Symmetric categorial grammar, or Lambek-Grishin calculus (LG), arose
out of the generalization of Lambek's syntactic calculus proposed by V.N. Grishin in 1983.
LG and ACG share the same compositional architecture. The differences are:
- ACG source and target logics are intuitionistic systems: judgements are of the form
X ⊢ A, with A a single conclusion formula. The abstract syntax of LG is a
bilinear system with judgements X ⊢ Y allowing multiple conclusions.
In LG, one considers a composition operation ('Merge', fusion) and its dual (fission);
these two communicate via structure-preserving distributivity principles.
- Whereas ACG specifies linear order and phrase structure on a word-by-word
basis in the lexicon, LG encodes these aspects of grammatical structure in its
logical constants: the LG source logic is non-commutative and respects
constituent structure.
- Compositional interpretations for LG are constrained to pass through
a continuation-passing-style translation. The image of the CPS translation
is a linear lambda term, in Curry-Howard correspondence with a MILL
proof, as in ACG.
The bilinear perspective of LG leads to a re-assessment of mild context-sensitivity:
semantic expressivity results from the CPS translation that makes the context an
explicit component of the computation; syntactic expressivity is obtained from
the distributivity laws that channel the flow of information between the
fusion/fission dimensions of grammatical structure.
Key notions to be covered:
- Symmetries: fusion, fission; linear distributivity.
- Bringing in the context: normal derivations, continuation-passing-style translation.
- Graphical calculus: LG proof nets and normal derivations.
- Modeling mild context-sensitivity: LG and the k-MCFG hierarchy.
Tuur and René on Moot's encoding of TAG in LG, and on his
analysis of LG in terms of Hyperedge Replacement Grammar; Julia and Vlasta on the
category theoretic background of categorial grammar.