Recognising foreign-accented speech of varying intelligibility and linguistic complexity: insights from older listeners with or without hearing loss

Abstract

Objective

The goal of this study was to assess recognition of foreign-accented speech of varying intelligibility and linguistic complexity in older adults. It is important to understand the factors that influence the recognition of this commonly encountered type of speech, in a population that remains understudied in this regard.

Design

A repeated measures design was used. Listeners repeated back linguistically simple and complex sentences heard in noise. The sentences were produced by three talkers of varying intelligibility: one native American English, one foreign-accented talker of high intelligibility and one foreign-accented talker of low intelligibility. Percentage word recognition in sentences was measured.

Study sample

Twenty-five older listeners with a range of hearing thresholds participated.

Results

We found a robust interaction between talker intelligibility and linguistic complexity. Recognition accuracy was higher for simple versus complex sentences, but only for the native and high intelligibility foreign-accented talkers. This pattern was present after effects of working memory capacity and hearing acuity were taken into consideration.

Conclusion

Older listeners exhibit qualitatively different speech processing strategies for low versus high intelligibility foreign-accented talkers. Differences in recognition accuracy for words presented in simple versus in complex sentence contexts only emerged for speech over a threshold of intelligibility.

Keywords:

• Speech perception
• foreign-accented speech
• linguistic complexity
• behavioural measures

Disclosure statement

The authors declare no potential conflict of interest.

Notes

1 The length of HINT sentences ranges from 4 to 7 words and the vast majority of them have only one verb, with few sentences having two verbs. All the HINT sentences used in the present study were 4 to 6 words long and none of them had more than one verb.

2 These were a subset of the talkers used in Strori et al. (2020).

3 Generalized linear mixed effects models (GLMMs) with logistic regression are recommended to analyse data wherein the response variable is a proportion (or percentage) with values between 0 and 1 (Jaeger 2008; Bolker et al. 2009), such as in the present case. Using the beta-binomial distribution for the proportion response variable (instead of the more typically used binomial distribution) solves a critical problem attributed to the usage of the binomial distribution with proportion data, namely the contextual effect in sentences. Specifically, in the context of a sentence, each word represents a trial and the probability of success/failure (p) of a trial clearly affects that of the other trials, which runs counter to the assumptions of the binomial distribution where p is independent for each trial and is fixed. In the beta-binomial distribution, p is not fixed in each trial, but varies randomly following a beta distribution, thus accounting for the possibility of contextual effects in a sentence, unlike the binomial distribution. For further discussion of the advantages of using this distribution for analysing percentage/proportion data, see Crowder (1978); Hilbe (2013); and Muniz-Terrera et al. (2016).

4 The models were implemented with the 'glmmTMB' R package (Brooks et al. 2017) and subsequent pairwise comparisons between the conditions were performed with the 'emmeans' R package (Lenth 2019).

5 The binary factor Complexity was dummy coded as: 0 – Simple, 1 – Complex (default = Simple) and the 3-level factor Talker was contrast coded using the forward difference coding system (default = Native). Forward difference coding, a strategy for coding categorical variables in mixed effects modelling, compares the mean of the dependent variable on a specific level of the independent variable to the mean of the dependent variable for the next (adjacent) level of the independent variable. In the present models, Talker was contrast coded as follows: the first contrast compared differences between the native and high intelligibility non-native talker (2/3, −1/3, −1/3) and the second contrast compared differences between the high and low intelligibility non-native talkers (1/3, 1/3, −2/3).

6 The correlation between working memory capacity (WM) and amount of hearing loss (PTA) was assessed prior to including these factors in the regression models and it was revealed to be weak and non-significant, Pearson’s r = −0.02, p > .05. This allowed us to include these predictors simultaneously in the corresponding mixed effects logistic regression models.

7 These comparisons were implemented with the ‘emmeans’ R package (Lenth 2019), previously known as ‘lsmeans’.

Additional information

Funding

This research was supported by a grant from the Alumnae of Northwestern University. The authors thank Chun Liang Chan for his technical support and Kendra Marks for her assistance during data collection.