Semantically Rich Tools for Text Exploration: TEI and SEASR

Semantically Rich Tools for Text Exploration: TEI and SEASR
Ashton, Andrew Thomas, Brown University Library, Andrew_Ashton@brown.edu
Much of the existing work of researchers using the Text Encoding Initiative (TEI) guidelines has been made available on the web via processes such as XSLT transformations, which are intended to reproduce a work as an enhanced digital surrogate using HTML, PDF, or another publication format. Recent innovations in scholarly software design offer opportunities to exploit the semantic depth of TEI collections by creating new tools for textual analysis; tools that are designed not specifically for digital publication, but for creating mash-ups, data sets, and expressions of semantic data intended for machine-readability, rather than online readership. To explore these opportunities, the Brown University Library’s Center for Digital Scholarship (STG) and the Brown University Women Writers Project (WWP) - with the support of the National Endowment for the Humanities’ Digital Humanities Start-Up Grant program - are developing a prototype suite of software tools to explore TEI encoded texts in the Software Environment for the Advancement of Scholarly Research (SEASR). A demonstration of these tools, information about their availability, and a discussion about their usefulness to digital humanities projects, offers an opportunity to examine this approach to exploring digital texts.

Exploring Data Through Experimentation and Play
SEASR is a software environment for creating and sharing text and data mining tools. Textual analyses and visualizations created in SEASR can be shared on the web and adapted by other scholars to support their own research. The characteristic of SEASR that permits such exploration and interchange is its modular approach to software architecture. Modularity – the ability to share and reuse discrete pieces of software within a broader framework – enables relatively simple software tools to be combined to produce results that were previously the domain of complex and often proprietary software. Within the SEASR environment, analysis is modeled as a “flow”: a pipeline, or a series of “components” arranged to produce a specific analysis (see figure 1). Each component in a flow is a small piece of software that ingests and processes data, then passes it to the next component in the flow. Each component performs a single function, but through their recombination and sequencing SEASR enables innumerable potential analyses. Once a flow of components is constructed, it can be made publicly available as a web service and reused on other data. The SEASR framework permits components and flows to be integrated into the publication interface of a digital project (as a way of working with project data), or to be used as part of an independent web-based workbench through which a scholar could examine data drawn from multiple projects. In addition, scholars wishing to demonstrate a new analysis or method can create interactive web applications to make their work accessible and reproducible. And because flows can be flexibly combined and altered, they permit much greater latitude in scholarly inquiry, instead of channeling text analysis into predetermined sequences.

Designing a SEASR "flow"

Figure. Designing a SEASR "flow"

Full Size Image

Working with TEI
TEI’s markup logic makes the semantic information encoded in a text tractable for the purposes of extraction, analysis, and reuse. When used in tandem with other data mining and visualization tools available in the SEASR environment (e.g., MONK, Simile, OpenNLP), these TEI components for SEASR offer new venues for exploring questions germane to literary and textual scholars.

As an example, a research scenario involving the WWP texts might include examining the language of specific genres and structures, possibly in relation to other information axes (e.g. time, geography). For example, how does the language of lyric poetry change during the course of the long 18th century? How do women situate themselves as authors (in dedications, letters to the reader, and acknowledgements)? To what extent do women make use of cultural referents from European history in situating their work, and how does this change over time? In pursuing these questions a researcher might use SEASR (either within a project interface such as Women Writers Online, or within a separate workbench application) to:

Extract and separate a collection of texts by genre. In addition to offering text-level genre information (based on a basic differentiation between prose, verse, and drama, which MONK also exploits), like many TEI projects the WWP marks genre-specific structures within the text such as poems, dramatic speeches, letters, recipes, essays). SEASR modules can thus identify and extract single genres of interest (all lyric poetry, for instance) or divide the entire corpus by genre (verse, drama, prose, etc.) at a specified level of granularity.
Extract specific pieces of text (quotes, verse stanzas, dedications) for downstream analysis.
Distill from the selected texts or text pieces the personal names, and separate these by type (references to historical figures, mythological figures, biblical figures; place names; etc.)
Sort a subset of data chronologically.
Pass the data through a component that adds morphosyntactic information to each word.
Generate a visualization such as a stacked area chart for each genre that analyzes changes in the association of certain adjectives with personal names, differentiated by gender.
Conclusion
The primary outcome of this exploration is a set of SEASR components designed to extract and process many commonly encoded semantic features using the TEI P5 Guidelines. This poster session will include a demonstration of the TEI tools for SEASR, as well as a discussion of the challenges of developing broadly applicable scholarly software tools such as these. Specifically, it will examine the difficulties and opportunities in designing software to work within a broad and diverse community of practice, such as the TEI community. The software, along with a set of sample analyses, will be available for download in a publicly available repository. In addition, a white paper, including an analysis of likely use-cases for these tools, will be available at the conclusion of the grant.