... (Charniak,2001). For these reasons, we prefer enriching the syntactic dependencies by information from the left context.However, as mentioned in (Roark, 2001), oneway of conditioning the probabilities ... the richer syntactic tags. Since thetreebank parses allow us to annotate parent informa-tion onto the constituents, as Johnson did in (John-son, 1998), this richer predictive annotation can ... annotation can ex-tend information slightly beyond the left context.Under the equivalence classification inEq.( 2, 3, 4), the conditional information avail-able to the SLM model components is made up...