Boosting bonsai trees for efficient features combination: application to speaker role identification

Abstract

In this article, we tackle the problem of speaker role detection from broadcast news shows. In the literature, many proposed solutions are based on the combination of various features coming from acoustic, lexical and semantic information with a machine learning algorithm. Many previous studies mention the use of boosting over decision stumps to combine efficiently these features. In this work, we propose a modification of this state-of-the-art machine learning algorithm changing the weak learner (decision stumps) by small decision trees, denoted bonsai trees. Experiments show that using bonsai trees as weak learners for the boosting algorithm largely improves both system error rate and learning time.

Can We Predict How Challenging Spoken Language Understanding Corpora Are Across Sources, Languages, and Domains?

“…We use a glass-box classifier called Bonzaiboost 1 [8] based on boosting [14] where a set of weak classifiers made of small decision trees on the features of GF Ambiguity # of semantic labels acceptable for W # of Part-Of-Speech (POS) acceptable for W + POS label # of possible syntactic dependency for W + dependency label distance between W and the sentence syntactic root. utterance length (in words) % of words in S belonging to a concept Coverage # of occurrences of W in train # of occurrences of (W, l) in train is bigrams (W − 1,W ) and (W,W + 1) occurring in train?…”

Automatic quality estimation for speech translation using joint ASR and MT features

“…In a preliminary experiment, we will evaluate these features for quality assessment in ASR only (W CE ASR task). Two different classifiers will be used: a variant of boosting classification algorithm called bonzaiboost [14] (implementing the boosting algorithm Adaboost.MH over deeper trees) and the Conditional Random Fields [12].…”

“…This allowed us to observe that WCE performance decreases as ASR system improves. For reproducible research, most features14 and algorithms used in this paper are available through our toolkit called WCE-LIG. This package is made available on a GitHub repository 15 under the licence GPL V3.…”