MIT Press

Table 2

Tagger features for , , and .

	LEX	ORTHO	POS
τ_i	τ_i, w_i	τ_i, w_i[0]	τ_i, ϕ_i
τ_{i − 1}, τ_i	τ_i, w_{i − 1}	τ_i, w_i[0..1]	τ_i, ϕ_{i − 1}
τ_{i − 2}, τ_i	τ_i, w_{i + 1}	τ_i, w_i[0..2]	τ_i, ϕ_{i − 1}, ϕ_i
τ_{i − 2}, τ_{i − 1}, τ_i	τ_i, w_{i − 2}	τ_i, w_i[0..3]	τ_i, ϕ_{i + 1}
	τ_i, w_{i + 2}	τ_i, w_i[n]	τ_i, ϕ_i, ϕ_{i + 1}
	τ_i, w_{i − 1}, w_i	τ_i, w_i[n-1..n]	τ_i, ϕ_{i − 1}, ϕ_i, ϕ_{i + 1}
	τ_i, w_i, w_{i + 1}	τ_i, w_i[n-2..n]	τ_i, ϕ_{i − 2}
		τ_i, w_i[n-3..n]	τ_i, ϕ_{i − 2}, ϕ_{i − 1}
		τ_i, w_i ⊆ Digit	τ_i, ϕ_{i − 2}, ϕ_{i − 1}, ϕ_i
		τ_i, w_i ⊆ UpperCase	τ_i, ϕ_{i + 2}
		τ_i, w_i ⊆ Hyphen	τ_i, ϕ_{i + 1}, ϕ_{i + 2}
		τ_i, w_i ⊆ Hyphen	τ_i, ϕ_i, ϕ_{i + 1},ϕ_{i + 2}

	LEX	ORTHO	POS
τ_i	τ_i, w_i	τ_i, w_i[0]	τ_i, ϕ_i
τ_{i − 1}, τ_i	τ_i, w_{i − 1}	τ_i, w_i[0..1]	τ_i, ϕ_{i − 1}
τ_{i − 2}, τ_i	τ_i, w_{i + 1}	τ_i, w_i[0..2]	τ_i, ϕ_{i − 1}, ϕ_i
τ_{i − 2}, τ_{i − 1}, τ_i	τ_i, w_{i − 2}	τ_i, w_i[0..3]	τ_i, ϕ_{i + 1}
	τ_i, w_{i + 2}	τ_i, w_i[n]	τ_i, ϕ_i, ϕ_{i + 1}
	τ_i, w_{i − 1}, w_i	τ_i, w_i[n-1..n]	τ_i, ϕ_{i − 1}, ϕ_i, ϕ_{i + 1}
	τ_i, w_i, w_{i + 1}	τ_i, w_i[n-2..n]	τ_i, ϕ_{i − 2}
		τ_i, w_i[n-3..n]	τ_i, ϕ_{i − 2}, ϕ_{i − 1}
		τ_i, w_i ⊆ Digit	τ_i, ϕ_{i − 2}, ϕ_{i − 1}, ϕ_i
		τ_i, w_i ⊆ UpperCase	τ_i, ϕ_{i + 2}
		τ_i, w_i ⊆ Hyphen	τ_i, ϕ_{i + 1}, ϕ_{i + 2}
		τ_i, w_i ⊆ Hyphen	τ_i, ϕ_i, ϕ_{i + 1},ϕ_{i + 2}

All lexical (LEX), orthographic (ORTHO), and part-of-speech (POS) features are duplicated to also occur with τ_i − 1; e.g., {τ_i − 1, τ_i, w_i} as a LEX feature.