PMB release 5.1.0. Compared to release 5.0.1, the following addition/changes have been made:

* updating the data, note the train, test, validation sets are different from previous releases.
* removing Japanese data
* manually correcting long documents and changing the strategy for generating long test set

PMB release 5.0.1. Compared to release 5.0.0, the following addition/changes have been made:

* removing the ill-formed long documents
* fixing the problem with newline symbols

PMB release 5.0.0. Compared to release 4.0.0, the following addition/changes have been made:

* Switching from clause notation to sequence notation (SBN) for meaning representation
* Adding Japanese and Chinese data to the existing languages
* Replacing Bronze data with Copper data for nl, de, it and ja
* Reorganizing the train/development/test sets based on length distribution
* Reconstruct the structure of split sets and reformat the .sbn file in the format of ID/raw/sbn

PMB release 4.0.0. Compared to release 3.0.0, the following additions/changes have been made:

* Substantially more gold annotated meaning representations (DRSs): +2300 for English, +1000 for German, +700 for Italian and +450 for Dutch
* Substantially more silver/bronze DRSs: +30k/10k for English, +1k/30k for German, +1.3k/36k for Italian and +0.15k/7k for Dutch
* Inclusion of SBN format (simplified box notation) for each document
* Dutch and Italian now also have a gold standard training set for semantic parsing, instead of only a dev and test set
* Propositional complements have a slightly different representation than in previous releases. The thematic role referring to the proposition directly denotes the box label of the proposition, and the proposition is not embedded anymore.

PMB release 3.0.0. Compared to release 2.2.0, the following additions/changes have been made:

* Substantially more (ca. 3500) gold annotated meaning representations (DRSs): +2400 for English, +500 for German, +300 for Italian, +300 for Dutch.
* The way discourse relations are represented has changed significantly. Instead of {B REL B1 B2, B DRS B1, B DRS B2} we have now the simpler B1 REL B2.
* Explicit presupposition discourse relations are now included in DRS clause notation (PRESUPPOSITION)
* Propositional complements are now represented by the ATTRIBUTION discourse relation in DRS clause notation. Instead of {B PRP P B1} we have now {B proposition n.01 P, B ATTRIBUTION B1}.
* Many new documents added from three corpora: SICK, GMB (subset of actual GMB release), and Incidents.
* The .drs.box files are not included in this release.
* Documents come with Combinatory Categorial Grammar (CCG)-style derivations. The derivations are presented in .der.xml and .parse.tags files
* A script has been added that can extract individual layers with desired statuses (e.g. gold and silver semtags) and put them in CoNLL format. For details see src/extract_conll.py.
* German has a train/dev/test split for semantic parsing, instead of only dev/test
* The data in data/ is split by language (e.g. data/en/gold/..)

PMB release 2.2.0. Compared to release 2.1.0, the following additions/changes have been made:

* The addition of documents that are not manually checked, but fully automatically produced (bronze, 120622 en, 102998 de, 60209 it, 20554 nl)
* More documents that are fully manually checked (gold, 1374 en, 299 de, 139 it, 47 nl)
* Fewer documents that are partially manually checked for English (silver, 3343) due to a more strict selection process
* More documents that are partially manually checked for non-English (silver, 3433 de, 2070 it, 718 nl)
* There was a bug in resolving tense for multi-sentence documents. Now each sentence has its own tense.
* The CCG categories are included in the sentence derivation
* Spatial relations are added in the DRS; see: http://pmb.let.rug.nl/drs.php

PMB release 2.1.0. Compared to release 2.0.0, the following additions/changes have been made:

* More documents that are fully manually checked (gold, 630 en, 127 de, 67 it, 59 nl)
* More documents that are partially manually checked (silver, 5010 en, 77 de, 40 it, 15 nl)
* All DRSs in clause format are checked by a format checker (referee) and are in valid format, 
  there were 2 invalid gold DRSs in 2.0.0
* The recommendation to use part 00, 10, 20 and 30 as development set
* Naming of box-variables is made more consistent

PMB release 2.0.0. Compared to release 1.0.0, the following additions/changes have been made:

* More documents that are fully manually checked (1876 en, 407 de, 181 it, 133 nl)
* Documents are added from "Questions", the corpora used at QA@CLEF-2004 (http://clef-qa.fbk.eu/2004/resources.html)
* Addition of "silver data", documents that are partially manually checked (66693 en, 611 de, 266 it, 192 nl)
* Changes in general non-logical symbols (entity.n.01 instead of thing.n.12)
* Tuples are now called clauses
* Order of clauses in the DRSs is better aligned with the surface text
* Improved analysis of copula, wh-questions, intensifiers, vague quantifiers
* Proper names in DRSs keep their language-specific forms
