On December 4th, 2023, the NULab for Texts, Maps, and Networks hosted an online event entitled “Remembering Hester Piozzi’s Streatham: Place and Sentiment in Eighteenth-century Letters”, with presenter Dr. Cassie Ulph, Digital Development Officer at the University of Leeds. Dr. Ulph shared her recent project in which she mapped sentiment in relation to place through the letters of the eighteenth-century author and salonnière Hester Thrale Piozzi (c. 1740–1821). This work required a combination of machine learning approaches, entity recognition, sentiment analysis, geospatial mapping and literary-historical textual scholarship. During this online event, Dr. Ulph described the ongoing project and its challenges, as well as the future possibilities of her cross-disciplinary work.
Dr. Ulph established the relationship between authorship and space through evocative questions; for example, exactly which place receives the honor of being dubbed “the home of” a particular author, especially when that author lived in multiple homes? Which site becomes the site? Must there only be one site? Dr. Ulph evaluated these questions in connection to diarist, travel writer, biographer, historian, and lexicographer Hester Thrale Piozzi. Piozzi has been somewhat neglected, her reputation dwarfed at times by her famous friendship with Johnson. Though Piozzi occupied a variety of spaces that all bore impact on her writings, she did not receive the official honor of any historical place being singularly associated with her, a lack of recognition that Dr. Ulph claimed is connected to her literary neglect. A problem that Dr. Ulph faced in the attempt to recover Piozzi’s literary-spatial traces is that Streatham Park House, where Piozzi resided for a long period, no longer exists; the only historical commemoration devoted to what is left of that space is dedicated to Samuel Johnson. So, Dr. Ulph and her team arrived at the question that would drive this project: How can a place persist in memory and narrative even when the material site is gone?
This eight-week preliminary study treated Piozzi’s writing as a site of recovery for a sense of place in connection with narrative form and emotion. Through this pilot project, Dr. Ulph and her team engaged the idea that literary documents can contain spatial and emotional data, using Hester Piozzi’s letters to investigate this theory. In addition, this project sought to test whether machine learning and natural language processing methods are suitable for a textual corpus of this period and genre, that is, 18th-century letters. Specifically, they used Aspect-Based Sentiment Analysis (ABSA), a new subtype of sentiment analysis that allows for more customization and fine-tuning within machine learning. In other words, instead of performing a sentiment analysis over the whole unit of text, this method limited analysis to text paragraphs, where specific topics could be identified. Thus, the overall goals for the project were threefold: to transform the unstructured data of the letters into structured data that could be analyzed; to identify geographical locations within those letters that would be the subject of the aspect-based sentiment analysis; and to use that analysis to map sentiments associated with the identified locations and observe patterns over time.
The corpus at this stage of the project consisted of letters written by Piozzi from 1784-1821. These letters began at a point in Piozzi’s life when her relationship to Streatham was very complex, coinciding with the death of her husband Henry Thrale and the budding of a new relationship with her daughter’s music teacher, Gabriel Piozzi, whom she later married. The data set was inherently one-sided, as the letters only represent Hester’s correspondence. Other difficulties that arose in the text analysis process were the variations in how Piozzi identified places. For example, sometimes “Cumberland” signified the place, other times it referenced the person the Duke of Cumberland; however, the machine flags both textual entries as one and the same. Similarly, there was one location in Wales that Piozzi spelled eight or nine different ways. Piozzi also occasionally referred to locations by their classical names even if she was referencing them contemporarily, which likewise complicated the process. Identifying and resolving these problems was a core accomplishment of the pilot.
The results of this pilot study demonstrated that neutral sentiments (80%) were the most commonly associated with place generally, but that positive (12%) and negative (8%) associations were present as well. This textual analysis also identified the key locations that were the most associated with sentiment in her letters, including: Bath, Brynbella, London, and of course, Streatham. For both Bath and Brynbella, Piozzi’s positive associations outweighed the negative associations by fairly large margins. For Bath, Piozzi’s positive sentiments comprised approximately 65% of the data, with negative sentiments at 35%. Similarly, Brynbella boasted 85% positive associations, and only 15% negative. However, London and Streatham maintain thinner margins, with London at approximately 52% positive sentiment and 48% negative, and Streatham maintaining roughly 46% positive association and 58% negative association. For London and Streatham, these positive and negative associations fluctuated consistently over time, rather than presenting a steady arc from one emotion to the other.
Dr. Ulph emphasized that these results were all first pass data and that this process has yet to be fine-tuned. Importantly, this pilot project served to prove the connection between emotion and place in narrative. Looking ahead, Dr. Ulph asserts, “I really want to do it again, and I want to do it better” so that she and her team may “harness another dimension of what literature can tell us about how place is experienced, processed, and remembered.”