Valence extraction using EM selection and co-occurrence matrices

Abstract

This paper discusses two new procedures for extracting verb valences from raw texts, with an application to the Polish language. The first novel technique, the EM selection algorithm, performs unsupervised disambiguation of valence frame forests, obtained by applying a non-probabilistic deep grammar parser and some post-processing to the text. The second new idea concerns filtering of incorrect frames detected in the parsed text and is motivated by an observation that verbs which take similar arguments tend to have similar frames. This phenomenon is described in terms of newly introduced co-occurrence matrices. Using co-occurrence matrices, we split filtering into two steps. The list of valid arguments is first determined for each verb, whereas the pattern according to which the arguments are combined into frames is computed in the following stage. Our best extracted dictionary reaches an $F$-score of 45%, compared to an $F$-score of 39% for the standard frame-based BHT filtering.Comment: 24 pages, 3 table

Walenty: gruntowny składniowo-semantyczny słownik walencyjny języka polskiego

“…Prace Zespołu Inżynierii Lingwistycznej IPI PAN sytuują się na granicy językoznawstwa teoretycznego, lingwistyki korpusowej, leksykografii elektronicznej i -oczywiście -przetwarzania języka naturalnego. W drugiej połowie pierwszej dekady XXI wieku zespół podejmował prace nad automatycznym wydobywaniem informacji walencyjnych z jedynie morfoskładniowo znakowanych korpusów (Przepiórkowski, Fast 2005;Dębowski 2009;Przepiórkowski 2009;Hajnicz 2011). Prace doprowadziły do powstania nowych algorytmów i narzędzi, lecz wydobywane informacje walencyjne obarczone były zbyt dużym błędem, by stać się podstawą gruntownego słownika walencyjnego wysokiej jakości.…”

An Unsupervised Model for Instance Level Subcategorization Acquisition

“…A number of recent works have applied modern machine learning techniques to SCF induction, including point-wise co-occurrence of arguments (Debowski, 2009), a Bayesian network model (Lippincott et al, 2012), multi-way tensor factorization (Van de Cruys et al, 2012) and Determinantal Point Processes (DPPs) -based clustering (Reichart and Korhonen, 2013). However, all of these systems induce type-level SCF lexicons and, except from the system of (Lippincott et al, 2012) that is not capable of learning traditional SCFs, they all rely on supervised parsers.…”

“…Most existing systems rely on handwritten rules (Briscoe and Carroll, 1997;Korhonen, 2002;Preiss et al, 2007) or simple cooccurrence statistics (O'Donovan et al, 2005;Chesley and Salmon-Alt, 2006;Ienco et al, 2008;Lenci et al, 2008;Altamirano and Alonso i Alemany, 2010;Kawahara and Kurohashi, 2010) applied to the grammatical dependency output of supervised statistical parsers. Even the handful of recent systems that use modern machine learning techniques (Debowski, 2009;Lippincott et al, 2012;Van de Cruys et al, 2012;Reichart and Korhonen, 2013) use supervised parsers to pre-process the data 2 .…”

A Hebrew verb–complement dictionary

“…Rather, we treated each potential complement independently of other complements, and only attempted to determine the strength of its own attachment to the verb. This approach is similar to the one used by Dȩbowski (2009), who also collected single possible complements from a (Polish) corpus; however, Dȩbowski (2009) also defined a second filtering step in which full frames are constructed.…”