SpanSn

SpanSn is a software library for span-based text annotation and processing. It provides a formal representation of contiguous spans within token sequences and offers utilities for creating, validating, and comparing spans. The library is intended to support natural language processing tasks that rely on span-level annotations, such as named entity recognition, event extraction, and coreference tagging.

Core concepts include the Span object, which records a start and end index, and the SpanSet or SpanTree collections that manage multiple spans with operations for overlap checks, containment, union, and intersection. The design emphasizes immutability where possible and supports lazy evaluation to minimize recomputation in large documents. Interoperability with tokenizers and label schemes allows seamless integration into existing NLP pipelines.

SpanSn provides bindings for several programming languages and can be integrated with common data formats used in NLP, such as CoNLL-style annotations and JSON-based annotation schemas. It offers serialization options, indexing structures for fast range queries, and adapters for popular NLP frameworks, making it easier to store, filter, and retrieve span data.

Applications range from annotation tooling and corpus management to model training for span-based labeling tasks. The project is maintained by an international community of contributors and has been adopted by various open-source NLP toolchains. The project emphasizes clear documentation and a stable API to support long-term use.