Parsing Ill-Formed Inputs with Constraint Graphs

Logic Grammars for Diagnosis and Repair

“…In this paper we shall argue that we can get similar and maybe even better results than those obtained in the recent declarative work surveyed, through CHR grammars' [13] inherent treatment of ambiguity and abduction with considerably less work from both the implementors and the users: a) as in [9] or [8], we can build non-connected structures showing all partial successful analyzes (but without having to replace trees by graphs as in the former, or needing extra machinery as in the latter). Moreover, we can repair illformed input as well, whereas [9] merely detects it, even though it is capable of parsing sentences containing associated elements such as hesitations, repetitions, etc. (these are largely ignored after being detected).…”

“…Blache, for instance, uses constraint graphs, built from constraints over categories such as linear precedence (e.g. "det must precede n"), exclusion (which restricts co-occurrence between sets of categories), uniqueness (for categories which cannot be repeated in a phrase) [9]. Kakas et al propose a technique for database repair which could perhaps be adapted to syntactic error repair as well, and which involves interleaving abduction and the resolution steps of a CLP framework [21].…”

Logic grammars for diagnosis and repair

“…In this paper we shall argue that we can get similar and maybe even better results than those obtained in the recent declarative work surveyed, through CHR grammars' [13] inherent treatment of ambiguity and abduction with considerably less work from both the implementors and the users: a) as in [9] or [8], we can build non-connected structures showing all partial successful analyzes (but without having to replace trees by graphs as in the former, or needing extra machinery as in the latter). Moreover, we can repair illformed input as well, whereas [9] merely detects it, even though it is capable of parsing sentences containing associated elements such as hesitations, repetitions, etc.…”

“…Blache, for instance, uses constraint graphs, built from constraints over categories such as linear precedence (e.g. "det must precede n"), exclusion (which restricts co-occurrence between sets of categories), uniqueness (for categories which cannot be repeated in a phrase) [9]. Kakas et ai.…”

“…Moreover, we can repair illformed input as well, whereas [9] merely detects it, even though it is capable of parsing sentences containing associated elements such as hesitations, repetitions, etc. (these are largely ignored after being detected).…”