tutorial_cqp [CompoNet Wiki]

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== Looking at KWIC lines with CQP ======

by Marco Baroni

===== The First Session =====

Invoke CQP (press enter/return key after this and all other commands):

<code>cqp -e</code>

or:

<code>cqp -eC</code>

Exit from CQP:

<code>exit</code>

(If you did that, please enter again!)

<code>show corpora;</code>

While in CQP, keep in mind that some things work like on the Unix terminal -- in particular, you can recall previous commands with the upwards-pointing arrow, and you navigate the kwic results with more/less-like syntax (space to move to next page, q to quit, etc.)

Select corpus (remember the semi-colon at the end of each command), e.g.:

<code>BNCV4;
REPUBBLICA;
etc.
</code>

(Notice that tab completion works for corpus names.)

Let's stick to the BNC, for now.

A quick way to know how many tokens there are in a corpus:

<code>info;</code>

Simple kwic:

<code>"food";
"food" %c;
"good" "food";</code>

If you have problems seeing accented characters (as in vowels with umlaut in German or with accents in Italian and Spanish), try:

<code>set Pager more;</code>

You can move through the kwic results like with a standard Unix pager: **space** to see next page, **b** to go back one page, **q** to exit kwic display.

Whenever **q** does not work, use **ctrl+c** to interrupt any command.

To see the frequency of occurrence of your last query:

<code>size Last;</code>

If you have too many results, it is a good idea to take a look at a random sample... First, "save" query into a variable:

<code>A = "often";</code>

Then, "reduce" A to the desired number of randomly selected contexts, e.g.:

<code>reduce A to 20;</code>

Finally, take a look at these contexts:

<code>cat A;</code>

Change context size:

<code>set Context 60;
set Context 5 words;
set Context s;
set Context 3 s;
set Context default;</code>

Other visualization options:

<code>show +pos;
show +lemma;
show -pos -lemma;
show -cpos;
set PrintStructures text_domain;
set PrintStructures "";</code>

===== Exploiting morpho-syntactic annotation =====

Doing queries using morphosyntactic annotation (if you've been experimenting with show and set, now it's a good moment to go back to a normal-looking kwic-display):

<code>[word = "obsessive"] [pos = "NN.*"]; 
[word = "obsessive" %c] [pos = "NN.*"];

[word = "cause"];
[lemma = "cause"];
[lemma = "cause" & pos = "VV.*"];</code>

<file>Practice time:
- look for candidate N+N compounds in Italian with "donna" as head (at the lemma level)
- select a random sample of 100 hits
</file>

No need to try the following now, but here is how you can extract a frequency list for a collocate extracted from a "flexible" context (from the BNC, in this specific case):

<code>[lemma = "cause" & pos = "VV.*"][pos="AT0"]?[pos="AJ.*"]*[pos="NN.*"];
count by lemma on matchend;</code>

The former is something you **cannot** do with cwb-scan-corpus.

===== Regular Expressions =====

Nouns ending in izzazione (in la Repubblica):

<code>A = [lemma = ".*izzazione" & pos = "NOUN"];</code>

===== Structural attributes =====

The lemma "opportunist" used by women and men in the BNC:

<code>[lemma="opportunist"] :: match.text_author_sex="Female";
[lemma="opportunist"] :: match.text_author_sex="Male";</code>

===== Saving results =====

Save the results to an output file:

<code>cat Last > "myconc.txt";</code>

or, if you saved results in a variable:

<code>cat A > "myconc.txt";</code>

<file>Practice:
- save 100 random concordance lines for the donna+NOUN pattern in an external text file, with extended context (e.g., a 3 sentence window)</file>

===== Useful links =====

==== Stefan Evert's CQP tutorial ====

[[http://www.ims.uni-stuttgart.de/projekte/CorpusWorkbench/CQPTutorial/html]] \\
[[http://www.ims.uni-stuttgart.de/projekte/CorpusWorkbench/CQPTutorial/cqp-tutorial.pdf]]

==== Some Web-based interfaces ====

Serge Sharoff's Internet Corpora: [[http://corpus.leeds.ac.uk/internet.html]]

CucWeb: [[http://ramsesii.upf.es/cgi-bin/cucweb/search-form.pl?lang=en_US]]

SSLMITDev: [[http://sslmitdev-online.sslmit.unibo.it/corpora/corpora.php]]

The Word Sketch Engine uses a syntax that is almost identical to the
one of CQP!