This page describes various approaches to paraphrasing using the DELPH-IN tools. This page was initiated by FrancisBond and DarrenAppling; please feel free to make additions or corrections as you see fit.

• Strict: parse and then generate

• Some variation: parse and then rephrase (see RmrsParaphrasing)

• More variation: create a monolingual Machine Translation system (see MachineTranslationTop)

In this case, the monolingual machine translation should use the external MRS exposed by the Semantic Interface (SEM-I), not the internal MRS, for the following reasons:

1. the SEM-I is the interface to outside resources, such as WordNet, which may also be used in generation

2. the internal MRS is subject to restructuring as the quest for ever better linguistic analysis proceeds

3. the mtr.tdl used in MT refers to the external MRS

The SEM-I should, therefore, expose enough information for the paraphrase engine, including some hierarchies among types, this must be done manually at present.

Bond, Francis, Eric Nichols, Darren Scott Appling, Michael Paul (2008)

• Improving Statistical Machine Translation by Paraphrasing the Training Data, In IWSLT-2008.