05.03.2013 Views

PhD thesis - School of Informatics - University of Edinburgh

PhD thesis - School of Informatics - University of Edinburgh

PhD thesis - School of Informatics - University of Edinburgh

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Appendix A<br />

Evaluation Metrics and Notation<br />

This appendix explains the evaluation metrics and statistical significance tests used for<br />

the experiments presented in Chapters 3, 4 and 5 and explains how they are calculated.<br />

It also specifies various notations in order to avoid confusion.<br />

A.1 System Evaluation Metrics<br />

In the broad sense, English inclusion detection can be regarded as an information ex-<br />

traction task, where the aim is to identify all English inclusions occurring in text that<br />

is written primarily in a different language. The English inclusion classifier’s per-<br />

formance is evaluated intrinsically on seen and unseen evaluation data against gold<br />

standard annotation. The evaluation measures used for this intrinsic evaluation are ac-<br />

curacy and F-score which are calculated using the conlleval script written by Erik<br />

Tjong Kim Sang. 1 The identification <strong>of</strong> English inclusions is therefore evaluated in a<br />

similar way to named entity recognition (NER), but for single tokens. A useful way<br />

to illustrate how accuracy and F-score are computed is via a contingency table <strong>of</strong> the<br />

gold standard annotation and the system output (see Table A.1). The positive and<br />

negative annotations <strong>of</strong> the gold standard are compared against those produced by the<br />

system. The positive and negative labels which are correctly predicted by the system<br />

with respect to the gold standard are called true positives (TP) and true negatives<br />

(TN), respectively. A wrongly predicted positive label is called false positive (FP) and<br />

1 This script is freely available at: http://www.cnts.ua.ac.be/conll2000/chunking/<br />

conlleval.txt<br />

177

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!