RegCCRF

Software accompanying the paper Constraining Linear-chain CRFs to Regular Languages

RegCCRF

Typ
Tool
Autor

Sean Papay, Roman Klinger, Sebastian Padó

Beschreibung

In structured prediction, a major challenge for models is to represent the interdependencies within their output structures.  For the common case where outputs are structured as a sequence, linear-chain conditional random fields (CRFs) are a widely used model class which can learn local dependencies in output sequences.  However, the CRF's Markov assumption makes it impossible for these models to capture nonlocal dependencies, and standard CRFs are unable to respect nonlocal
constraints of the data (such as global arity constraints on output labels).  We present a generalization of CRFs that can enforce a broad class of constraints, including nonlocal ones, by specifying the space of possible output structures as a regular language L. The resulting regular-constrained CRF (RegCCRF) has the same formal properties as a standard CRF, but assigns zero probability to all label sequences not in L. Notably, RegCCRFs can incorporate their constraints during training, while related models only enforce constraints during decoding.

Referenz

Sean Papay, Roman Klinger, Sebastian Padó: "Constraining Linear-chain CRFs to Regular Languages".

Download
Dieses Bild zeigt Roman Klinger

Roman Klinger

PD Dr.

Akademischer Oberrat/Privatdozent

Dieses Bild zeigt Sebastian Padó

Sebastian Padó

Prof. Dr.

Lehrstuhlinhaber Theoretische Computerlinguistik, Geschäftführender Direktor des IMS

Zum Seitenanfang