This preview shows pages 1–3. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Search & Decoding Project CMU & SRI Language model Toolkits Introduction The main objective of this project is to study the performance of the CMU and SRI language model toolkits by creating various statistical language models and calculating the perplexity against test corpora and finally comparing the outcomes. While creating these language models we will look at different discounting strategies and study the effect of smoothing on the language models and which strategy is the most effective. Corpus In order to build the statistical language models we use the Corpora provided by National Language Toolkit. ‘Nltk-data-0.o.zip’ Install Cygwin The two toolkits require a set of Unix software tools as part of creating the language models. Cygwin is a free software that simulates a Unix environment on a Windows platform. Cygwin can be downloaded from: The Cygwin Setup instructions are fairly straightforward, After download Cygwin installation file “setup.exe” from the website, we can run “setup.exe” and click “Install from Internet”. Select a downloadable site. Then, we chose install all Cygwin packages (or chose install by default). We should make sure a few packages are installed correctly for the toolkits to work. Here are the most important software packages that have to be installed:-gcc versoin 3.4.3 or higher-GNU make-TCL toolkit , version 7.3 or higher-Tcsh-gzip : to read/write compressed file-GNU awk ( gawk ): to interpret many of the utility script Once Cygwin is installed we can begin installation of the SRILM and CMU toolkits. After downloading the appropriate file from SRILM and CMU websites, we will unpack or unzip the files as follows: cd /cygdrive/c/cygwin/srilm tar zxvf srilm.tgz Similarly for the CMU toolkit. Then we will run and install the toolkits: Make World The steps for creating language models and running perplexity test are slightly different for the two packages and will use differing commands. Building language model with SRILM We first have to create the and n-gram count and an estimate of the language model- Command for generating the 3-gram count file: ./ ngram-count –vocab abc/en.txt -text abc/rural.txt -order 3 -write abc/rural.count -unk Discounting Strategies We can then create the language models with the four discounting strategies. Discounting is the process of replacing the original counts with modified counts so as to redistribute the probability mass from the more commonly observed events to the less frequent and unseen events. If the actual number of occurrences of an event E (such as a bigram or trigram occurrence) is c ( E ), then the modified count is d ( c ( E )) c ( E ), where d ( c ( E )) is known as the discount ratio....
View Full Document
This note was uploaded on 02/11/2012 for the course ECE 5527 taught by Professor Staff during the Fall '11 term at FIT.
- Fall '11