Used a recent version of SweFN (2014-12-03).
More tolerant filtering of once-used patterns (only if the frame has a more frequently used pattern).
Slightly more elaborated decisions on object FEs. Introduced heuristics for specific frequently used FEs: Agent and its subtypes indicate the subject while Time, Manner and Place should not be the subject or object. This resolves a lot cases where there are conflicting subject or object FEs.
Provided a partial alignment between BFN and SweFN LUs (a shared multilingual lexicon on top of the monolingual lexicons).
In the result, there are 869 shared semantico-syntactic valence patterns of 483 frames.
Coverage: 77.5% of both BFN and SweFN sentences evoking the shared frames.
Semantico-syntactic types: 541 (339 FEs of type NP, 159 FEs of type Adv, 17 FEs of type VP, 17 FEs of type S and 9 FEs of type QS).
LUs: 3,432 for English (3,350 have linearization) and 1,899 for Swedish (1,789 have linearization).
Aligned LUs: 703 English with 900 Swedish (21% and 47% respectively; used as target verbs in 25% of BFN sentences and 36% of SweFN sentences).
A lot of changes to support LUs as MWEs.
The language-specific implementation LUs is done in a more robust way, and verb constructors are extracted from several RGL modules (DictL, DictionaryL, LexiconL, IrregL, StructuralL) considerably improving the coverage.
In the result, the grammar comprises 723 shared semantico-syntactic verb valence patterns of 426 frames.
LUs: 3,060 for English (2,986 – 97.6% – have linearization) and 1,323 for Swedish (1,252 – 94.6% – have linearization).
FEs of type S (embedded sentence) are distinguished from FEs of type QS (embedded question).
Consequently, verbs of type VS and VQ are distinguished as well.
A bug related to passive voice pattern extraction has been fixed in the SweFN processor.
In the result, there are 720 shared semantico-syntactic valence patterns of 422 frames (~80 changes).
Coverage: 70.2% of BFN sentences and 68.9% of SweFN sentences evoking the shared frames.
Semantico-syntactic types (subcategorized FEs): 483 (306 of type NP, 141 of type Adv, 15 of type S, 13 of type VP and 8 of type QS).
LUs: 3,012 for English (>60 changes) and 1,251 for Swedish (>50 changes).
Extended English and Swedish lexicons.
2,755 (92%) out of 2,996 English verbs and 1,211 (96%) out of 1,257 Swedish verbs have linearization.
Includes an implementation of the Swedish lexicon and an extended implementation of the English lexicon by reusing the monolingual RGL dictionaries.
2,482 (93%) out of 2,663 English verbs and 1,021 (96%) out of 1,063 Swedish verbs have linearization.
Minor bug fixes and improvements.
The set of shared semantico-syntactic valence patterns is slightly reduced resulting in 714 patterns of 421 frames.
The number of LUs has increased as target verbs evoking non-shared patterns that subsume the shared patterns are also included: 2,663 English verbs (linearization for 2,373), and 1,063 Swedish verbs.
First public release following the LREC 2014 and CNL 2014 publications.
Supported languages: English and Swedish.
Comprises 717 shared semantico-syntactic valence patterns (frame building functions) of 423 frames.
Covers 69.4% of sentences in BFN and 68.9% in SweFN evoking the shared frames.
The patterns use 486 syntactically subcategorized FEs (semantico-syntactic types): 308 of type NP, 149 of type Adv, 16 of type S, and 13 of type VP.
LUs: 2,609 English verbs (linearization for 2,321), and 1,041 Swedish verbs. A concrete implementation of the Swedish lexicon is not included yet.
FE – frame element
LU – lexical unit (entry)
MWE – multi-word expression