This exercise covers the material from Day 4, for working with dictionaries using quanteda. We will also be using the quanteda package, which contains some additional corpora not in the base quanteda package.
Always remember to install a fresh copy of both packages, and from the dev branch for quanteda:
if (!require(devtools)) install.packages("devtools", dependencies=TRUE)
devtools::install_github("kbenoit/quanteda", dependencies=TRUE, ref="dev")
devtools::install_github("kbenoit/quantedaData")Getting used to dictionaries
Creating a simple dictionary.
Dictionaries are named lists, consisting of a “key” and a set of entries defining the equivalence class for the given key. To create a simple dictionary of parts of speech, for instance we could define a dictionary consisting of articles and conjunctions, using:
posDict <- list(articles = c("the", "a", "and"),
conjunctions = c("and", "but", "or", "nor", "for", "yet", "so"))To let this define a set of features, we can use this dictionary when we create a dfm, for instance:
require(quanteda)## Loading required package: quanteda
## Loading required package: data.tableposDfm <- dfm(inaugCorpus, dictionary=posDict)## Creating dfm from a corpus: ... done.posDfm[1:10,]## features
## docs articles conjunctions Non_Dictionary
## 1789-Washington 178 73 1179
## 1793-Washington 15 4 116
## 1797-Adams 344 192 1782
## 1801-Jefferson 232 109 1385
## 1805-Jefferson 256 126 1784
## 1809-Madison 166 63 946
## 1813-Madison 169 63 978
## 1817-Monroe 458 174 2738
## 1821-Monroe 577 195 3687
## 1825-Adams 448 150 2317Weight the posDfm by term frequency using tf(), and plot the values of articles and conjunctions (actually, here just the coordinating conjunctions) across the speeches. (Hint: you can use docvars(inaugCorpus, "Year")) for the x-axis.)
Is the distribution of normalized articles and conjunctions relatively constant across years, as you would expect?
Hierarchical dictionaries.
Dictionaries may also be hierarchical, where a top-level key can consist of subordinate keys, each a list of its own. For instance, list(articles = list(definite="the", indefinite=c("a", "and")) defines a valid list for articles. Make a dictionary of articles and conjunctions where you define two levels, one for definite and indefinite articles, and one for coordinating and subordinating conjunctions. (A sufficient list for your purposes of subordinating conjuctions is “although”, “because”, “since”, “unless”.)
Output the results and examine them. What happened to the hierarchies, to make them into “features”?
Getting used to thesauruses
A “thesaurus” is a list of feature equivalencies specified in the same list format as a dictionary, but which—unlike a dictionary—returns all the features not specified as entries in the thesaurus.
If we wanted to count pronouns as equivalent, for instance, we could use the thesaurus argument to dfm in order to group all listed prounouns into a single feature labelled “PRONOUN”.
mytexts <- c("We are not schizophrenic, but I am.", "I bought myself a new car.")
myThes <- list(pronouns = list(firstp=c("I", "me", "my", "mine", "myself", "we", "us", "our", "ours")))
myDfm <- dfm(mytexts, thesaurus=myThes)## Creating dfm from character vector ... done.myDfm[1:2,]## features
## docs PRONOUNS.FIRSTP are not schizophrenic but am bought a new car
## text1 1 1 1 0 1 0 0 1 0 2
## text2 2 0 0 1 0 0 1 0 0 2Notice how the thesaurus key has been made into uppercase — this is to identify it as a key, as opposed to a word feature from the original text.
Try running the articles and conjunctions dictionary from the previous exercise on as a thesaurus, and compare the results.
Using regular expressions
Regular expressions are very important concepts in text processing, as they offer tools for searching and matching text strings based on symbolic representations. For the dictionary and thesaurus features, we can define equivalency classes in terms of regular expressions. There is an excellent tutorial on regular expressions at http://www.regular-expressions.info.
This provides an easy way to recover syntactic variations on specific words, without relying on a stemmer. For instance, we could construct a tax dictionary as follows:
econDict <- list(tax=c("tax", "charge"), cuts=c("cut", "^aust", "budget"))
econTexts <- c("The new budget raises income taxes and introduces a cuteness tax.",
"The new budgetary era of austerity is upon us, taxing us with surcharges.",
"We live in a new caustic era of taxes and charges.")
econDfm <- dfm(econTexts, dictionary=econDict)## Creating dfm from character vector ... done.econDfm[1:3,]## features
## docs tax cuts Non_Dictionary
## text1 1 1 9
## text2 0 0 13
## text3 0 0 11Observe the results. Why are some words picked up, and others not? Contrast that to the results if dfm() is rerun with the dictionary_regex=TRUE option. Is “caustic” in text3 not picked up under the “cuts” key, and if not, why not? How could the regular expression be modified in order to pick it up?
Why does “tax” as a regular expression pick up all of the tax words?
How could we modify the terms in the cuts key to not count cuteness?
Replicating a published dictionary analysis
Populism dictionary.
Here we will create and implement the populism dictionary from Rooduijn, Matthijs, and Teun Pauwels. 2011. “Measuring Populism: Comparing Two Methods of Content Analysis.” West European Politics 34(6): 1272–83. Appendix B of that paper provides the term entries for a dictionary key for the concept populism. Implement this as a dictionary, and apply it to the same UK manifestos as in the article.
Hint: You can get a corpus of the UK manifestos from their article using the following:
require(quantedaData)## Loading required package: quantedaDatadata(ukManifestos)
ukPopCorpus <- subset(ukManifestos, (Year %in% c(1992, 2001, 2005) &
(Party %in% c("Lab", "LD", "Con", "BNP", "UKIP"))))
summary(ukPopCorpus)## Corpus consisting of 11 documents.
##
## Text Types Tokens Sentences Country Type Year Language
## UK_natl_1992_en_Con 3829 29560 1605 UK natl 1992 en
## UK_natl_1992_en_Lab 2295 11355 623 UK natl 1992 en
## UK_natl_1992_en_LD 2979 17381 939 UK natl 1992 en
## UK_natl_2001_en_Con 2473 13196 721 UK natl 2001 en
## UK_natl_2001_en_Lab 3523 28711 1602 UK natl 2001 en
## UK_natl_2001_en_LD 3263 21177 1232 UK natl 2001 en
## UK_natl_2005_en_BNP 4575 25214 1058 UK natl 2005 en
## UK_natl_2005_en_Con 1842 7687 420 UK natl 2005 en
## UK_natl_2005_en_Lab 3512 23806 1557 UK natl 2005 en
## UK_natl_2005_en_LD 2832 16081 840 UK natl 2005 en
## UK_natl_2005_en_UKIP 2201 8882 425 UK natl 2005 en
## Party
## Con
## Lab
## LD
## Con
## Lab
## LD
## BNP
## Con
## Lab
## LD
## UKIP
##
## Source: /home/paul/Dropbox/code/quantedaData/* on x86_64 by paul.
## Created: Tue Sep 16 16:17:33 2014.
## Notes: .Create a dfm of the populism dictionary on the UK manifestos. Use this dfm to reproduce the x-axis for the UK-based parties from Figure 1 in the article. Suggestion: Use dotchart(). You will need to normalize the values first by term frequency within document. Hint: Use tf() on the dfm.
You can explore some of these terms within the corpus to see whether you think they are appropriate measures of populism. How can you search the corpus for the regular expression politici* as a “keyword in context”?
Laver and Garry (2000) ideology dictionary.
Here, we will apply the dictionary of Laver, Michael, and John Garry. 2000. “Estimating Policy Positions From Political Texts.” American Journal of Political Science 44(3): 619–34. Using the pre-built Laver and Garry (2000) dictionary file, which is distributed by Provalis Research for use with its Wordstat software package from Provalis, we will apply this to the same manifestos from the UK manifesto set.
To do this, you will need to:
download and save the Wordstat-formatted dictionary file LaverGarry.cat;
load this into a dictionary list using readWStatDict();
build a dfm for the corpus subset for the Labour, Liberal Democrat, and Conservative Party manifestos from 1992 and 1997; and
try to replicate their measures from the “Computer” column of Table 2, for Economic Policy. (Not as easy as you thought—any ideas as to why?)
Fun with the Regressive Imagery Dictionary and the LIWC.
Try analyzing the inaugural speeches from 1980 onward using the Linguistic Inquiry and Word Count. The LIWC dictionary is available from the file LIWC2001_English.dic available from the course Readings folder (see the Dropbox link distributed by email). To load it, you can use the quanteda function readLIWCdict() which reads LIWC-formatted dictionaries. Note: In the current implementation, this takes up to a minute to create the dfm, a while because of the size and complexity of the LIWC dictionary. Compare the speeches based on the (normalized) Anger measure.
Try the same thing but using the Regressive Imagery Dictionary, from Martindale, C. (1975) Romantic progression: The psychology of literary history. Washington, D.C.: Hemisphere. You can download the dictionary from http://www.provalisresearch.com/Download/RID.ZIP, formatted for WordStat. Compare the Presidents based on the level of “Icarian Imagery.” Which president is the most Icarian?
Mega-extra credit
(and possibly leading to me hiring you as a post-doc) Examine the WordStat sentiment dictionary which has a series of relative rules here: http://provalisresearch.com/Download/WSD.zip. How could these rules be implemented for dictionaries and dfm creation in quanteda?