This dissertation uses digital humanities methods and computational linguistic tools to perform a sentiment analysis on a corpus of 18th century Moravian English personal memoirs. This historical corpus contains important similarities to modern natural language text, requiring similar computational tools and approaches. The methodology of this sentiment analysis, considering the human social and cultural context in the design and implementation process of computational tools has compelling relevance to modern society and computing. The Moravian "Lebenslauf" or memoir represents a unique autobiographical genre, providing rare insight into social and personal lives of individuals living in an 18th century historical community. Sentiment analysis was performed using a custom lexicon and rule-based approach, adapting existing technology to a challenging and previously untested domain. Several different methods of sentiment scoring, analysis, and visualization were used, including narrative trendlines, word frequency, topic identification, and comparison of changing sentiment across time. Results reflect historical Moravian society, some community specific and others timeless and universally relatable as human feelings, and concepts. Split positive and negative sentiment is associated with illness and death, reflecting a positive perception as Moravians compared personal suffering to suffering of Christ during the passion while death meant salvation and "going home". Increasing life expectancy and decreasing young deaths from infectious disease, correlated with increasing negative sentiment, and decreasing positive sentiment associated with illness and death. Analysis by gender reflected unique Moravian gender constructs in 18th century Moravian society, often affecting sentiment. Moravian "blood and wounds" theology, peaking from the 1730s-1750s was detected using topic-specific sentiment vocabulary. Methodology relied on consideration of human social and cultural context and awareness of possible bias and limitations in computational tools during design, implantation, and analysis. These considerations in design allowed for creation of better domain-adapted tools. This methodology is directly relevant and consequential to modern computing problems involving interaction with human context, including serious concerns about ethics, privacy, and bias associated with computational tools and algorithms. Similar methodological approaches considering human context in computational design, implantation, and use can greatly improve future computational technology as it becomes increasingly indispensable to modern society.
|