System Demonstrations

This page lists system demonstrations from several JHU projects.

LOME: Large Ontology Multilingual Extraction (2021)

LOME is a system for information extraction. It identifies events and entities and predicting structural information around them, like types, coreference, and temporal relations. In addition, it can process raw text in any language. We release our system on Docker and we have lite-version in a web demo here. We also have a system description paper which describes and benchmarks the performance of individual components.

For now, we are only releasing a Docker container. It can be found on Dockerhub and instructions are provided there.

InFillmore: Frame-Guided Language Generation with Bidirectional Context (2021)

Infillmore is a system for interactive story writing using guided text generation based on Frame Semantics. The system accepts bidirectional document context and infills text according to FrameNet frames inferred by the model and/or chosen by a user. We provide a Google Colab-based web demo here. A paper describing the method plus experiments and example use cases is found here.

SchemaBlocks: Human Schema Curation via Causal Association Rule Mining (2022)

SchemaBlocks is an annotation toolkit for event schemas, or structured knowledge sources defining typical real-world scenarios (e.g., going to an airport). Different components of a schema, such as events, relations, and entities, are represented as nested blocks, which provides an intuitive and accessible alternative to annotating directly in the domain-specific language (e.g. XML or JSON). We provide a web-based demo of the system here. Combining SchemaBlocks with a semi-automatic schema induction method, we collect and evaluate 232 event schemas, which we describe in detail in our paper.