Systematic Literature Review for Computational Detection of Anthropomorphic Language

With the rise of the large language models (LLMs), there has been an increased focus on the anthropomorphic characterization of AI research, journalism, and public discussion. Although anthropomorphism (giving human attributes to non-human entities) has been a topic of decades-long study for psychologists, anthropomorphic framing in natural language has recently become an independent area of research. However, anthropomorphic framing is expressed in the observable linguistic characteristics including certain pronouns, verbs and syntactic structures, which makes natural language processing (NLP) tools very useful for the detection and thus, also a major interest of research compared to theoretical analyses. This systematic literature review provides a review of previous work on the computational detection of anthropomorphic framing, including both key works on the problem and the background literature to motivate the problem. In the review, the authors employ a strict methodology and extensive quality assessment to categorize the existing studies into 5 methodological categories: crowdsourcing and classification, automated metric based detection, supervised and unsupervised detection by classifiers, taxonomy and intervention. This review offers insights into the present status of research and offers researchers and practitioners of AI a reference point for designing more accurate detection techniques and more trustful and transparent human-AI interactions. The review also reveals a number of research gaps that continue to hinder progress towards understanding and mitigating the impact of anthropomorphic framing, such as the absence of a large-scale publicly available dataset, limited cross-lingual approaches, and the absence of benchmarking of conversational data generated by LLMs.