Part-of-speech tagging using decision trees

Abstract

{ Imarquez, horacio}@ Is i. upc. esA b s t r a c t . We have appfied inductive learning of statistical decision trees to the Natural Language Processing (NLP) task of morphosyntactic disambiguation (Part Of Speech Tagging). Previous work showed that the acquired language models are independent enough to be easily incorporated, as a statistical core of rules, in any flexible tagger. They are also complete enough to be directly used as sets of POS disambiguation rules. We have implemented a quite simple and fast tagger that has been tested and evaluated on the Wall Street Journal (WSJ) corpus with a remarkable accuracy. In this paper we basically address the problem of tagging when only small training material is available, which is crucial in any process of constructing, from scratch, an annotated corpus. We show that quite high accuracy can be achieved with our system in this situation. In addition we also face the problem of dealing with unknown words under the same conditions of lacking training examples. In this case some comparative results and comments about close related work are reported. I n t r o d u c t i o n a n d S t a t e of t h e A r tPOS Tagging is a very well known NLP problem which consists of assigning to each word of a text the proper morphosyntactic tag in its context of appearance. Figure 1 shows the correct part of speech assignment to the words of a sentence, together with the list of valid labels for each word taken in isolation 1. The base of POS tagging is that being most words ambiguous regarding their POS, they can be almost completely disambiguated taking into account an adequate context.Starting with the pioneer tagger T A G G I T (Greene & Rubin 71), used for an initial tagging of the Brown Corpus (BC), a lot of efforts have been devoted to improve the quality of the tagging process in terms of accuracy and efficiency. Existing taggers can be classified into three main groups according to the kind of knowledge they use: linguistic, statistic and machine-learning family. Of course *

Analysis of continuous neuronal activity evoked by natural speech with computational corpus linguistics methods

“…Book and audio book consist of a total number of 40460 tokens (number of words) and 6117 types (number of unique words). The distribution of single word classes and bi-gram word class combinations occurring in the (audio) book were analysed and compared to a number of German reference corpora [59], and in addition, other German novels, by applying part-of-speech (POS) tagging [60][61][62] as implemented in the python library spaCy [63]. The similarities, or dissimilarities respectively, of all distributions are visualized using multi-dimensional scaling (MDS) [64][65][66][67].…”

Syntactic Wordclass Tagging

“…In principle, however, it would be possible to produce a complete tagger on the basis of a learned statistical decision tree. Recently, this approach has indeed been explored (Marquez & Rodriguez, 1998). (Rumelhart et al 1986) are the most popular neural network architecture.…”

Classifiers combination to arabic morphosyntactic disambiguation

“…Márquetz et al [38] develop decision tree tagger for English POS tagging. They use non-incremental supervised learning from examples of TDIDT (Top Down Induction of Decision Tree) to construct the decision tree.…”