| NEWS {koRpus} | R Documentation |
tokenize()/treetag(): as indicated by unit tests in tm.plugin.koRpus, the
nchar(type="width") issue wasn't fully fixed yet
read.corp.LCC()/read.corp.celex()/readTagged(): changed how encoding is
applied to files to ensure no re-encoding takes place on windows, which
might break UTF-8 encoded characters and result in failure to correctly read
files
text descriptives: R-devel changed how nchar(type="width") counts newline
characters, therefore the counting of characters with normalized space
had to be adjusted
lex.div()/MTLD(): calculations were slightly off (~0.5%) due to an
incorrect stage of applying means to the forward/backward calculations; MTLD-MA
remains unaffected (thanks to akira murakami for reporting the issue)
treetag()/tokenize(): added a check to doc_id which is expected to be a
character string; especially if it was manually set to 0 issues were
reported
fixed some URLs (https if available)
class kRp.TTR: dropped the mean value from the "factors" list of MTLD
results
readability(): flat=TRUE now stores results in a list named by doc_id
like lex.div() already did
summary(): features "lex_div" and "readability" are now supported for
kRp.txt objects, a new "flat" argument was added
readability()/lex.div(): dropped "Note:" from validity warnings as it is
already a warning
updated unit test standards
readability(): new formula "Gutierrez" for spanish texts, also added to
the shiny web app
readability()/fucks(): the oldest bug so far, present since the first
version of the package: Fucks' formula doesn't determine word length by
characters but syllables; references were updated. the index has been on the
list of "needs validation" and still remains there. the erroneous formula
likely came from the documentation of TextQuest, as the initial scope of
koRpus, when it wasn't even a package yet, was to validate the calculations
of various readability tools (thanks to berenike herrmann for the hint)
cTest(): don't freak out if there's text left after the last sentence
ending punctuation
textTransform(): the argument "paste=TRUE" was broken
readability.num(): solved issue of missing "txt.file" object and
undefined language; "lang" can now also be set in the "text.features" if needed
kRp_TTR(): validity check was missing "sd" in names of the MSTTR slot
the package now installs a sample text that is used in many examples
many examples now use a sample text and can therefore omit the \dontrun{} clause they were previously enclosed in
class definitions now use the initialize method instead of prototype()
kRp.text.analysis(): deprecated since 0.13-1, removed the code
treetag(): allow for lexicon files to be optional and not return an error
if none is found (which was the case with the newly added file name
checks)
treetag(): use "-lex" argument for lexicon files if no lookup command is
given
treetag(): always add lookup command from manual options even if a preset
is used
read.corp.custom(): calculation failed if caseSens=FALSE
tokenize()/treetag(): force UTF-8 encoding on read texts to prevent
windows from misunderstanding characters
treetag() et al.: drastically increased the speed of calculating
descriptive statistics (can be 100x faster for very large texts)
updated the language package templates
treetag(): the "utf8" check for lexicon files led to path errors if the
lexicon was NULL
unit tests: jumbledWords() randomly created false positives, fixed by
setting a seed
#freeRealityWinner
docTermMatrix(): numbers were calculated correctly, but possibly added to
the wrong columns, leading to a completely wrong document term matrix
treetag(): a dumb misorderig of calls suppressed the "utf8" check for
abbreviation files introduced with 0.11-5
treetag(): also added a "utf8" check for lexicon files and ".txt" file
extensions (which might be missing in newer versions of TreeTagger)
correct.tag(): stopped method from adding tag descriptions to objects
that didn't have them yet
kRp_readability()/kRp_corp_freq(): properly initialize the slots
readability() wrapper functions: fixed a bunch of readability.num() calls
including an unused hyphen argument
readability(): HTML documentation had a wrong formula for LIX (LaTeX was
correct)
textTransform(): now recounting letters when scheme is "normalize" as it
might have altered word lengths; the calculations of some data in the desc
slot (all.chars, lines, normalized.space) as now also done relative to
the old values, because they can't be correctly recalculated from a mere
vector of tokens
docTermMatrix(): optimzed calculation speed drastically
read.corp.custom(): re-wrote most of the code, now based on
docTermMatrix() and thereby up to 50 times faster; also removed the now unused quiet
argument, as well as methods using the directory path or lists of tagged
texts, because using methods of the tm.plugin.koRpus package instead is much
more efficient now
show(): simplified the code for kRp.text class objects and unified the
horizontal positioning of resulting values
show(): generalized the handling of factor columns to be able to deal
with unexpected columns
tokenize(), treetag(): always generate a doc_id if none was given; also
improved the examples
readability(): added some ASCII versions of the formulae to the
documentation
readability(): the code of the internal workhorse kRp.rdb.formulae() was
cleaned up, now using the new helper functions validate_parameters(),
check_parameters() and rdb_parameters(), saving ~350 lines of code
updated unit test
kRp.text: new replacement class for kRp.tagged, kRp.txt.freq,
kRp.txt.trans; the TT.res slot was renamed into "tokens", additional columns in the
data frame are now ok, new slots "features" and "feat_list" to host
analysis results like readability or lexical diversity, and the "desc" slot now
always contains elements named by doc_id
docTermMatrix(): new method to calculate document term matrices from TIF
compliant token data frames and koRpus objects
doc_id(), hasFeatures(), hasFeatures()<-, features(), features()<-,
corpusReadability(), corpusReadability()<-, corpusHyphen(), corpusHyphen()<-,
corpusLexDiv(), corpusLexDiv()<-, corpusFreq(), corpusFreq()<-,
corpusCorpFreq(), corpusCorpFreq()<-, corpusStopwords(), corpusStopwords()<-: new
getter/setter methods for kRp.text objects
dependencies: the Matrix package was added to imports for docTermMatrix()
validate_df(): new internal method to check data frames for expected
columns
readability(): new argument "keep.input" to define whether hyphen objects
should be preserved in the output or dropped
hyphen(), lex.div(): new argument "as.feature" to store results in the
new "feat_list" slot of the input object rather than returning it directly
fixObject(): new methods to convert old objects of deprecated classes
kRp.tagged, kRp.txt.freq, kRp.txt.trans, and kRp.analysis
split_by_doc_id(): new method transforms a kRp.text object with multiple
doc_ids into a list of single-document kRp.text objects
[[/[[<-: gained new argument "doc_id" to limit the scope to particular
documents
describe()/describe()<-: now support filtering by doc_id
kRp.tagged, kRp.txt.freq, kRp.txt.trans, kRp.analysis: these classes were
special cases of kRp.text, and since all their information can now be
part of kRp.text objects, they are no longer used; they are actually still
present, but considered deprecated and should be converted using fixObject()
readability(), freq.analysis(): removed the methods that could be called
on files directly instead of objects of class kRp.text. this simplifies
the code and it's probably not too much to ask users to call tokenize() or
treetag() directly instead of doing this internally with less control
freq.analysis(): removed the "tfidf" argument; as it turned out, its
value was never effectively used, the tf-idf was always calculated, and it
seemed like a reasonable default anyway
kRp.text.analysis(): now deprecated, just use lex.div() and
freq.analysis() to the same effect
query(): method was broken for tagged objects
textTransform(): method was broken
class kRp.txt.trans: renamed column "token.old" into "token.orig", which
is what was actually used by textTransform(); also added a validity test
for those column names to prevent confusion
readTagged(): adjusted default encoding
query(): new method for objects of class data.frame, which is now used if
query() is being called on koRpus class objects
query(): now also supports all numerical queries for tagged texts that
were previously only available for frequency objects
filterByClass(): a new method for tagged text objects, replacing the
kRp.filter.wclass() function, which is now deprecated
pasteText(): like filterByClass(), but replacing kRp.text.paste()
readTagged(): like filterByClass(), but replacing read.tagged()
readTagged(): new argument mtx_cols for new tagger="manual" setting,
allowing to import data POS tagged with third party tools
textTransform(): new scheme "normalize" to replace tokens by given query
rules with a defined value or the result of a provided function
diffText()/diffText()<-: new getter/setter methods for the "diff" slot of
transformed text objects
originalText(): new method to revert text transformations and get the
original text
kRp.POS.tags(): now includes universal POS tags by default
new unit tests for many methods, including query(), textTransform(),
readTagged(), filterByClass(), pasteText(), diffText(), originalText(),
jumbleWords(), and clozeDelete()
tokenize(): now a S4 method for objects of class character and connections
treetag(): now a S4 method for objects of class character and connections
class kRp.txt.trans: "diff" slot now also lists the transformations done
to the tokens in a new list element called "transfmt", the changed tokens
in a data frame called "transfmt.equal" and normalization details in a
list called "transfmt.normalize"
language support: if you try using a preset but the langauge package wasn't loaded or even installed, a more elaborate error message is returned with hopefully useful hints to what to try next
jumbleWords(): now a S4 method, no longer a function; the resulting
object is now also of class kRp.txt.trans if the input was a tagged text
object, preserving the original tokens
clozeDelete(): now returns an object of class kRp.txt.trans, dropping the
additional data frame in "desc"; this ist much more consistent with other
text transformations in the packahe
cTest(): like clozeDelete() now returns an object of class kRp.txt.trans,
dropping the additional data frame in "desc"
moved class union definition kRp.taggedText to its own file and updated the import calls on a number of files accordingly
textTransform(): moved the whole code segment that combines the
transformed text into the returned object to a separate internal function so it can
be re-used by other text transforming methods
cTest(): changed method signature from kRp.tagged to class union
kRp.taggedText
summary(): changed method signature from kRp.tagged to class union
kRp.taggedText
plot(): changed method signature from kRp.tagged to class union
kRp.taggedText
lex.div(): removed the validation warning for MATTR, implementation has
been validated by kevin cunningham and katarina haley
restructured source code files
set.kRp.env()/treetag(): now throws an error if you try to combine a
language preset with TreeTagger's batch files as the tagger to use; some users
seem to be confused about what to configure, and this error message
hopefully helps them to understand why treetag() must fail in these cases
treetag(): newer versions of TreeTagger will no longer have "utf8" in
their parameter and abbreviation files. since we never know what version of
TreeTagger we're dealing with, treetag() will from now on look for files
with "utf8" if specified in the language package, but not fail if none is
found, but also try for a non-labelled file and replace the file name on the
fly if one is found
grapheme clusters: in UTF-8, certain characters in some languages are
shown as a single character, but technically are several characters combined.
nchar() counts all combined parts individially, which in most use cases
for this package is not what one expects. it now uses nchar(type="width")
for a letter count that is much closer to user's expectations
set.lang.support(): explicitly set the sorting method for factor levels
to "radix" as the new default "auto" (R >= 3.5) produced unstable results
with different setups; hence some of the test standards also had to be
updated
templates: incomplete package name in license header
read.BAWL(): updated download URL and added DOI
the startup check for available language packages was reduced to short
hints to available.koRpus.lang() and install.koRpus.lang()
the startup message can now be suppressed by adding "noStartupMessage=TRUE" to the koRpus options in .Rprofile
treetag()/tokenize(): fixed an issue with sentence numbering which was
triggered if all sentences were of equal length
query(): method failed for columns which are now factors
treetag(): koRpus no longer fails with an error if unknown tags are
found. there will be a warning, but you can continue to work with the object
depends on R >= 3.0.0 now
improved available.koRpus.lang() to make it more obvious how to install
language support packages, and which
session settings done with set.kRp.env() or queried by get.kRp.env() are
no longer stored in an internal environment but the global .Options; this
also allows for setting defaults in an .Rprofile file using options()
in the docs, improved the link format for classes, omitting the "-class" suffix
set.lang.support(): the levels of tag, wclass, and desc are now
automatically sorted; test standards had to be adjusted accordingly
set.lang.support(): new argument "merge"; it is now possible to add or
update single POS tag definitions
new class object contructors kRp_tagged(), kRp_TTR(), kRp_txt_freq(),
kRp_txt_trans(), kRp_analysis(), kRp_corp_freq(), kRp_lang(), and
kRp_readability() can be used instead of new("kRp.tagged", ...) etc.
this is a testing release introducing major changes in the way language support is handled (see other changes in this log). tl;dr: you must install additional koRpus.lang.** packages to fully restore the previous functionality, i.e., all supported languages. see ?install.koRpus.lang
treetag(): with TT.tknz=FALSE, the last letter of a text was truncated
due to a missing newline at the end of the tempfile (thanks to adam
spannbauer for both reporting and fixing it)
treetag(): hopefully fixed a nasty encoding issue on windows, again
treetag(): fixed an issue that could be triggered by hard to tokenize
texts exceeding a default limit of summary() for factors
treetag()/tokenize(): silenced warnings of readLines() for missing final
EOL of input files
language support: while the sylly package is released on CRAN now, its
separate language packages were not allowed to be published there as well. a
special repository was therefore set up on gitub and added via the
"Additional_repositories" field to the DESCRIPTION file. however, not having the
sylly.XX packages on CRAN made it necessary to further modularize the
package and complete remove all out-of-the-box language support (see removed
section). all these support packages for language are now being resolved
by installing from that repo instead of CRAN.
package loading: when koRpus is being loaded, it now checks for available (i.e. already installed) language packages. if none are found, it asks you to install one. i'm sorry for the unconvenience
vignette is now in RMarkdown/HTML format; the SWeave/PDF version was dropped
tif_as_tokens_df(): new method to get TT.res in fully TIF compliant format
new functions available.koRpus.lang() and install.koRpus.lang() for more
convenient handling of language support packages.
language support: koRpus previously supported some languages directly (de, en, es, fr, it, and ru). this support had to be removed and is now available as separate language packages via https://undocumeantit.github.io/repos/l10n
kRp.lang: fixed the show() and summary() methods to omit country
information which was dropped from the UDHR data a while ago
treetag(): windows users might run into problems because of differences
between the file separators R uses internally when they are also used in
shell() calls. this hasn't been an issue earlier, but is worked around now
anyway. hope this doesn't cause new issues...
kRp.tagged: the TT.res data.frame of the object class has new columns
"doc_id", "idx" (index), and "sntc" (sentence), with "doc_id" now being the
first column before "token" to comply with the Text Interchange Formats
proposed by rOpenSci
kRp.tagged: in TT.res, the columns "tag", "wclass" and "desc" are no
longer character vectors but factors. this doesn't actually change the class
definition, as TT.res just has to be a data.frame, but it reduces the
object size especially for larger texts, and makes it much simpler to do
analysis with these objects
tokenize()/treetag()/read.tagged(): these functions now add token index
and sentence number to the resulting objects; document ID is added if
provided
kRp.lang: depending on the information available in the UDHR data, the
show() and summary() methods' output is now dynamically adjusted; summary()
now also lists the columns "iso639-3" and "bcp47" by default
treetag(): debug output for tokenize() looks a little nicer
kRp.text.transform(): the old function is now deprecated and was replaced
by a proper S4 method called textTransform(). the old one will work for
the moment, but you'll get a warning
the tt slot in class kRp.TTR gained two new entries called "type.in.txt"
and "type.in.result", which will contain a list of all types with the
index where it is to be found in the original text or the lex.div() results
respectively, if type.index=TRUE; the indices might differ because the
result might be stripped of certain word classes
treetag()/tokenize(): internal workflow for adding word class and
description of tags was modularized for more detailed control. you can now toggle
whether you want the verbose description of each tag added directly to
objects with the new argument "add.desc". it is set in the environment by
set.kRp.env() and defaults to FALSE, making the objects about 5% smaller in
memory.
kRp.corp.freq: the class gained a new slot called "caseSens", documenting
whether the frequency statistics were calculated case sensitive (see
read.corp.*() below).
validity check for objects of class kRp.tagged is a bit more liberal when
TT.res doesn't have all expected columns and suggests to call fixObject()
(see below) instead of failing with an error
adjusted unit tests
summary(): method for class kRp.TTR now also supports the logical "flat"
argument
new "[" and "[[" methods can be used to directly address the data.frames
in tagged or hyphenated objects. that is, you don't have to call
taggedText() or hyphenText() first, it will be done internally
new "[" and "[[" methods have also been added for objects of classes
kRp.TTR and kRp.readability for quick access to their summary() results (index
by measure)
treetag(): a new check will throw an informative error message if
TreeTagger didn't return something the function can use
lex.div() et al.: new option "type.index" to produce the indices
described above in the "changed" section
hyphen(): new option "as" to set the return value class, still defaults
to "kRp.hyph", but can also be "data.frame" or "numeric"
new shortcut methods hyphen_df() and hyphen_c() use different defaults
for "as"
treetag()/tokenize(): new option "add.desc" (see changed section)
taggedText(): new option "add.desc" to (re-)write the "desc" column in
the data.frame, useful if it was omitted during treetag()/tokenize() but you
want to add it later without retagging everything
read.corp.LCC()/read.corp.celex(): added new option "caseSens" to toggle
whether frequency statistics should be calculated case sensitive or
insensitive
new method fixObject() can upgrade old tagged objects from previous
koRpus releases, i.e. add missing columns and adjust data types where needed
hyphen(): all parts of the package that were specific for hyphenation
were removed as they are now part of the new sylly package. this includes the
class definitions (kRp.hyph.pat and kRp.hyphen) and methods (correct(),
hyphen(), show() and summary()) for those classes, as long as they in turn
are not specific to koRpus. the hyphenation definitions were also removed
from the language support files, as they are now part of individual
language packages for the sylly package (sylly.en, sylly.de, etc.) that this
package now depends on. you should, however, notice no difference in using
the package, everything should just work like it did before this split.
the standard generics for describe() and language() were removed because
they are now defined in the sylly package
leftover typo in lang.support-en.R referencing "utf8-tokenize.pl" instead of "utf8-tokenize.perl" in the windows preset and a call to grep that is not present in Treetagger's *.bat file
readability(): fixed a minor issue with the internal handling of wrongly
tagged dashes in the FOG formula (shouldn't have any effect on results)
if no encoding is provided and treetag() needs to write temporary files,
output file encoding is now forced into UTF-8
hyphen(): caching now uses an environment instead of a data.frame. this
means that old cache files will need to be changed as well. hyphen() will
try to convert them on the fly, but if this fails you should remove the old
files
hyphen(): cached results are now looked up much more efficient, speeding
up the process drastically (about 100 times faster in my benchmarks!)
hyphen(): hyphenation patterns are now internally converted to
environments which speeds up uncached runs (or first runs with cache) noticeably
readability(): default parameters are now always fetched by the internal
function default.params(), individually for each index
source code: moved all wrapper functions for readability() and lex.div()
from individual source files to one wrapper file, respectively. the source
tree became a bit overcrowded over the years
new options redability(index="validation") and
lex.div(measure="validation") show current the status of validation. this info was previously only
available as comments in the source code and is now directly available.
WSFT(): deprecated wrapper, was replaced by nWS() in 2012
windows users could run into an error of an undefined object
(TT.call.file) when using treetag()
CRAN doesn't accept leading zeroes in version numbers any longer and asked me to change 0.07 into 0.7. i'd rather play this safe, so i'm jumping right to 0.10 to keep the versioning consistent fo all users. the reason for this policy change was not explained to me, could be anything from "we think it looks ugly" to "it breaks our build systems".
allowing treetag() to run even when a defined lexicon file is not found.
this previously resulted in an error and now causes only a warning message.
the show method for Flesch Brouwer was not working properly
if a cache file for hyphen is set but not existing, it will be created automatically
the manual page for the wrapper function ELF() attributed the index to
Farr, when it was in fact Fang (as correctly said in ?readability);
vigilantly spotted by Mario Martinez
calling lex.div() on untagged character vectors didn't really work yet
guess.lang() had problems with newer UDHR files which included comments
in the index.xml file
shiny app: was omitting the row names of tables in newer versions of shiny
treetag() appended the abbreviation list two times in english preset
TT.options checks in treetag() do no longer ask for mandatory options if
TT.cmd is not "manual"
updated shiny app: disabling FOG by default (faster), adding Brouwer and MTLDMA.steps options, adding dutch and portuguese by default, disabled language selection in language guessing tab
shiny app: using fluidPage() now
shiny app: set tables to use bootstrap striped layout
reaktanz.de supports HTTPS now, updated references
new summary() method for kRp.hyph objects
new show() methods for kRp.hyph and kRp.taggedText objects
new methods tokens() and types() to quickly get tokens and types of a text
the treetag() function actually omittet options for the tokenizer due to
a never updated variable and a wrong setting later on; this has been the
case for years – interesting that no-one ever noticed this
read.corp.LCC() can now digest newer LCC archives, omitting the
*-meta.txt file if none is present, and also supporting *-words.txt files with
duplicate columns
some typos in the ChangeLog...
fixed manual page for class kRp.corp.freq
the support for non-UTF-8 presets for was removed, since TreeTagger is only endorsing UTF-8 encoding itself for a while; the old preset names will continue to work for the time being, but if possible you should already rename them from "<lang>-utf8" into just "<lang>" in your scripts
removed options corp.rm.class and corp.rm.tag from method hyphen() for
character strings
massively improved the speed of hyphen by using a new method for exploding words into their sub-parts. in benchmark tests (text with ~30.000 words) the new method only takes about 15% of the time without cache, and about 50% with cache
massively improved the speed of lex.div() by reducing unnecessary
computations. in benchmark tests (see above) the new method is more than 100
times faster, which also makes readability() three times as fast with standard
indices. if you disable the FOG index, readability() is now finished in
an instant, too. see the new index="fast" option below
tokenize() now uses data.table() instead of data.frame() internally,
leading to an increase in speed of about 20%
new slots "bigrams" and "cooccur" in S4 class kRp.corp.freq
cleaned up code
removed the never used variable TT.tknz.opts.def in the language support
set.lang.support() now checks for duplicate tag definitions and throws an
error if any were found
renamed class and method files to set some environment first
moved several internal hyphenation functions to koRpus-internal.hyphen.R
moved several internal readability functions to koRpus-internal.rdb.formulae.R
read.corp.LCC() can now import the information on bigrams and
co-occurences of tokens in a sentence
language support now also uses TT.splitter, TT.splitter.opts, and TT.pre.tagger, which was needed mostly to implement the TreeTagger script for portuguese (available in the separate package koRpus.lang.pt), but also for updates of languages that were already supported
updated the RKWard plugin (UTF-8 defaults, added dutch and portuguese, added Brouwer formula)
new unit tests for lex.div(), tokenize() and readability()
new options to set index="fast" in readability() to drop FPG from the
defaults for faster calculations
new option MTLDMA.steps to increase the step size for MTLD-MA. this diverts from the original proposal, but if your text is long enough, you will get a very good estimate and only need a fraction of the computing time
fixed the Douma formula: based on available literature, the factor for average sentence length was set to 0.33, but the original paper reported it as 0.93
fixed the documentation for tokenize(), roxygen2 had problems with an
escaped double quote
corrected some problems with umlauts in the docs
new template for a roxyPackage script to make it easy to build packages from language support scripts
additional validation for ARI, flesch (en), flesch-kincaid, SMOG and FOG, via http://wordscount.info/wc/jsp/clear/analyze_readability.jsp
new Flesch parameters to calculate readability according to Brouwer (NL),
can be invoked as index "Flesch.nl-b", "Flesch.Brouwer", or Flesch
paremeters set to "nl-b"
now the manual is actually documenting all the various Flesch formulas, i.e., listing all parameter values, so that it's easier for users to check what is being calculated
workaround for missing POS tag "NS" for english texts
made guess.lang() compatible with recent format of UDHR archives, now
using ISO 639-3 codes as language identifier
tokenize() and treetag() weren't able to cope with text that only
consisted of a single token
declared import from graphics package to satisfy CRAN checks
updated rkwarddev script according to recent development in the rkwarddev package
some basic validity checks of treetag()s "TT.options" moved to an
internal function checkTTOptions(), which is now also called by set.kRp.env()
guess.lang() doesn't warn about missing EOL in the UDHR texts any longer
added a README.md file
new option "no.unknown" can be passed to the "TT.options" of treetag(),
to toggle the "-no-unknown" switch of TreeTagger
new option "validate" for set.kRp.env() to enable/disable checks
actually query for supported POS tags in internal function
is.supported.lang(). the function previously looked for supported languages in the
available presets, which failed if there was no preset named like the language
abberviation
made hyphen() not split words after first or before last character,
therefore min.length was increased to 4 accordingly
adjusted test standards to changed hyphen results
read.tagged() does now also accept matrix objects, see
https://github.com/unDocUMeantIt/koRpus/issues/1
read.corp.custom() calculated the in-document frequency wrong if analysis
was performed case insensitive
updated some more links in the docs (?kRp.POS.tags)
correct.tag() now accepts all objects of class union kRp.taggedText
query() now uses "%in%" instead of "==" to match character strings
against "query"
exported the previously internal function set.lang.support(), to prepare
for the possibility of third party package to add new languages
initial support to manually extend the languages supported by the
package. you can now add new languages on-the-fly in a running session, or in a
more sustainable manner by providing a language package (using the same
methods, basically). key to this is the now globally available function
set.lang.support(), and there's also two commented template scripts installed
with the package, see the "templates" folder
read.corp.custom() was buggy when dealing with tagged objects
suppress message stating text language in summary() for readability
objects if "flat=TRUE"
changed the following functions into S4 methods: readability(),
lex.div(), hyphen(), read.corp.custom() and freq.analysis()
removed long since deprecated function kRp.freq.analysis()
splitted the code of the monolithic internal function for
read.corp.custom() into several subfunctions to get more flexibility
read.corp.custom() now also supports analysis of lists of tagged objects
removed option "fileEncoding" from the signature of read.corp.custom(),
but it can still be used as part of the "..." options; this was neccessary
because treetag() uses "encoding" instead
new option "tagger" now also available in read.corp.custom()
there is now a mailing list to discuss the koRpus development: https://ml06.ispgateway.de/mailman/listinfo/korpus-dev_r.reaktanz.de
changed "selected" values of checkboxGroupInput() in the shiny file ui.R
to comply with the changes made in shiny 0.9.0
function kRp.text.transform() was missing some columns in TT.res
fixing this ChangeLog: the parameter for Szigriszt (Flesch ES) is not
"es2", as reported in the log to koRpus 0.05.3, but "es-s"!
calling readability for "ARI.NRI" without hyphenation didn't work,
allthough ARI doesn't need syllables
updated some broken links in the docs (?kRp.POS.tags, ?guess.lang)
added imports for 'utils' and 'stats' packages to comply with new CRAN checks
added a otherwise useless definition of "text" to the body of
guess.lang(), also to satisfy R CMD check
replaced the RKWard plugin with a modularized rewrite (rkwarddev script)
some code cleaning in internal function kRp.rdb.formulae() and
freq.analysis(), mostly replacing @ by slot()
new readability formula tuldava(), kindly suggested by peter grzybek
the shiny app has gained support for Tuldava and Szigriszt (Flesch ES) formulae and log.base parameter (lexical diversity)
set.kRp.env() does now check whether a language preset is valid
removed Snowball from the list of suggested packages, as it is deprecated and fully replaced by SnowballC
re-generated all docs with roxygen2 3.1.0, which can now handle S4 class definitions properly
replaced all tabs in the source code by two space characters
new tf-idf feature: read.corp.custom() now calculates idf, then
freq.analysis() can use that to calculate tf-idf, kindly suggested by sandro tsang
new columns "inDocs" and "idf" in slot "words" of class kRp.corp.freq
new columns "tf", "idf" and "tfidf" in slot "words" of class kRp.txt.freq
PCRE 8.34 caused the tests to fail because of problems with regular
expressions in internal tokenizing function tokenz(); fixed by ensuring that
"-" is being escaped as "\\-"
due to a logical bug in calls to internal functions, the "lemmatize"
argument if lex.div() didn't really have any effect
using file names with readability() and its wrappers was broken, works
again now
the "tt" slot in class kRp.TTR gained two new entries, "lemmas" and
"num.lemmas", kindly suggested by roberto trunfio
show() method for kRp.TTR objects now also lists the number of lemmas (if
found)
parameters of Flesch formulae were slightly changed to be more accurate (from rounded values of 206.84 to 206.835) where applicable
Flesch-Szigriszt and Fernandez-Huerta have been validated against INFLESZ v1.0, so the warning was removed
readability.num() now gracefully accepts a single number of syllables for
formulae who don't need to know more
added a proper GPL notice at the beginning of each R file
adjustet tests according to the changes made
alternative Flesch parameters for spanish texts according to Szigriszt
were added as parameters="es2", kindly suggested by carlos ortega
this is the first version of the package with slightly reduced sources on CRAN – the debian directory, GPL license file and hyphenation pattern ChangeLog had to be removed. if you want the full sources to this package, please use the packages provided at http://reaktanz.de/?c=hacking&s=koRpus
added two previously undocumented (and hence missing) italian tags "FW"
and "LS"
removed some ::: operators which were not neccessary
updated slot "param" of kRp.TTR objects to include "min.tokens",
"rand.sample", "window" and "log.base"
moved some parts of treetag() and kRp.text.paste() to internal functions
for easier re-use of its functionality
support for marco baroni's TreeTagger tagset for italian was added
added SnowballC to the suggested packages, as tokenize() and treetag()
can also use SnowballC::wordStem() for stemming
new function read.tagged() can be used to import already tagged texts
new argument "apply.sentc.end" in function treetag()
new argument "log.base" in functions lex.div() and lex.div.num()
DRP() readability formula tried to fetch a non-existing variable and
hence didn't calculate; this also fixed a problem with summary(), if DRP
results were expected in the object; tests had to be corrected as well
textFeatures() gets number of letters and TTR again
MTLD calculation (lex.div()) now counts a factor as full if it is <
factor.size, it was implemented as <= factor.size before (thanks to scott
jarvis for insight on the details)
summary() for kRp.TTR objects always showed MTLD, even if it was empty
vignette now describes the use of taggedText() and describe(), instead of
direct access to slots
readability() now assumes that if there's any text, it represents at
least one sentence, even if no sentence ending punctuation can be found
"quiet=TRUE" in readability(), readability.num(), lex.div() and
lex.div.num() will now also suppress all warnings regarding validation status
MTLD calculation (lex.div()) was optimized and takes less than half of
the time it used to. it also gained a new boolean argument "detailed", which
is FALSE by default. this means that the full factor results are skipped
now, which boosts performance even more (six times as fast as before)
the caching mechanism for hyphen() was restructured into internal
functions, allowing for better access to the cached data
set.kRp.env() and get.kRp.env() have new signatures, namely, all
previously hardcoded parameters have been replaced by the more flexible "...".
usage stays the same, so there's no need to change any scripts, as long as
you called all parameters by name, not only by position!
object class kRp.corp.freq can now have additional columns in slots
"words" and "desc". this flexibility allows for using this class with valence
data as well
query() now examines the desired columns to decide whether character or
numeric operations are to be done
performance of hyphen() has been massively improved if cache=TRUE
guess.lang() now also standardizes the difference values; this was added
to the respective summary() method, which also produces nicer output
the source code was re-organized a bit, to ensure classes and methods are found in an appropriate order; the collate roclet of roxygen2 had problems with this when running in R 3.0.0
new function read.BAWL() to import BAWL-R data
new demo application for use with the "shiny" package, can be found in
$SRC/inst/shiny
lex.div() now supports a new method for calculating MTLD (MTLDMA,
moving-average)
new getter method hyphenText() to access the "hyphen" slot in kRp.hyphen
objects
getter methods language() and describe() for kRp.hyphen objects also added
added "quiet" argument to lex.div.num()
guess.lang() can now analyze a given text directly, not only from files
set.kRp.env() can now explicitly unset parameters in the environment
set.kRp.env() and get.kRp.env() know a new parameter,
"hyphen.cache.file", which can be set to a file name to read from/write to the hyphenation
cache. this way you can easily restore cached hyphenation rules over
sessions. if this parameter is set, it will be used by hyphen() automatically if
called with "cache=TRUE"
removed some non-ASCII characters, mostly from comments, to keep the package on CRAN; some author names are now spelled wrong, though...
optimized tokenize() to also detect prefixes/suffixes of the defined
heuristics if they co-occur with punctuation
re-saved hyph.fr.rda with explicitly UTF-8 ecoded vectors
renamed LICENSE to LINCENSE.txt, so it won't get installed, as demnanded by Writing R Extensions
the language specific heuristics "en" and "fr" in tokenize() were renamed
into "suf" and "pre". but they are still available, with "fr" now
activating both "suf" and "pre".
read.hyph.pat() now explicitly sets vector encoding to UTF-8 with
Encoding()<-, to ensure that the generated objects don't cause warnings from R
CMD check if they're included in packages
internally replaced paste(..., sep="") with paste0(...)
added new getter/setter methods taggedText(), taggedText()<-, describe(),
describe()<-, language() and language()<- for tagged text objects
added is.taggedText() test function
added a warning to treetag() if "TT.options" is not a list (because this
will likely render the options meaningless if they *contain* a list).
tokenize() can now apply a list of patterns/replacements to given texts
via the new "clean.raw" attribute, and even supports perl-like regular
expressions. the replacements are done before the texts are tokenized, so this
can be tried to globally clean up bad characters or simply replace
strings, etc.
tokenize() and treetag() have a new option "stopwords" to enable stopword
detection
kRp.filter.wclass() can now remove detected stopwords
tokenize() and treetag() have a new option "stemmer" to interface with
stemmer functions/methods like Snowball::SnowballStemmer()
added support for french (thanks to alexandre brulet)
a typo in Spache calculation (substraction instead of addition of a constant) lead to wrong results
Spache now counts unfamiliar words only once, as explained in the original article
old Spache formula was missing in readability(index="all")
validated Linsear Write, Dale-Chall (1948) and Spache (1953) results and removed warnings
status messages of hyphen() and lex.div() have been replaced by a space
saving prograss bar added
added tests for lex.div(), hyphen() and readability()
tests should now work on any machine
using utf8-tokenizer.perl now in all UTF-8 presets, also on windows systems. the script is part of the windows installer of TreeTagger 3.2 (at least since june 2012)
correct.*() methods now also update the descriptive statistics in corrected objects
there's now a class union "kRp.taggedText" with the members "kRp.tagged",
"kRp.analysis", "kRp.txt.freq" and "kRp.txt.trans"
advanced summary() statistics for objects returned by clozeDelete()
clozeDelete(offset="all") now iterates through all cloze variants and
prints the results, including the new summary() data
clozeDelete() now uses the new class union "kRp.taggedText" as signature
read.corp.custom() now uses table(), "quiet" is TRUE by default, the new
option "caseSens" can be used to ignore character case, and "corpus" can
now also be a tagged text object
summary() for objects of class kRp.txt.freq was broken
as("kRp.tagged") for objects of class kRp.txt.freq was broken
elaborated documentation for method cTest()
added new method clozeDelete()
added new list "cTest" in desc slot of the objects returned by cTest(),
which lists all words that were changed (in clozeDelete() this list is
called "cloze")
added new function jumbledWords() and new method cTest()
kRp.text.paste() now also removes superfluous spaces at the end of texts
(i.e., before the last fullstop)
koRpus now suggests the "testthat" package and uses it for automatic tests
treetag() and tokenize() now also accept input from open connections
treetag() shouldn't fail on file names with spaces any more
added features:
kRp.corp.freq class objects now include the columns 'lttr', 'lemma', 'tag' and 'wclass'
query() for corpus frequency objects now returns objects of the same
class, to allow nested queries
the 'query' parameter of query() can now be a list of lists, to
facilitate nested requests more easily
query() can now invoke grepl(), if 'var' is set to "regexp"; i.e., you
can now filter words by regular expressions (inspired by suggestions after
the koRpus talk at TeaP 2012)
fixed bug in summary() for tagged objects without punctuation
renamed kRp.freq.analysis() to freq.analysis() (with wrapper function for
backwards compatibility)
readability.num() can now directly digest objects of class kRp.readability
data documentation hyph.XX is now a roxygen source file as well
cleaned up summary() and show() docs
adjustements to the roxygen2 docs (methods)
code cleanup: initialized some variables by setting them NULL, to avoid
needless NOTEs from R CMD check (hyphen(), and internal functions
frqcy.by.rel(), load.hyph.pattern(), tagged.txt.rm.classes() and
text.freq.analysis())
re-formatted the ChangeLog so roxyPackage can translate it into a NEWS.Rd file
prep for CRAN release:
0.04-26 was short-lived...
really fixed plot docs
removed usage section from hyph.XX data documentation
renamed text.features() to textFeatures()
encapsulated examples in set.kRp.env()/get.kRp.env() in \dontrun{}
re-encoded hyph.XX data objects to UTF-8
replaces non-ASCII characters in code with unicode escapes
fixed plot docs
prep for inital CRAN release
re-compressed all hyphenation pattern data files, using xz compression
lifted the R dependency from 2.9 to 2.10
compressed LCC tarballs are now detected automatically
kRp.freq.analysis() now also lists the log10 value of word frequencies in
the TT.res slot
in the desc slot of kRp.txt.freq class objects, the rather misleading
list elements "freq" and "freq.wclass" were more adequately renamed to
"freq.token" and "freq.types", respectively
unmatched words in frequency analyses now get value 0, not NA
fixed wrong signature for option "tagger" in kRp.text.analysis()
fixed kRp.cluster() which still called some old slots
fixed bug for attempts to calculate value distribution texts without any sentence endings
all readability wrapper functions now also accept a list of text features for calculation
class kRp.readability now inherits kRp.tagged
readability() now checks for presence of a hyphen slot and re-uses it, if
no new hyphen object was provided; this in addition to the previous
change enables one to re-analyze a text more efficiently, as already calculated
results are also preserved
letter and character distribution in kRp.tagged desc slot now include columns with zero values if the respective values are missing (e.g., no words with five letters, but some with six, etc.)
added summary method for class kRp.tagged, summarizing main information from the desc slot
added plot method for class kRp.tagged
show method for kRp.readability now lists unfamiliar words for Harris-Jacobson
cleaned up code of lex.div.num() a bit
added precise RGL formula option to FORCAST
removed validation warnings from several indices, because results have
been checked against those of other tools, and were comparable, so the
implementations of these measures are assumed to be correct: - lex.div(): TTR,
MSTTR, C, R, CTTR, U, Maas, HD-D, MTLD (thanks a lot to scott jarvis &
phil mccarthy for calculating sample texts!) - readability(): ARI, ARI NRI,
Bormuth, Coleman-Liau, Dale-Chall, Dale-Chall PSK, DRP,
Farr-Jenkins-Paterson, Farr-Jenkins-Paterson PSK, Flesch, Flesch PSK, Flesch-Kincaid, FOG,
FOG PSK, FORCAST, LIX, RIX, SMOG, Spache, Wheeler-Smith
moved all calculation from readability() to an internal function
kRp.rdb.formulae(). to make it easier to write a similar function to lex.div.num()
for the readability fomulas as well
added readability.num()
adjusted exsyl calculation for ELF to the approach used in other
measures, which also results in a change of its default "syll" parameter from 1 to
2; also corrected a typo in the docs, the index was proposed by Fang, not
Farr
readability results now list letter distribution, not character distribution in desc slot
the desc slot from readability calculations was enhanced so that it can
directly be used as the txt.features parameter for readability.num()
docs were polished
further fixes to the Wheeler-Smith implementation. according to the original paper, polysyllabic words need to be counted, and the example given shows that this means words with more than one syllable, not three or more, as Bamberger & Vanecek (1984) suggested
fixed HD-D, previous results are now labelled as ATTR in the HDD slot
adjusted HD-D.char calculation for small number of tokens (probabilities are now set to 1, not NaN)
added MATTR characteristics
show() for lex.div() objects now also reports SD for characteristics
MTLD now uses a slightly more efficient algorithm, inspired by the one used for MATTR
MSTTR now also reports SD of TTRs
differentiated the word class adposition into pre-, post- and circumposition in the language support for german and russian
added both Tränke-Bailer formulae to readability(), incl. wrapper
traenkle.bailer() and show()/summary() methods
Coleman formulae now also count only prepositions as such
fixed Wheeler-Smith (thanks to eleni miltsakaki)
added Moving Average TTR (MATTR) to lex.div(), incl. wrapper MATTR() and
show()/summary() methods
added "rand.sample" and "window" to the parameters returned by lex.div()
further re-arranged the code of readability() and lex.div() to make it
easier to maintain
summary(flat=TRUE) for readability objects is now a numeric vector
added five harris-jacobson readability formulae, incl. wrapper
harris.jacobson() and show()/summary() methods
updated vignette
MTLD characteristics are now twice as fast
classes "kRp.txt.freq" and "kRp.txt.trans" now simply extend
"kRp.tagged", and "kRp.analysis" extends "kRp.txt.freq"
removed internal function check.kRp.object() (globally replaced by
inherits())
fixed letter count issue in readability()
fixed bugs in loading word lists in readability()
fixed crash if index="all" in readability()
reordered default kRp.readabilty slot order alphabetically, as well as
show() and summary() for readability results
renamed results of the Neue Wiener Sachtextformeln from WSTF* to nWS* in
readability object methods show() and summary() for consistency
renamed WSFT() to nWS() for the same reason
cleaned up roxygen comments for more roxygen2 compliance
added missing word exclusion to Gunning FOG measure
added sentence length, word length, distribution of characters and
letters to "desc" slot of class kRp.tagged and readability() results, where
missing
both syllable (hyphen()) and character distributions gained inversed
cummulation for absolute numbers and percentages, so this one table now makes
it easy to see how many words with more/equal/less characters/syllables
there are in a text
changed internals of kRp.freq.analysis() and readability() to re-use
descriptives of tagged text objects
NOTE: this also changed the names of some result elements in their "desc"
slots for overall consistency ("avg.sent.len" is now "avg.sentc.length",
"avg.word.len" became "avg.word.length", and instances of "num.words",
"num.chars" etc. lost the "num." prefix). in case you accessed these
directly, check if you need to adopt these changes. this is a first round of
changes towards 0.05, see the notes to 0.04-17 below!
replaced the english hyphenation parameter set with a new one, which was made with PatGen2 especially for koRpus
tokenize() will now interpret single letters followed by a dot as an
abbreviation (e.g., of a name), not a sentence ending, if heuristics include
"abbr"
fixed bug which caused hyphen() to drop syllables if only one pattern
match was found
added cache support to the correct method of class kRp.hyphen
added number of words and sentences to "desc" slot of class kRp.tagged
elaborated treetag() error message if no TreeTagger command was specified
NOTE: koRpus 0.05 will likely merge some object classes similar to kRp.tagged, i.e. kRp.txt.freq and kRp.txt.trans, into one class for tokenized text, either replacing or inheriting those classes
added slot "desc" to class kRp.tagged, to have descriptive statistics
directly available in the object
added support for descriptive statistics to tokenize() and treetag()
added function text.features() to extract a 9-features set from texts for
authorship detection (inspired by a talk at the 28C3)
hyphen() can now cache results on a per session basis, making it
noticeably faster
manage.hyph.pat() is now an exported function
added initial support for italian (thanks to alberto mirisola)
added italian hyphenation patterns
changed min.length from 4 to 3 in hyphen() and manage.hyph.pat()
hyphen now considers hyphenating before last letters of a word
tuned hyph.en (with contributions by laura hauser)
fixed check for existing tokenizer, tagger and parameter file in treetag()
fixed MTLD calculation for texts which don't make even one factor
added new internal function manage.hyph.pat() to add/replace/remove
pattern entries for hyphenation
added number of tokens per factor and standard deviation to MTLD results (thx to aris xanthos for the suggestion)
added column "token" to slots MTLD$all.forw and MTLD$all.back of
lex.div() results, so you can verify the results more easily
slot HDD$type.probs of lex.div() results is now sorted (decreasing)
removed warnings of missing encoding, since enc2utf() seems to do a
pretty good job
added support for the newer LCC .tar archive format
changed vignette accordingly
for consistency, changed "words" and "dist.words" into "tokens" and
"types" in class kRp.corp.freq, slot desc
added lgeV0 and the relative vocabulary growth measures suggested by Maas
to lex.div(); furthermore, a is now reported instead of a^2
added lgV0 and lgeV0 to lex.div.num()
show method for class kRp.TTR now excludes Inf values from charasteristics values
added function lex.div.num(), calculates TTR family measures by numbers
of tokens and types directly
cleaned up lex.div() code a little
fixed missing 'input.enc' information if treetag() option 'treetagger' is
not "manual" but a script
enhanced encoding handling internally if none was specified
changed default value of 'case.sens' to FALSE in lex.div(), as this seems
to be more common
changed default value of 'fileEncoding' from "UTF-8" to NULL and use
enc2utf() internally if no encoding was defined
tokenize() now converts all input to UTF-8 internally, to prevent
conflicts later on (treetag() does that since 0.04-7 already)
added an experimental feature to treetag() to replace TreeTagger's
tokenizer with tokenize()
fixed bugs in treetag(): "debug" now works without "manual" config as
well, and global TT.options are now found if no preset was selected
added "encoding" option to treetag() and defaults to the language presets
fixed some option check and file path issues in treetag()
fixed package description for R 2.14
fixed dozends of small glitches in the docs which caused warnings during package checks
fixed bug in getting the right preset: mixed "lang" and "preset" during
the modularization
modularized language support by the internal function set.lang.support(),
this should make it much easier to add new languages in the future,
because it means to add only one R file. hyphen(), kRp.POS.tags() and treetag()
now use this new method
added CITATION file
fixed duplicate "PREP" definition in spanish POS tags, which caused
treetag() to consume lots of RAM
fixed superfluous "es" definitions in treetag()
added support for spanish (thanks to earl brown)
docs can be created from source by roxygen2 (but all class docs are static, until '@slot' works again)
added support for autodetection of headlines and paragraphs in tokenize()
added support to revert autodetected headlines and paragraphs in
kRp.text.paste()
updated RKWard plugin to use tokenize()
added parameters for formula C and simplified formula to SMOG
enhanced readability formulas (like adding age levels to Flesch.Kincaid, grade levels to LIX)
removed the duplicate Amstad index (is now just Flesch.de)
added the full RKWard plugin as inst/rkward, so both get updated simultanously
added experimental internal functions to import result logs from Readability Studio and TextQuest
integrated internal tags to kRp.POS.tags(), so tokenize() can return
valid kRp.tagged class objects, i.e. substitute TreeTagger if it's not
available
consequently renamed 'treetagger' option into 'tagger' in readability(),
kRp.freq.analysis() and kRp.text.analysis()
lots of small fixes
added a simple tokenize() function
first working version of read.corp.custom()
added "..." option to readability, kRp.freq.analysis and
kRp.text.analysis, to configure treetag()
added TT.options to the get/set environment functions
changed default values for treetag() (for readability)
fixed bug in internal check.file() function (mode="exec" returned TRUE
too soon)
added warning messages to readability() and lex.div() to make people
aware these implemetations are not yet fully validatied
introduced release dates in this ChangeLog ;-) (reconstructed them for earlier releases from the time stamps on the server)
added "desc" slot with some statistics to class kRp.hyphen and hyphen()
added grading information for Flesch and RIX measures
fixed grading for Wheeler-Smith formula
introduced "quiet" options for hyphen(), lex.div() and readability()
further improved the vignette, elaborated on the examples
fixed typo in kRp.POS.tags("ru"): "Vmis-sfa-e" tags no longer a "vern",
but a "verb"
removed XML package dependency again, by writing a small parser (there was no windows binary for the XML package, which was obviously a problem...)
fixed "quiet" option in guess.lang()
fixed bug in calculation of sentence lengths in kRp.freq.analysis()
(counted punctuation as words)
tweaked hyph.en patterns to get better results
solved a small charset issue in treetag()
fixed hyphen() output if doubled hyphenation marks appeared
elaborated the vignette a little (including some references)
added support for zipped LCC database archives to read.corp.LCC()
improved handling of unknown POS tags: now causes an error dump for debugging
added query() method to search in objects of class kRp.tagged
de-factorized treetag() output
fixed hyphenation problems (remove all non-characters for hyphen())
fixed missing "”" and "$" POS tags in kRp.POS.tags("en")
renamed kRp.guess.lang() to guess.lang()
guess.lang() now gzips only in memory by default, saves about 1/8 of
processing time - added option "in.mem" to switch back to previous behavious
(temporary files)
added internal function is.supported.lang() as a possible wrapper for
guessed ULIs
added internal functions roxy.description() and roxy.package() to ease
development
added support for automatic language determination: - changed internal
function compression.ratio() to txt.compress() - added internal function
read.udhr() - added kRp.guess.lang() and class kRp.lang
added class kRp.txt.trans for results of kRp.text.transform()
enhanced function kRp.text.transform(), most notably calculate differences
added function kRp.text.paste()
added function kRp.text.transform()
fixed hyphen() bug (leading dots in words caused functions to fail)
added kRp.filter.wclass()
added TODO list to the sources
fixed another bug in frequency analysis with corpus data (superfluous class definition)
fixed missing POS tags: refinement of english tags (extra tags for "to be" and "to have")
added more to the vignette
added .Rinstignore file to clean up the doc folder
began to write a vignette
fixed treetag() failing on windows machines (hopefully...)
added TRI readability index
fixed bug in frequency analysis with corpus data (wrong class definition)
fixed bug in Bormuth implementation (didn't fetch parameters)
fixed missing Flesch indices in summary method
corrected display of FOG indices in summary method (grade instead of raw)
added compression.ratio() to internal functions
enhanced query() methods
fixed some typos and smaller bugs
initial public release (via reaktanz.de)