We herein present a language-model-based evaluator for deletion-based sentence
compression and view this task as a series of deletion-and-evaluation operations using the
evaluator. More specifically, the evaluator is a syntactic neural language model that is first
built by learning the syntactic and structural collocation among words. Subsequently, a
series of trial-and-error deletion operations are conducted on the source sentences via a
reinforcement learning framework to obtain the best target compression. An empirical study …

… (2) Compared with Evaluator-LM, Evaluator-SLM brings further improvement, suggesting
syntax feature may enable model to learn more unseen but reasonable word collocations. (eg
noun is usually followed by verb rather than adjective). … Task: delete words in the original
sentence to form short compression that remains readable and preserves its underlying meaning.
… The total reward is used to reinforce the policy network in search of the best compression. …