cwb-ccc 0.9.6

CWB wrapper to extract concordances and collocates

Collocation and Concordance Computation

• Introduction
• Installation
• Usage
  • Defining your corpus
  • Extracting concordance lines
  • Dealing with anchored queries
  • Calculating collocates
  • Extracting keywords
  • Argument queries
• Acknowledgements

Introduction

This module is a wrapper around the IMS Open Corpus Workbench (CWB). Main purpose of the module is to extract concordance lines, calculate keywords and collocates, and run queries with several anchor points.

If you want to extract the results of queries with more than two anchor points, the module requires CWB version 3.4.16 or later.

Installation

You can install this module with pip from PyPI:

pip3 install cwb-ccc

You can also clone the repository from github, cd in the respective folder, and use setup.py:

python3 setup.py install

Usage

Corpus Setup

All methods rely on the Corpus class, which establishes the connection to your CWB-indexed corpus:

from ccc import Corpus
corpus = Corpus(
	corpus_name="EXAMPLE_CORPUS",
	registry_path="/path/to/your/cwb/registry/"
)

This will raise a KeyError if the named corpus is not in the specified registry.

If you are using macros and wordlists, you have to store them in a separate folder (with subfolders wordlists and macros). Make sure you specify this folder via lib_path when initializing the corpus.

If you want to compare your query results according to meta data, set the s_meta parameter to the structural attribute that links your data base (e.g. "text_id").

You can use the cqp_bin to point the module to a specific version of cqp (this is also helpful if cqp is not in your PATH).

By default, the cache_path points to "/tmp/ccc-cache". Make sure that "/tmp/" exists and appropriate rights are granted. Otherwise, change the parameter when initializing the corpus (or set it to None).

Concordancing

Before you can display concordances, you have to run a query with the corpus.query() method, which accepts valid CQP queries such as

query = '[lemma="Angela"]? [lemma="Merkel"] [word="\\("] [lemma="CDU"] [word="\\)"]'
corpus.query(query)

The default context window is 20 tokens to the left and 20 tokens to the right of the query match and matchend, respectively. You can change this via the context parameter.

Note that queries may end on a "within" clause (s_query), which will limit the matches to regions defined by this structural attributes. Additionally, you can specify an s_break parameter, which will cut the context. NB: The implementation assumes that s_query regions are confined by s_break regions, and both of them are within the s_meta regions.

Now you are set up to get the query concordance:

concordance = corpus.concordance()

You can access the query frequency breakdown via concordance.breakdown:

type	freq
Angela Merkel ( CDU )	2253
Merkel ( CDU )	29
Angela Merkels ( CDU )	2

All query matches and their respective s_meta identifiers are listed in concordance.meta (if s_meta=None, it will use the CQP identifiers of the s_break parameter as s_id):

match	s_id
48349	A44847086
48856	A44855701
52966	A44847097
53395	A44847526
...	...

You can use concordance.lines() to get concordance lines. This method returns a dictionary with the cpos of the match as keys and the entries one concordance line each. Each concordance line is formatted as a pandas.DataFrame with the cpos of each token as index:

cpos	offset	word	anchor
48344	-5	Eine	None
48345	-4	entsprechende	None
48346	-3	Steuererleichterung	None
48347	-2	hat	None
48348	-1	Kanzlerin	None
48349	0	Angela	None
48350	0	Merkel	None
48351	0	(	None
48352	0	CDU	None
48353	0	)	None
48354	1	bisher	None
48355	2	ausgeschlossen	None
48356	3	.	None

You can decide which and how many concordance lines you want to retrieve by means of the parameters order ("first", "last", or "random") and cut_off. You can also provide a list of matches (from concordance.meta.index) to get a dict of specific concordance lines.

You can specify a list of additional p-attributes besides the primary word layer to show via the p_show parameter of concordance.lines() (these will be added as additional columns).

Anchored Queries

The concordancer detects anchored queries automatically. The following query

concordance.query(
	'@0[lemma="Angela"]? @1[lemma="Merkel"] [word="\\("] @2[lemma="CDU"] [word="\\)"]'
)

thus returns DataFrames with appropriate anchors in the anchor column:

cpos	offset	word	anchor
48344	-5	Eine	None
48345	-4	entsprechende	None
48346	-3	Steuererleichterung	None
48347	-2	hat	None
48348	-1	Kanzlerin	None
48349	0	Angela	0
48350	0	Merkel	1
48351	0	(	None
48352	0	CDU	2
48353	0	)	None
48354	1	bisher	None
48355	2	ausgeschlossen	None
48356	3	.	None

Collocation Analyses

After executing a query, you can use the corpus.collocates() class to extract collocates for a given window size (symmetric windows around the corpus matches):

query = '[lemma="Angela"] [lemma="Merkel"]'
corpus.query(query, s_break='s', context=20)
collocates = corpus.collocates()

collocates() will create a dataframe of the context of the query matches. You can specify a smaller maximum window size via the mws parameter (this might be reasonable for queries with many hits). You will only be able to score collocates up to this parameter. Note that mws must not be larger than the context parameter of your initial query.

By default, collocates are calculated on the "lemma"-layer, assuming that this is a valid p-attribute in the corpus. The corresponding parameter is p_query (which will fall back to "word" if the specified attribute is not annotated in the corpus).

Using the marginal frequencies of the items in the whole corpus as a reference, you can directly annotate the co-occurrence counts in a given window:

collocates.show(window=5)

The result will be a DataFrame with lexical items (p_query layer) as index and frequency signatures and association measures as columns:

item	O11	f2	N	f1	O12	O21	O22	E11	E12	E21	E22	log_likelihood	...
die	1799	25125329	300917702	22832	21033	25123530	275771340	1906.373430	20925.626570	2.512342e+07	2.757714e+08	-2.459194	...
Bundeskanzlerin	1491	8816	300917702	22832	21341	7325	300887545	0.668910	22831.331090	8.815331e+03	3.008861e+08	1822.211827	...
.	1123	13677811	300917702	22832	21709	13676688	287218182	1037.797972	21794.202028	1.367677e+07	2.872181e+08	2.644804	...
,	814	17562059	300917702	22832	22018	17561245	283333625	1332.513602	21499.486398	1.756073e+07	2.833341e+08	-14.204447	...
Kanzlerin	648	17622	300917702	22832	22184	16974	300877896	1.337062	22830.662938	1.762066e+04	3.008772e+08	559.245198	...

For improved performance, all hapax legomena in the context are dropped after calculating the context size. You can change this behaviour via the drop_hapaxes parameter of collocates.show().

By default, the dataframe is annotated with "z_score", "t_score", "dice", "log_likelihood", and "mutual_information" (parameter ams). For notation and further information regarding association measures, see collocations.de. Available association measures depend on their implementation in the association-measures module.

The dataframe is sorted by co-occurrence frequency (O11), and only the first 100 most frequently co-occurring collocates are retrieved. You can change this behaviour via the order and cut_off parameters.

Keyword Anayses

For keyword analyses, you will have to define a subcorpus. The natural way of doing so is by selecting text identifiers (on the s_meta annotations) via spreadsheets or relational databases. If you have collected a set of identifiers, you can create a subcorpus via the corpus.subcorpus_from_ids() method:

corpus.subcorpus_from_ids(ids)
keywords = corpus.keywords()
keywords.show()

Just as with collocates, the result will be a DataFrame with lexical items (p_query layer) as index and frequency signatures and association measures as columns.

Argument Queries

Argument queries are anchored queries with additional information. (1) Each anchor can be modified by an offset (usually used to capture underspecified tokens near an anchor point). (2) Anchors can be mapped to external identifiers for further logical processing. (3) Anchors may be given a clear name:

anchor	offset	idx	clear name
0	0	None	None
1	-1	None	None
2	0	None	None
3	-1	None	None

Furthermore, several anchor points can be combined to form regions, which in turn can be mapped to identifiers and be given a clear name:

start	end	idx	clear name
0	1	"0"	"person X"
2	3	"1"	"person Y"

Example: Given the definition of anchors and regions above as well as suitable wordlists, the following complex query extracts corpus positions where there's some correlation between "person X" (the region from anchor 0 to anchor 1) and "person Y" (anchor 2 to 3):

query = (
	"<np> []* /ap[]* [lemma = $nouns_similarity] "
	"[]*</np> \"between\" @0:[::](<np>[pos_simple=\"D|A\"]* "
	"([pos_simple=\"Z|P\" | lemma = $nouns_person_common | "
	"lemma = $nouns_person_origin | lemma = $nouns_person_support | "
	"lemma = $nouns_person_negative | "
	"lemma = $nouns_person_profession] |/region[ner])+ "
	"[]*</np>)+@1:[::] \"and\" @2:[::](<np>[pos_simple=\"D|A\"]* "
	"([pos_simple=\"Z|P\" | lemma = $nouns_person_common | "
	"lemma = $nouns_person_origin | lemma = $nouns_person_support | "
	"lemma = $nouns_person_negative | "
	"lemma = $nouns_person_profession] | /region[ner])+ "
	"[]*</np>) (/region[np] | <vp>[lemma!=\"be\"]</vp> | "
	"/region[pp] |/be_ap[])* @3:[::]"
)

NB: the set of identifiers defined in the table of anchors and in the table of regions, respectively, should not overlap.

It is customary to store these queries in json objects (see an example in the repository).

You can use the concordancer to process argument queries and display the results:

# read the query file
import json
query_path = "tests/gold/query-example.json"
with open(query_path, "rt") as f:
	query = json.loads(f.read())

# query the corpus and initialize the concordancer
corpus.query(query['query'], context=None, s_break='tweet', match_strategy='longest')
concordance = corpus.concordance()

# show results
concordance.show_argmin(query['anchors'], query['regions'])

The show_argmin method returns the result as a dict with the following keys:

• "nr_matches": the number of query matches in the corpus.
• "holes": a global list of all tokens of the entities specified in the "idx" columns (default: lemma layer).
• "meta": the meta ids of the concordance lines.
• "settings": the query settings.
• "matches": a list of concordance lines. Each concordance line contains:
  • "position": the corpus position of the match
  • "df": the actual concordance line as returned from Concordance().query() (see above) converted to a dict
  • "holes": a mapping from the IDs specified in the anchor and region tables to the tokens or token sequences, respectively (default: lemma layer)
  • "full": a reconstruction of the concordance line as a sequence of tokens (word layer)

Acknowledgements

The module relies on cwb-python, thanks to Yannick Versley and Jorg Asmussen for the implementation. Special thanks to Markus Opolka for the implementation of association-measures and for forcing me to write tests.

This work was supported by the Emerging Fields Initiative (EFI) of Friedrich-Alexander-Universität Erlangen-Nürnberg, project title Exploring the Fukushima Effect.

Further development of the package has been funded by the Deutsche Forschungsgemeinschaft (DFG) within the project Reconstructing Arguments from Noisy Text, grant number 377333057, as part of the Priority Program Robust Argumentation Machines (SPP-1999).