Non-Contiguous Tree Parsing

Pairing structural descriptions in MT, syntax-semantics interfaces and so on becomes more difficult the more structurally different are the languages involved; there is, implicitly or explicitly, a process of ‘tree parsing’, where a structural description is split into component smaller trees for transfer rules to be applied. Recent work has looked at the construction of transfer rules, using both symbolic and statistical approaches, that require the pairing of groups of several contiguous nodes in structural descriptions. We look at the case where pairings of groups of non-contiguous nodes are necessary, and present an efficient dynamic programming algorithm based on TAG and drawing on compiler theory for a decomposition into appropriate groupings. We then examine the formal properties of this algorithm, and show that it is linear in the number of nodes in the tree and has the same complexity as existing algorithms requiring only groupings of contiguous nodes.