rustomata

Framework for (weighted) automata with storage

main functionalities:

• parsing of LCFRSs using their Chomsky-Schützenberger characterisation
• parsing of PMCFGs/MCFGs/LCFRSs and CFGs using their automata characterisation
• approximation of MCFGs/LCFRSs and CFGs using several approximation strategies
• coarse-to-fine parsing of MCFGs/LCFRSs and CFGs using their automata characterisation (under construction)

quick start

• create a grammar file

  cat <<EOF > example.mcfg
  initial: [S]
  S → [[Var 0 0, Var 1 0, Var 0 1, Var 1 1]] (A, B)
  A → [[T a, Var 0 0],  [T c, Var 0 1]     ] (A   )   # 0.5
  A → [[],  []                             ] (    )   # 0.5
  B → [[T b, Var 0 0],  [T d, Var 0 1]     ] (B   )   # 0.5
  B → [[],  []                             ] (    )   # 0.5
  EOF

• construct a tree-stack automaton from the given grammar, print out that automaton

  cargo run mcfg automaton grammar.gr

• construct a tree-stack automaton from the given grammar and recognise the word "a a b c c d", print out a best (maximum weight) accepting configuration (if there is one)

  echo "a a b c c d" | cargo run mcfg parse grammar.gr

the grammar formats

Rustomata can deal with multiple context-free grammars (short: MCFGs) and context-free grammars (short: CFGs). MCFGs are expressively equivalent to linear context-free rewriting systems (LCFRSs) and simple range concatenation grammars (sRCG). The algorithms of rustomata use automata (tree-stack automata, short: TSA, for PMCFGs; and pushdown automata, short: PDA, for CFGs) or the Chomsky-Schützenberger representation to deal with grammars. The weights are (for now) assumed to be from the algebra (ℝ₊, ⋅, 1) of non-negative reals with multiplication; any parser considers greater values to be better.

an example of an MCFG (in the notation of an sRCG):

initial non-terminals: only S

productions:
S(x₁y₁x₂y₂) ← A(x₁, x₂) B(y₁, y₂)  with weight 1
A(ax₁, cx₂) ← A(x₁, x₂)            with weight 0.5
A(ε, ε)     ← ε                    with weight 0.5
B(by₁, dy₂) ← B(y₁, y₂)            with weight 0.5
B(ε, ε)     ← ε                    with weight 0.5

• the same grammar in rustomata's notation:

  initial: [S]
  
  S → [[Var 0 0, Var 1 0, Var 0 1, Var 1 1]] (A, B)
  A → [[T a, Var 0 0],  [T c, Var 0 1]     ] (A   )   # 0.5
  A → [[],  []                             ] (    )   # 0.5
  B → [[T b, Var 0 0],  [T d, Var 0 1]     ] (B   )   # 0.5
  B → [[],  []                             ] (    )   # 0.5
  

• Lines that contain no non-whitespace characters are ignored by the parser.
• Lines whose first non-whitespace character is a % are also ignored (comments).
• A single production may not contain newline characters.
• A production may be followed by a comment (starting with %).
• The weight definition (e.g. # 1.0) may be omitted. Rustomata then assumes a weight of 1.0.

an example of a CFG:

initial non-terminals: S

S → a S b    with weight 0.4
S → ε        with weight 0.6

• the same grammar in rustomata's notation:

  initial: [S]
  
  S → [T a, Nt A, T b]  # 0.4
  S → []                # 0.6
  

• The parser specifics of MCFGs also apply for CFGs.

constructing automata

• create a tree-stack automaton that is equivalent to the given MCFG:

  cargo run mcfg automaton example.mcfg

• create a push-down automaton that is equivalent to the given CFG:

  cargo run cfg automaton example.cfg

recognition functionality

• parse an MCFG (internally constructing an tree-stack automaton):

  cargo run mcfg parse grammar.gr

• parse a CFG (internally constructing a pushdown automaton):

  cargo run cfg parse grammar.gr

Chomsky-Schützenberger parsing for LCFRS

construct a Chomsky-Schützenberger representation (binary file) of a given LCFRS

• for LCFRS files in rustomata's format

  cargo run -- csparsing extract examples/example.pmcfg > example.cs

• for LCFRS files in discodop's format

  cargo run -- csparsing extract --disco \
                                 --lexer ⟨some_lexer_file_from_discodop⟩ \
                                 ⟨some_gzipped_grammar_from_discodop⟩ \
                                 > example.cs

  • …or without lexer

    cargo run -- csparsing extract --disco \\
                                   ⟨some_gzipped_grammar_from_discodop⟩ \
                                   > example.cs

  • …or with unzipped grammar

    cargo run -- csparsing extract --disco \
                                   --unzipped \
                                   --lexer ⟨some_lexer_file_from_discodop⟩ \
                                   ⟨some_unzipped_grammar_from_discodop⟩ \
                                   > example.cs

• parse a space separated word using a Chomsky-Schützenberger representation

  echo "a a b c c d" | cargo run -- csparsing parse example.cs

approximation

Rustomata contains several approximation strategies, allowing the transformation of automata with storage into other automata with storage. Available are

multiple context-free → context-free

• approximation of an MCFG (via a tree-stack automaton) by a pushdown automaton:

  cargo run approximation tts automaton example.mcfg

• parse a word with the approximation automaton for the given MCFG:

  cargo run approximation tts parse example.mcfg

context-free → context-free

• approximate a CFG (via a pushdown automaton) by a pushdown automaton using an equivalence relation on the non-terminal symbols. An equivalence relation is defined by specifying equivalence classes:

  S [S]
  N [A,B]
  R *
  

  with * matching the remaining non-terminals. To get the approximation automaton:

  cargo run approximation relabel automaton example.cfg example.classes

• to parse with the approximation pushdown automaton:

  cargo run approximation relabel parse example.cfg example.classes

context-free → recognisable

• approximation of a CFG (via a pushdown automaton) by a finite state automaton using a restriction of the underlying pushdown to height k:

  cargo run approximation ptk automaton example.cfg k

• parse with the approximation finite state automaton:

  cargo run approximation ptk parse grammar.gr k

coarse-to-fine parsing

currently being refactored