DELPH-IN (Deep Linguistic Processing with HPSG - INitiative) is an international collaboration of computational linguists who develop open-source tools for deep linguistic processing of human language. The initiative aims to combine linguistic and statistical methods to computationally understand the meaning of texts and utterances. Its tools are based on two formalisms: head-driven phrase structure grammar (HPSG) and minimal recursion semantics (MRS). Since 2005, DELPH-IN has held an annual summit, an unconference where participants share work, seek feedback, and agree on standards.
Technologies and resources
DELPH-IN provides a suite of tools for grammar engineering and parsing. The Linguistic Knowledge Builder (LKB) is a grammar engineering environment for constructing HPSG grammars. The PET parser (Platform for Experimentation with efficient HPSG processing Techniques) produces HPSG parse trees with MRS outputs. The ACE processor (Answer Constraint Engine) efficiently processes DELPH-IN grammars and can also generate natural language sentences. The LOGON infrastructure supports transfer-based machine translation, offering hybrid rule-based and stochastic translations.
Grammars and linguistic resources
DELPH-IN maintains a catalogue of hand-crafted computational HPSG grammars for multiple languages. The LinGO Grammar Matrix is an open-source starter kit for rapid prototyping of broad-coverage grammars, providing a library of common language phenomena. The CLIMB libraries (Comparative Libraries of Implementations with Matrix Basis) extend the Grammar Matrix to test alternative analyses across languages.
Corpora and treebanks
DELPH-IN also provides annotated corpora. The MRS Test Suite contains representative sentences for testing MRS phenomena, available in over ten languages. WikiWoods is a parsed corpus with rich syntacto-semantic annotations for the English Wikipedia. DeepBank annotates the one million words of 1989 Wall Street Journal text (the same as the Penn Treebank) using the English Resource Grammar. The Cathedral and the Bazaar compilation, based on Eric Raymond's essay, serves as a multilingual test suite.
Community and usability
DELPH-IN's open-source culture benefits the Natural language processing community, but usability has been a challenge for newcomers. Developers are working to improve documentation and tutorials. The initiative is related to broader efforts in Artificial intelligence and Machine learning, though it focuses on symbolic and hybrid approaches rather than Neural network methods. Its resources are used in research and applications requiring deep semantic analysis, complementing shallow techniques like those in Large language models.
See also
- Head-driven Phrase Structure Grammar
- Minimal Recursion Semantics