Towards Automated Multilingual Autobiographical Interview Scoring: A Machine Translation Approach

The Autobiographical Interview (AI) is one of the most used method for quantifying episodic memory from memory narratives. Its manual scoring, however, is limited and demands considerable time and effort. These issues have been recently addressed by the use of natural language processing (NLP) allowing for automatic, unbiased and less time-consuming scoring of memory performance in the context of the AI. Since existing NLP models for AI analysis are primarily designed for English narratives, this study investigated whether machine translation could enable their effective use on German narratives without significantly reducing classification accuracy. We evaluated model performance on a total of 1,920 German narratives generated by 258 individuals across two datasets. We further provided an open-source analysis code. We demonstrate that correlations between predicted and manually rated internal and external details were highly significant and comparable to existing studies performed with English narratives. Machine translation of German narratives allows for high classification accuracy and low frequency of misclassification without a meaningful impairment of model performance compared to English narratives. We observed no performance differences across translation models, indicating lightweight models perform equally well. Our findings support the use of NLP as an alternative cost- and time-efficient approach for automated AI scoring in non-English-speaking populations. Our results highlight the potential of language models for large-scale memory research across languages. We conclude that NLP enables more detailed analyses of episodic memory narratives beyond traditional manual scoring.