V2.2.1.1 - update requirements

This is a maintenance release with updated requirements.txt

V2.2.0 - compatibility updates and new settings

• bump depedit to 2.3.0 (with variable definition support, see https://corpling.uis.georgetown.edu/depedit/)
• python 3.8 compatibility (timing module)
• support flair 0.6.1 (version 0.4.5--0.6.1 APIs now supported, given sequence tagger model trained with appropriate version)
• add speaker compatible check for global 1st/2nd speaker coref consistency in input with gold speaker information
• expose sentence object to coref rules (e.g. samesentence / !samesentence constraints in a model's coref_rules.tab)
• added configuration options:
  • no_overlap setting to allow/forbid nested markable coreference
  • force_single_type_chains setting to enforce majority entity type in a mixed type chain
• bug fixes

V2.1.2 - support newer versions of flair

Sequence labeler syntax now supports newer versions of flair too

V2.1.1 - library updates and some additions

• support flair 0.45
• support crf_suite-based sequence tagging (see utils/train_crf.py)
• new config.ini options
  • sequencer_override_thresh - allows confidence threshold for sequencer entity classification to ignore KB
  • default_atomic_named_entities - determines entity types in entities.tab which are assumed to be atomic (often places)
  • max_suffix_length for substring based morphological entity type feature

V2.1.0.2 - Neural entity sequence labeling (bugfix 2)

Bugfix 2 for 2.1.0

V2.1.0.1 - Neural entity sequence labeling (bugfix)

Bug fix for 2.1.0

V2.1.0 - Neural entity sequence labeling

This major update introduces many changes, most importantly:

• Support for neural sequence labeling in entity classification - note that the knowledge base still constrains possible entity types for known lemmas, but a sequence tagger can now disambiguate options and classify pronouns
• Support for XGBoost as a coref classifier
• Update English model to use Universal Dependencies as input
• New configuration parameters in model config.ini
• Upgraded DepEdit with new parse post-editing rule possibilities
• Various bug fixes

V2.0.2

Minor adjustment to conll_sent output format (can now include entities)

V2.0.1.0 - Bug fix release

Bugfix release and minor additions

• Fix incorrect module import in Python 3 when not running standalone (#82)
• Fix incorrect document name setting to 'untitled' when not running standalone (#83)
• Incorrect ordering of coreference chain in certain cataphora cases fixed
• options.tab now optional
• Added remove_infixes and non_extend_pos settings for interfixed core_text expansions (e.g. for Semitic construct state articles)
• Better Python 2/3 compatibility

V2.0.0.2

Major new feature release:

• Python 3 compatibility
• Adds support for pluggable machine learning classifiers
• Multithreading - concurrent batch processing of multiple documents
• New model files:
  • Added similar.tab to models for semantic similarity to OOV items
  • Added nominalizations.tab for irregular verb nominalizations used in verbal antecedent matching
  • Made all model files optional except config.ini and coref_rules.tab
• Support conllu input with multitoken units and virtual tokens from enhanced trees
• Added option to use sentence type (s_type) annotations (e.g. declarative, question, imperative)
• Added negated head and negated parent features (used e.g. to decrease coref likelihood for didn't see [a dog])
• New DepEdit version with new operators
• New utils/ folder with tools to create model files and classifiers
• Added a starter model udx as a building block for new models based on UD data