parallelism

Description of the dataset ted-metrics

From the PDTB-3, we excluded:

• relations involving split connectives like "on the one hand ... on the other hand ...",
• relations displaying one or two split text-spans,
• relations involving an overlap of the two text-spans (or one text-span containing the other),
• relations whose text-spans were made of two or more sentence,
• relation involving an act of speech,
• AltLex, AltlexC, Hypophora, EntRel and NoRel relations.

We end up with 39271 relations, available in the pandas DataFrame ted-metrics, whose columns are:

• INDEX: an primary key for each discourse relation
• RelationType: Explicit or Implicit in PTDB-3's terminology
• Conn: the connective
• Provenance: refer to PDTB-3 manual
• Layout: ORDERED (Arg1 and then Arg2) or SWITCHED (Arg2 and then Arg1)
• Arg1 & Arg2: textspans
• 1Rel, 2Rel, 3Rel: PDTB-3 three-level taxonomy of discourse relations
• NumTokensArgX: Number of tokens in the textspan X
• NxGX: NetworkX graph of the constituency tree of textspan X
• SizeX: number of nodes in the simplified (cf .ipynb files for the definition of simplified) constituency tree of textspan X
• TED: Tree Edit Distance value
• MatchSeq: mapping sequence between the trees at format [ (1,1), (2,3), (3,None) ...]
• matched: number of mapped nodes in MatchSeq i.e. tuples without None
• deleted: number of deleted nodes i.e. tuples with None on the right side
• inserted: number of inserted nodes i.e. tuples with None on the left side

The two following columns refer to Sanders et al. 2018 terminology:

• Polarity: pos (positive) or neg (negative)
• Operation: add (additive) or cau (causal) or add/cau (in case a manual review is required)

Desciption of the .ipynb files

First constituency-trees allows to display and save constituency trees obtained thanks to Stanza parser. We recommend to have a look at the documentation to understand how this class-oriented library actually works.

Second tree-edit-distance allows to compute the TED value thanks to the edist library. We recommend to have a look at the documentation and tutorials provided to understand how this class-oriented library actually works.

The mapping from PDTB-3 to CCR theory of Sanders et al. 2018