and found that one, whisper, performed at or above human listener accuracy. However, the content of whisper's responses diverged substantially from human responses, especially at lower signal-to-noise ratios, suggesting both opportunities and limitations for ASR--based speech intelligibility modeling.