Syntactic Dependency Relationships and the Extraction of Grammatical Triples
Digital Humanities NLP

Syntactic Dependency Relationships and the Extraction of Grammatical Triples

In recent research, extracted triples have become increasingly important for the analysis of large, textual corpora. By giving insight into the linguistic features of a corpus, extracted triples have supported rich interpretations of some of the most relevant problems of our time. They have been used to track the development of anti-vaccination narratives originating on “Mommy Blogs” (Tangherlini et. al, 2016), descriptions of climate change across online news sources (Alashri et al., 2018), to model scientific literature on COVID-19 (Papadopoulos et. al, 2020), and to trace how Parliamentary discourses on the rise of peoples’ rights have changed over time and may be influencing the politics of our present moment (Guldi, 2022).

The growing importance of triples extraction for analyzing large corpora has put the quality of extracted triples under new scrutiny, however. Triples outputs are known to have large amounts of erroneous triples. The need for a method of triples extraction designed for textual analysis is clear: the extraction of erroneous triples poses a risk because erroneous triples can be unfactual and even analogous to misinformation. Disciplines such as history, literature, and the social sciences, rely on accurate representations of events. In some cases, misrepresentations of language can be as problematic as describing a historical event that never occurred.

The present research proposes a method of triples extraction, posextract, which has been designed to meet the increasing need for high-accuracy triples outputs for the analysis of text. We propose a solution aimed at reducing errors related to: a) ungrammatical extractions; b) double counting; and c) the missed detection of triples.

Software package coming soon Article coming soon

You may also like

See all Digital Humanities
PANGeA

PANGeA

PANGeA is a system that uses large language models (LLMs) to create narrative content for turn-based RPGs based on game designers' high-level criteria. It introduces a novel validation system for handling free-form text input during development and gameplay, employing "self-reflection" techniques, enabling small/local LLMs to perform comparably to foundational models. It enriches player-NPC interactions by generating personality-biased non-playable characters (NPCs). It improves AI accuracy through crowdsourcing mechanics. PANGeA houses a server with a custom memory system that provides context for LLM generation. The server's REST interface enables integration with any game engine.

GAME-KG

GAME-KG

Knowledge graphs (KGs) can augment large language models (LLMs) while also providing an explainable set of facts that can be inspected by a human. Explainability is valuable for fields that may otherwise avoid LLMs due to hallucinations, such as human trafficking analysis. Creating KGs poses challenges, however. KGs parsed from documents may include explicit connections (those directly stated in a document) but miss implicit connections (those evident to a human, but not directly stated). This research introduces GAME-KG, an approach to modifying explicit and implicit KG connections by crowdsourcing feedback through video games.

Course Materials: Foundations and Applications of Humanities Analytics

Course Materials: Foundations and Applications of Humanities Analytics

Computational methods allow researchers to systematically analyze and interpret large volumes of social, political, and cultural data, uncovering underlying patterns and insights at scale. These course materials, made for the Santa Fe Institute, are designed to equip humanities researchers with computational and quantitative tools. The course aims to foster a supportive community, build practical skills, and diversify the field of humanities analytics by welcoming participants from various backgrounds and stages of their academic careers.