TL;DR
This paper analyzes the linguistic patterns of a Russian blogger who died by suicide, using the LIWC software to track changes in their language in the weeks and months leading up to their death. The study highlights the potential of computational linguistics to offer insights into the psychological state of individuals at risk of suicide, particularly by examining the dynamics of their idiostyle over time.
What Problem Does This Paper Address?
Suicide is a significant global health concern, and identifying individuals at high risk is a critical challenge. While computational linguistics has shown promise in analyzing online texts for suicide risk, most research has focused on English-language data and often relies on manual annotation, which can be subjective. Furthermore, there's a limited understanding of how linguistic patterns change over time as an individual approaches suicide. This paper addresses this gap by analyzing the dynamic idiostyle of a Russian-speaking blogger who died by suicide, aiming to provide a more objective and nuanced understanding of the linguistic markers associated with suicidal ideation and behavior.
Methods
The researchers employed the Linguistic Inquiry and Word Count (LIWC) software, specifically its Russian version, to analyze the blogger's texts. LIWC categorizes words into over 80 psychologically meaningful variables, such as emotion, self-reference, and social processes. In addition to the standard LIWC dictionaries, the authors developed custom Russian dictionaries to capture specific linguistic nuances.
A key methodological aspect was the correlation analysis performed between the frequency of LIWC variables and the number of days prior to the author's death. This allowed them to identify linguistic parameters that showed a significant relationship with the proximity of suicide. Data visualization, through graphs showing the intensity of posting and the frequency of specific linguistic parameters over time, was also crucial for understanding the dynamics.
Datasets
The primary dataset for this study consisted of blog entries from two LiveJournal blogs of a Russian software engineer who died by suicide. These blogs, "light_medelis" and "Im_diary," were part of the RusSuiCorpus, a collection of texts from Russian individuals who died by suicide. The study focused on texts written within a specific timeframe leading up to the author's death. No new benchmark datasets were introduced.
Key Insights and Ideas
- Dynamic Idiostyle is Key: The study emphasizes that analyzing the dynamics of an individual's language, rather than just static patterns, is crucial for understanding their psychological state and potential suicidal ideation.
- Pronoun Shifts are Significant: A notable finding is the strong correlation between the number of days prior to death and the frequency of pronouns, particularly personal pronouns ("I") and overall pronouns. This suggests a shift in self-focus or social engagement as suicide approaches.
- Social Withdrawal Indicators: The decrease in words describing social processes in the period leading up to suicide aligns with sociological theories of suicidogenesis, suggesting increasing social isolation.
- Verbs and Conjunctions Show Varied Trends: A drop in common verbs was observed in both blogs, while the behavior of conjunctions varied between the two diaries, with one showing an increase (largely due to "and") and the other a decrease.
- Positive Emotions: A Complex Picture: While some studies suggest a decrease in positive emotion words, this study found that in one blog, positive emotion words increased in the final period, potentially reflecting a psychological shift or acceptance of the decision to die.
Key Results
The correlation analysis revealed that 8 out of 66 chosen text parameters correlated with the number of days prior to death in the personal diary (PD1), and 9 in the other blog (PD2). Specifically, parameters like common verbs, personal pronouns, overall pronouns, and words describing social processes showed positive correlations, meaning their frequency dropped as the date of death approached.
Visualizations showed distinct periods of peaks and drops in posting intensity for both blogs. The study also highlighted that as the date of suicide approached, the proportion of words describing social processes dropped, consistent with theories of increasing social isolation. The percentage of words describing the writer ("Ego") and the pronoun "I" remained consistently high in peak periods for PD1, but dropped significantly in the final period.
Limitations and Future Work
The authors acknowledge that this is a case study of a single individual who publicly discussed their depression and suicidal intentions. They emphasize the need for future work to compare this individual's writing with blogs of people who did not die by suicide, as well as those who died by suicide without discussing their plans. This would help identify universal linguistic patterns.
Future work should also focus on:
- Expanding the Russian LIWC dictionary and creating specialized dictionaries for suicide-related studies.
- Incorporating linguistic complexity and syntactic parameters into the analysis.
- Performing multivariate analysis to better understand how different linguistic parameters predict the time course of suicide.
- Investigating the dynamics of text parameters over different time periods (e.g., a year prior to death) and considering factors beyond just the number of days prior to death, such as events affecting posting intensity.
- Analyzing texts from different genres and subgenres, as linguistic patterns may vary.
Summary updated