There has been a surge of interest regarding the alignment of large-scale language models with human language comprehension behaviour. The majority of this research investigates comprehension behaviours from reading isolated, written sentences. We propose studying the perception of dialogue, focusing on an intrinsic form of language use: spoken conversations. Using the task of predicting upcoming dialogue turns, we ask whether turn plausibility scores produced by state-of-the-art language models correlate with human judgements. We find a strong correlation for some but not all models: masked language models produce stronger correlations than auto-regressive models. In doing so, we quantify human performance on the response selection task for open-domain spoken conversation. To the best of our knowledge, this is the first such quantification.We find that response selection performance can be used as a coarse proxy for the strength of correlation with human judgements, however humans and models make different response selection mistakes. The model which produces the strongest correlation also outperforms human response selection performance. Through ablation studies, we show that pre-trained language models provide a useful basis for turn representations; however, fine-grained contextualisation, inclusion of dialogue structure information, and fine-tuning towards response selection all boost response selection accuracy by over 30 absolute points.
Commercial Automatic Speech Recognition (ASR) systems tend to show systemic predictive bias for marginalised speaker/user groups. We highlight the need for an interdisciplinary and context-sensitive approach to documenting this bias incorporating perspectives and methods from sociolinguistics, speech & language technology and human-computer interaction in the context of a case study. We argue evaluation of ASR systems should be disaggregated by speaker group, include qualitative error analysis, and consider user experience in a broader sociolinguistic and social context.
Current multimodal sentiment analysis frames sentiment score prediction as a general Machine Learning task. However, what the sentiment score actually represents has often been overlooked. As a measurement of opinions and affective states, a sentiment score generally consists of two aspects: polarity and intensity. We decompose sentiment scores into these two aspects and study how they are conveyed through individual modalities and combined multimodal models in a naturalistic monologue setting. In particular, we build unimodal and multimodal multi-task learning models with sentiment score prediction as the main task and polarity and/or intensity classification as the auxiliary tasks. Our experiments show that sentiment analysis benefits from multi-task learning, and individual modalities differ when conveying the polarity and intensity aspects of sentiment.