Abstract
The Adult Attachment Interview (AAI) is a protocol-based, semi-structured interview method widely used to measure adults’ states of mind with respect to attachment. Recently, transcripts of this interview have been used to code secure base script knowledge, which is script-like knowledge related to the way parents dealt with their distress during childhood (ie., child went to parent for comfort, parent provided instrumental and emotional support, child went back to play). Manually coding the verbatim transcripts is labor-intensive and requires a lot of centralized training. The potential integration of machine learning and natural language processing (NLP) techniques may automate certain aspects of AAI analysis, potentially optimizing the process. The aim of this research project is to explore the practical application of these technologies in analyzing AAI transcripts.
The project uses data from a pooled set of 12 studies originating from four countries. Upon reviewing the 1,410 AAI transcripts in this set (conducted in three languages), notable discrepancies in the administration of the interviews emerged, some of which may affect the suitability of the interview to assess secure base script knowledge. The first focus of this research project is therefore to develop a model to automatically assess the quality of the transcripts, first for English studies and then for all studies and languages. This model will prioritize evaluating interview characteristics, including instances of unintelligibility and non-adherence to the prescribed AAI protocol.
As a next step in the project, employing sentiment analysis will enable an investigation into the correlation between participant-provided adjectives and their corresponding narratives. Finally, this research project will explore the possibility of automatically coding secure base script knowledge in AAI transcripts.
By combining technological advances with nuanced human insights, this research project not only provides a pathway toward research studies at scale, but also presents an opportunity to achieve a deeper understanding of emotional and cognitive dimensions within attachment narratives.
The project uses data from a pooled set of 12 studies originating from four countries. Upon reviewing the 1,410 AAI transcripts in this set (conducted in three languages), notable discrepancies in the administration of the interviews emerged, some of which may affect the suitability of the interview to assess secure base script knowledge. The first focus of this research project is therefore to develop a model to automatically assess the quality of the transcripts, first for English studies and then for all studies and languages. This model will prioritize evaluating interview characteristics, including instances of unintelligibility and non-adherence to the prescribed AAI protocol.
As a next step in the project, employing sentiment analysis will enable an investigation into the correlation between participant-provided adjectives and their corresponding narratives. Finally, this research project will explore the possibility of automatically coding secure base script knowledge in AAI transcripts.
By combining technological advances with nuanced human insights, this research project not only provides a pathway toward research studies at scale, but also presents an opportunity to achieve a deeper understanding of emotional and cognitive dimensions within attachment narratives.
| Original language | English |
|---|---|
| Publication status | Published - 11 Jan 2024 |
| Event | HumanCLAIM Workshop: The Human Perspective on Cross-Lingual AI Models - Amsterdam, Netherlands Duration: 10 Jan 2024 → 11 Jan 2024 https://clap-lab.github.io/workshop |
Workshop
| Workshop | HumanCLAIM Workshop |
|---|---|
| Country/Territory | Netherlands |
| City | Amsterdam |
| Period | 10/01/24 → 11/01/24 |
| Internet address |
Keywords
- Adult Attachment Interview
- secure base script knowledge
- natural language processing
- machine learning