Sample DELPH-IN outputs using ERG
15-June-2011

Guide to annotations
--------------------
For each sentence, four different annotations are given, showing distinct
perspectives on the parser analysis which was manually identified from the
parse forest of analyses produced by the parser:
 - Derivation tree showing which rules and which lexical entries were applied.
 - Labeled parse tree showing constituent structure, isomorphic to derivation.
 - MRS showing the semantic representation composed by the parser.
 - Elementary dependencies showing a simplified view of the semantics

Derivation tree
 Each level of parenthesized nesting indicates a syntactic constituent, where
 - the first integer is the unique edge number assigned in the parse chart
 - the next identifier is the name of the syntactic rule or lexical entry
 - the next (decimal) number is the likelihood from the statistical model
 - the last two digits show the token span of the constituent, starting at 0
 The leaves of each constituent are the strings showing the surface forms of
 the words comprising that constituent, including punctuation if any.

Labeled parse tree
 Each level of nesting again indicates a syntactic constituent, with the
 labels drawn from a coarse-grained inventory of mostly familiar category
 terms, augmented with a "/" suffix on those categories which contain or
 head a phrase containing a gap.  These parse labels are not manipulated
 by the parser during parsing but applied afterwards on the basis of the
 full feature structures. Each parse tree is isomorphic to the
 corresponding derivation tree.

MRS
 The Minimal Recursion Semantics output shown here is the "indexed" format
 which uses linear ordering of arguments for each predicate to encode roles,
 where the first argument has the role label ARG0, the second has ARG1, etc.
 Each elementary predication records its handle (e.g. "h7") used in scope
 constraints, its predicate name (e.g. _copy_n_of_rel), the character span
 of the constituent in the input sentence that corresponds to this 
 predication, and a list of variables, one for each of its semantic arguments,
 including an inherent argument for each  predication.  Variables of four 
 types are used: `h' for handles, `e' for  events and states (and also for 
 adverbial predications), `x' for referential indices, and `i' for individuals
 which are the underspecification of events and indices.  Where the grammar
 assigned constraints on properties of event and index variables, these are
 recorded following the variable in the predication where it is the inherent
 argument.  For `x' variables (referential indices), the properties are (in
 order, and when known) Person (1,2,3), Number (SG,PL), Gender (M,F,N), 
 Pronoun-type (STD_PRON [standard pronoun], REFL [reflexive]), Individuated
 (+,- [where singular count nouns are +, while plural and mass nouns are -]).
 For `e' (event/state) variables, the properties are Sentence Force (PROP 
 [proposition], QUES [question], COMM [command]), Tense (PRES, PAST, FUT),
 Mood (INDICATIVE, SUBJUNCTIVE), Progressive (+,-), Perfect (+,-).

Elementary dependencies
 Omitting scopal constraints, each dependency records its characteristic
 variable (its inherent argument), followed by the predicate name (suppressing
 the `_rel' suffix seen in the MRS), followed by the character span of the
 corresponding constituent in the input sentence, followed by a list of
 role-argument pairs, where the argument consists of its variable name and
 predicate name.  The inclusion of the variable with each argument preserves
 clarity in cases where the same semantic predicate is introduced multiple
 times in a single sentence.