Yet another LaTeX filter

emacs, filter, html-report, languagetool, latex, parser, python-3, vim
pip install yalafi==1.3.0


YaLafi: Yet another LaTeX filter

Installation | Example application | Interface to Vim | Interface to Emacs | Filter actions | Principal limitations | Usage under Windows | Inclusion of own macros | Package interface | Handling of displayed equations | Command-line of pure filter | Differences to Tex2txt | Remarks on implementation

This Python package extracts plain text from LaTeX documents. The software may be integrated with a proofreading tool and an editor. It provides

  • mapping of character positions between LaTeX and plain text,
  • simple inclusion of own LaTeX macros and environments with tailored treatment,
  • careful conservation of text flows,
  • replacement of equations with detection of trailing interpunction.

The sample Python application script yalafi/shell/ from section Example application integrates the filter with LanguageTool. It

  • sends the extracted plain text to the proofreader,
  • maps position information in returned messages back to the LaTeX text,
  • generates results in several formats.

You may

  • create a proofreading report in text or HTML format for a complete document tree,
  • check LaTeX texts directly in the editors Emacs and Vim via plug-ins Emacs-langtool and vim-grammarous,
  • run the script as emulation of a LanguageTool server with integrated LaTeX filtering.

For instance, the LaTeX input

Only few people\footnote{We use
\textcolor{red}{redx colour.}}
is lazy.

will lead to the text report

1.) Line 2, column 17, Rule ID: MORFOLOGIK_RULE_EN_GB
Message: Possible spelling mistake found
Suggestion: red; Rex; reds; redo; Red; Rede; redox; red x
Only few people is lazy.    We use redx colour. 
2.) Line 3, column 1, Rule ID: PEOPLE_VBZ[1]
Message: If 'people' is plural here, don't use the third-person singular verb.
Suggestion: am; are; aren
Only few people is lazy.    We use redx colour. 

This is the corresponding HTML report:

HTML report

In some sense, this project relates to software like OpenDetex, pandoc, plasTeX, pylatexenc, TeXtidote, and tex2txt.

The tool builds on results from Tex2txt, but differs in the internal processing method. Instead of using recursive regular expressions, a simple tokeniser and a small machinery for macro expansion are implemented; see sections Differences to Tex2txt and Remarks on implementation.

Beside the interface from section Package interface, application Python scripts like yalafi/shell/ from section Example application can access an interface emulating from repository Tex2txt by from yalafi import tex2txt.

The pure LaTeX filter can be directly used in scripts via a command-line interface, it is described in section Command-line of pure filter.

If you use this software and encounter a bug or have other suggestions for improvement, please leave a note under category Issues, or initiate a pull request. Many thanks in advance.

Happy TeXing!

Back to top


Choose one of the following possibilities.

  • Use python -m pip install [--user] yalafi. This installs the last version uploaded to PyPI. Module pip itself can be installed with python -m ensurepip.
  • Place yalafi/ or a link to it in the current directory.
  • Place yalafi/ in a standard directory like /usr/lib/python3.?/ or ~/.local/lib/python3.?/site-packages/.
  • Place yalafi/ somewhere else and set environment variable PYTHONPATH accordingly.

Back to top

Example application

Remark. You can find examples for tool integration with Bash scripts in Tex2txt/

Example Python script yalafi/shell/ has been copied with minor changes from repository Tex2txt and subdivided into several files. It will generate a proofreading report in text or HTML format from filtering the LaTeX input and application of LanguageTool (LT). It is best called as module as shown below, but can also be placed elsewhere and invoked as script. On option '--server lt', LT's Web server is contacted. Otherwise, Java has to be present, and the path to LT has to be specified with --lt-directory. Note that from version 4.8, LT does not fully support 32-bit systems any more. Both LT and the script will print some progress messages to stderr. They can be suppressed with python ... 2>/dev/null.

python -m
                [--lt-directory dir] [--as-server port]
                [--output mode] [--link] [--context number]
                [--include] [--skip regex] [--plain-input]
                [--list-unknown] [--language lang] [--encoding ienc]
                [--replace file] [--define file] [--python-defs module]
                [--extract macros] [--disable rules] [--lt-options opts]
                [--single-letters accept] [--equation-punctuation mode]
                [--server mode] [--lt-server-options opts]
                [--textgears apikey] [--no-config]
                latex_file [latex_file ...] [> text_or_html_file]

Option names may be abbreviated. If present, options are also read from a configuration file designated by script variable 'config_file' (one option per line, possibly with argument), unless --no-config is given. Default option values are set at the Python script beginning.

  • --lt-directory dir
    The directory of the local LT installation, may be omitted on option '--server lt' or if script variable 'ltdirectory' is set appropriately. For instance, the directory has to contain file 'languagetool-server.jar'. The LT zip archive, for example, can be obtained from the LT download page. See also the script comment at variable 'ltdirectory'.
  • --as-server port
    Emulate an LT server listening on the given port, for an example see section Interface to Emacs. The fields of received HTML requests (settings for language, rules, categories) overwrite option values given in the command line. The internally used proofreader is influenced by options like --server. Other options like --single-letters remain effective.
  • --output mode
    Mode is one of 'plain', 'html', 'xml', 'json' (default: 'plain' for text report). Variant 'html' generates an HTML report, see below for further details. Mode 'xml' is intended for a Vim plug-in, compare section Interface to Vim.
  • --link
    In an HTML report, left-click on a highlighted text part opens a Web link related to the problem, if provided by LT.
  • --context number
    Number of context lines displayed around each marked text region in HTML report (default: 2). A negative number shows the whole text.
  • --include
    Track file inclusions like \input{...}. Script variable 'inclusion_macros' contains a list of the corresponding LaTeX macro names.
  • --skip regex
    Skip files matching the given regular expression. This is useful, e.g., for the exclusion of figures on option --include.
  • --plain-input
    Assume plain-text input, do not evaluate LaTeX syntax. This cannot be used together with options --include or --replace.
  • --list-unknown
    Only print list of unknown macros and environments seen outside of maths parts.
  • --language lang
    Language code as expected by LT (default: 'en-GB'). The first two letters are passed to function 'tex2txt()' from module yalafi.tex2txt that uses 'en' in case of an unknown language.
  • --encoding ienc
    Encoding for LaTeX input and files from options --define and --replace (default: UTF-8).
  • --replace file
    File with phrase replacements to be performed after the conversion to plain text. Per line, a '&' sign separated by space splits two parts: the first part is replaced by the second part. Space in the first part is interpreted as arbitrary space not breaking the paragraph. A '#' sign marks the rest of line as comment.
  • --define file
    Read macro definitions as LaTeX code (using \newcommand).
  • --python-defs module
    Modify default definitions in file yalafi/ by function 'modify_parameters()' in the given module. For an example, compare file
  • --extract macros
    Only check arguments of the LaTeX macros whose names are given as comma-separated list. This is useful for check of foreign-language text, if marked accordingly. Internally used for detection of file inclusions on --include.
  • --disable rules
    Comma-separated list of ignored LT rules, is passed as --disable to LT (default: 'WHITESPACE_RULE').
  • --lt-options opts
    Pass additional options to LT, given as single string in argument 'opts'. The first character of 'opts' will be skipped and must not be '-'. Example: --lt-options '~--languagemodel ../Ngrams --disablecategories PUNCTUATION'. Some options are included into HTML requests to an LT server, see script variable 'lt_option_map'.
  • --single-letters accept
    Check for single letters, accepting those in the patterns given as list separated by '|'. Example: --single-letters 'A|a|I|e.g.|i.e.||' for an English text, where the trailing '||' causes the addition of equation replacements from script variable 'equation_replacements'. All characters except '|' are taken verbatim, but '~' and '\,' are interpreted as UTF-8 non-breaking space and narrow non-breaking space.
  • --equation-punctuation mode
    This is an experimental hack for the check of punctuation after equations in English texts, compare section Equation replacements in English documents. An example is given in section Differences to Tex2txt. The abbreviatable mode values indicate the checked equation type: 'displayed', 'inline', 'all'.
    The check generates a message, if an element of an equation is not terminated by a dot '.', and at the same time is not followed by a lower-case word or another equation element, both possibly separated by a punctuation mark from ',;:'. Patterns for equation elements are given by script variables 'equation_replacements_display' and 'equation_replacements_inline', corresponding to member variables 'Parameters.math_repl_display' and 'Parameters.math_repl_inline' in file yalafi/
  • --server mode
    Use LT's Web server (mode is 'lt') or a local LT server (mode is 'my') instead of LT's command-line tool. Stop the local server (mode is 'stop', currently only works under Linux and Cygwin).
    • LT's server: Server address is set in script variable 'ltserver'. For conditions and restrictions, please refer to
    • Local server: If not yet running, then start it according to script variable 'ltserver_local_cmd'. Additional server options can be passed with --lt-server-options. See also This may be faster than the command-line tool used otherwise, especially for a large number of LaTeX files. The server will not be stopped at the end (use '--server stop').
  • --lt-server-options opts
    Pass additional options when starting a local LT server. Syntax is as for --lt-options.
  • --textgears apikey
    Use the TextGears server, see Language is fixed to American English. The access key 'apikey' can be obtained on page, but the key 'DEMO_KEY' seems to work for short input. The server address is given by script variable 'textgears_server'.
  • --no-config
    Do not read config file, whose name is set in script variable 'config_file'.

Dictionary adaptation. LT evaluates the two files 'spelling.txt' and 'prohibit.txt' in directory


Additional words and words that shall raise an error can be appended here. LT version 4.8 introduced additional files 'spelling_custom.txt' and 'prohibit_custom.txt'.

HTML report. The idea of an HTML report goes back to Sylvain Hallé, who developed TeXtidote. Opened in a Web browser, the report displays excerpts from the original LaTeX text, highlighting the problems indicated by LT. The corresponding LT messages can be viewed when hovering the mouse over these marked places, see the introductory example above. With option --link, Web links provided by LT can be directly opened with left-click. Script option --context controls the number of lines displayed around each tagged region; a negative option value will show the complete LaTeX input text. If the localisation of a problem is unsure, highlighting will use yellow instead of orange colour. For simplicity, marked text regions that intertwine with other ones are separately repeated at the end. In case of multiple input files, the HTML report starts with an index.

Back to top

Interface to Vim

For the Vim plug-in [vim-grammarous], it is possible to provide an interface for checking LaTeX texts. With an entry in ~/.vimrc, one may simply replace the command that invokes LanguageTool. For instance, you can add to your ~/.vimrc

let g:grammarous#languagetool_cmd = '/home/foo/bin/yalafi-grammarous'
map <F9> :GrammarousCheck --lang=en-GB<CR>

A proposal for Bash script /home/foo/bin/yalafi-grammarous (replace foo with user name ;-) is given in yalafi-grammarous. It has to be made executable with chmod +x .... Please adapt script variable ltdir, compare option --lt-directory in section Example application. If you do not want to have started a local LT server, comment out the line defining script variable use_server.

Troubleshooting for Vim interface. If Vim reports a problem with running LT, you can do the following. In ~/bin/yalafi-grammarous, comment out the final ... 2>/dev/null. For instance, you can just place a '#' in front: ... # 2>/dev/null. Then start, with a test file t.tex,

$ ~/bin/yalafi-grammarous t.tex

This should display some error message, if the problem goes back to running the script, Python, or LanguageTool.

Installation of vim-grammarous. Download and unzip vim-grammarous. Create a directory ~/.vim/pack/bundle/start/. Place vim-grammarous/ under this directory.

Here is the introductory example from above:

Vim plug-in

Back to top

Interface to Emacs

The Emacs plug-in [Emacs-langtool] may be used in two variants. First, you can add to ~/.emacs

(setq langtool-bin "/home/foo/bin/yalafi-emacs")
(setq langtool-default-language "en-GB")
(setq langtool-disabled-rules "WHITESPACE_RULE")
(require 'langtool)

A proposal for Bash script /home/foo/bin/yalafi-emacs (replace foo with user name ;-) is given in yalafi-emacs. It has to be made executable with chmod +x .... Please adapt script variable ltdir, compare option --lt-directory in section Example application. If you do not want to have started a local LT server, comment out the line defining script variable use_server.

Server interface. This variant may result in better tracking of character positions. In order to use it, you can write in ~/.emacs

(setq langtool-http-server-host "localhost"
      langtool-http-server-port 8082)
(setq langtool-default-language "en-GB")
(setq langtool-disabled-rules "WHITESPACE_RULE")
(require 'langtool)

and start as server in another terminal with

$ python -m --as-server 8082 [--lt-directory /path/to/LT]

The server will print some progress messages and can be stopped with CTRL-C. Further script arguments from section Example application may be given. If you add, for instance, '--server my', then a local LT server will be used. It is started on the first HTML request received from Emacs-langtool, if it is not yet running.

Troubleshooting for Emacs interface. If Emacs reports a problem with running LT, you can apply the steps from [Troubleshooting for Vim interface] to ~/bin/yalafi-emacs.

Installation of Emacs-langtool. Download and unzip Emacs-langtool. Place file langtool.el in directory ~/.emacs.d/lisp/. Set in your ~/.profile or ~/.bash_profile (and log in again)

export EMACSLOADPATH=~/.emacs.d/lisp:

Here is the introductory example from above:

Emacs plug-in

Back to top

Filter actions

Here is a list of the most important filter operations. When the filter encounters a LaTeX problem like a missing end of equation, a message is printed to stderr. Additionally, the message is included into the filter output together with the mark from 'Parameters.mark_latex_error' in yalafi/ This mark should raise a spelling error from the proofreader at the place where the problem was detected.

  • A collection of standard LaTeX macros and environments is already included, but very probably it has to be complemented.
  • Macro definitions with \(re)newcommand in the input text are processed. Statement \LTmacros{file.tex} reads macro definitions from the given file. Further own macros with arbitrary arguments can be defined on Python level, see section Inclusion of own macros.
  • Unknown macros are silently ignored, keeping their arguments with enclosing {} braces removed. They can be listed with options --unkn and --list-unknown for yalafi and, respectively.
  • Environment frames \begin{...} and \end{...} are deleted. We implement tailored behaviour for environment types listed in 'Parameters.environment_defs' in file yalafi/, see section Inclusion of own macros. For instance, environment bodies can be removed or replaced by fixed text.
  • Text in heading macros as \section{...} is extracted with added interpunction, see variable 'Parameters.heading_punct' in file yalafi/ This suppresses false positives from LanguageTool.
  • For macros as \ref, \eqref, \pageref, and \cite, suitable placeholders are inserted.
  • Arguments of macros like \footnote are appended to the main text, separated by blank lines. This preserves text flows.
  • Inline maths material $...$ and \(...\) is replaced with text from the rotating collection in 'Parameters.math_repl_inline' in file yalafi/ Trailing interpunction from 'Parameters.math_punctuation' is appended.
  • Equation environments are resolved in a way suitable for check of interpunction and spacing. The argument of \text{...} is included into the output text. Versions \[...\] and $$...$$ are handled like environment displaymath. See also sections Handling of displayed equations and Parser for maths material.
  • We generate numbered default \item labels for environment enumerate.
  • For \item with specified [...] label, some treatment is provided. If the text before ends with a punctuation mark from collection 'Parameters.item_punctuation' in file yalafi/, then this mark is appended to the label. This works well for German texts, it is turned off with the setting 'item_punctuation = []'.
  • Letters with text-mode accents as '\`' or '\v' are translated to the corresponding UTF-8 characters.
  • Things like double quotes '``' and dashes '--' are replaced with the corresponding UTF-8 characters. Additionally, we replace '~' and '\,' by UTF-8 non-breaking space and narrow non-breaking space.
  • For language 'de', suitable replacements for macros like '"`' and '"=' are inserted, see method 'Parameters.init_language()' in file yalafi/
  • Macro \verb and environment verbatim are processed. Environment verbatim can be replaced or removed like other environments with an appropriate entry in 'Parameters.environment_defs' in yalafi/
  • Rare warnings from the proofreading program can be suppressed using \LTadd{}, \LTskip{}, \LTalter{}{} in the LaTeX text. Suitable macro definition there, e.g., for adding something that only the proofreader should see: \newcommand{\LTadd}[1]{}

Back to top

Principal limitations

The implemented parsing mechanism can only roughly approximate the behaviour of a real LaTeX system. We assume that only “reasonable” macros are used, lower-level TeX operations are not supported. If necessary, they should be placed in a LaTeX file “hidden” for the filter (compare option --skip of in section Example application). A list of remaining incompatibilities must contain at least the following points.

  • Mathematical material is represented by simple replacements. As the main goal is application of a proofreading software, we have deliberately taken this approach.
  • Parsing does not cross file boundaries. Tracking of file inclusions is possible though.
  • Macros depending on (spacing) lengths may be treated incorrectly.

Back to top

Usage under Windows

Both and yalafi can be directly used in a Windows command script or console. For example, this could look like

py -3 -m --server lt --output html t.tex > t.html


"c:\Program Files\Python\Python37\python.exe" -m --server lt --output html t.tex > t.html

if the Python launcher has not been installed.

Files with Windows-style line endings (CRLF) are accepted, but the text output of the pure LaTeX filter will be Unix style (LF only), unless a Windows Python interpreter is used.

Python's version for Windows by default prints Latin-1 encoded text to standard output. As this ensures proper work in a Windows command console, we do not change it for when generating a text report. All other output is fixed to UTF-8 encoding.

Back to top

Inclusion of own macros

Unknown macros and environment frames are silently ignored. As all input files are processed independently, it may be necessary to provide project-specific definitions in advance.

For macros, which may be declared with \newcommand, you can apply \LTmacros{file.tex} as a simple solution. This adds the macros defined in the given file, skipping all other content. For the “real” LaTeX, the macro \LTmacros has to be defined as \newcommand{\LTmacros}[1]{} that is in turn ignored by the filter.

If LaTeX files have to stay untouched, you can use options --defs and --define for yalafi and, respectively. Alternatively, one can add the definitions to member 'Parameters.macro_defs_latex' in file yalafi/ Here is a short excerpt from this file:

        self.macro_defs_latex = r"""

More complicated macros as well as environments have to be registered with Python code. This may be done with options --pyth and --python-defs for yalafi and, respectively; see the example in Alternatively, you can modify the collections 'Parameters.macro_defs_python' and 'Parameters.environment_defs' in yalafi/

Definition of macros

Macro(parms, name, args='', repl='', defaults=[], extract='')

  • parms: current object of type Parameters
  • name: macro name with '\'
  • args: string that codes the argument sequence
    • 'A': mandatory argument, may be a single token or a sequence enclosed in {} braces
    • 'O': optional argument in [] brackets
    • '*' optional asterisk
  • repl: replacement string as for \newcommand ('*' does count as argument), or a function (see file yalafi/ for examples)
  • defaults: an optional list of replacement strings for absent optional arguments
  • extract: like repl, but the resulting text is appended to the main text, separated by blank lines; for an example, see declaration of macro \footnote in 'Parameters.macro_defs_python' in yalafi/

Definition of environments

Environ(parms, name, args='', repl='', defaults=[], remove=False, add_pars=True, items=None)

Argument parms to defaults are the same as for Macro(), where the arguments are those behind the opening '\begin{xyz}'. This means that the environment name 'xyz' does not yet count as argument in args and repl.

  • remove: if True, then the complete environment body is skipped; a fixed replacement can be given in repl
  • add_pars: if True, then paragraph breaks (blank lines) are generated before and behind the environment body
  • items: for inclusion of specific \item labels; a generator taking a nesting level argument has to be specified; compare declaration of environment enumerate in yalafi/

Definition of equation environments

EquEnv(parms, name, args='', repl='', defaults=[], remove=False)

This is equivalent to Environ(), but maths material is replaced according to section Handling of displayed equations. Replacements in repl and defaults are still interpreted in text mode.

  • remove: if True, then a fixed replacement can be specified in repl, and trailing interpunction given by 'Parameters.math_punctuation' in file yalafi/ is appended

Back to top

Package interface

We comment the central function in file yalafi/ that uses the package interface to emulate the behaviour of script in repository Tex2txt.

 1  from . import parameters, parser, utils
 2  def tex2txt(latex, opts):
 3      def read(file):
 4          try:
 5              with open(file, encoding=opts.ienc) as f:
 6                  return True,
 7          except:
 8              return False, ''
 9      parms = parameters.Parameters(opts.lang)
10      if opts.defs:
11          parms.add_latex_macros(opts.defs)
12      if opts.pyth:
13          exec('import ' + opts.pyth)
14          exec(opts.pyth + '.modify_parameters(parms)')
15      if opts.extr:
16          extr = ['\\' + s for s in opts.extr.split(',')]
17      else:
18          extr = []
19      p = parser.Parser(parms, read_macros=read)
20      toks = p.parse(latex, extract=extr)
21      txt, pos = utils.get_txt_pos(toks)
22      if opts.repl:
23          txt, pos = utils.replace_phrases(txt, pos, opts.repl)
24      if opts.unkn:
25          txt = '\n'.join(p.get_unknowns()) + '\n'
26          pos = [0 for n in range(len(txt))]
27      pos = [n + 1 for n in pos]
28      return txt, pos
  • 3-8: This is an auxiliary function for the parser.
  • 9: The created parameter object contains all default settings and definitions from file yalafi/
  • 11: If requested by script option --defs, additional macros are included from the string opts.defs.
  • 14: On option --pyth, we call a function to modify the parameter object, see file for an example.
  • 15-18: If option --extr requests only extraction of arguments of certain macros, this is prepared.
  • 19: We create a parser object, the passed function is called on \LTmacros.
  • 20: The parsing method returns a list of tokens.
  • 21: The token list is converted into a 2-tuple containing the plain-text string and a list of numbers. Each number in the list indicates the estimated position of the corresponding character in the text string.
  • 23: If phrase replacements are requested by option --repl, this is done. String opts.repl contains the replacement specifications read from the file.
  • 25: On option --unkn, a list of unknown macros and environments is generated.
  • 27: This is necessary, since position numbers are zero-based in yalafi, but one-based in Tex2txt/

Back to top

Handling of displayed equations

Displayed equations should be part of the text flow and include the necessary interpunction. The German version of LanguageTool (LT) will detect a missing dot in the following snippet. For English texts, see the comments in section Equation replacements in English documents ahead.

Wir folgern
    a   &= b \\
    c   &= d
Daher ...

Here, 'a' to 'd' stand for arbitrary mathematical terms (meaning: “We conclude <maths> Therefore, ...”). In fact, LT complains about the capital “Daher” that should start a new sentence.

Trivial version

With the entry

    Environ(self, 'align', remove=True, add_pars=False),

in 'Parameters.environment_defs' of file yalafi/, the equation environment is simply removed. We get the following filter output that will probably cause a problem, even if the equation itself ends with a correct interpunction sign.

Wir folgern
Daher ...

Simple version

With the entry

    EquEnv(self, 'align', repl='  Relation', remove=True),

in 'Parameters.environment_defs', one gets:

Wir folgern
Daher ...

Adding a dot '= d.' in the equation will lead to 'Relation.' in the output. This will also hold true, if the interpunction sign ('Parameters.math_punctuation') is followed by maths space or by macros as \label and \nonumber.

Full version

With the default entry

    EquEnv(self, 'align'),

we obtain (“gleich” means equal, and setting language to English will produce “equal”):

Wir folgern
  V-V-V  gleich W-W-W
  W-W-W  gleich X-X-X.
Daher ...

The replacements like 'V-V-V' are taken from collection 'Parameters.math_repl_display' that depends on language setting, too. Now, LT will additionally complain about repetition of 'W-W-W'. Finally, writing '= b,' and '= d.' in the equation leads to the output:

Wir folgern
  V-V-V  gleich W-W-W,
  X-X-X  gleich Y-Y-Y.
Daher ...

The rules for equation parsing are described in section Parser for maths material. They ensure that variations like

    a   &= b \\
        &= c.


    a   &= b \\
        &\qquad -c.

also will work properly. In contrast, the text

    a   &= b \\
    -c  &= d.

will again produce an LT warning due to the missing comma after 'b', since the filter replaces both 'b' and '-c' by 'W-W-W' without intermediate text.

In rare cases, manipulation with \LTadd{} or \LTskip{} may be necessary to avoid false warnings from the proofreader.

Inclusion of “normal” text

In variant “Full version”, the argument of \text{...} (macro names: collection 'Parameters.math_text_macros') is directly copied. Outside of \text, only maths space like \; and \quad (see 'Parameters.math_space') is considered as space. Therefore, one will get warnings from the proofreading program, if subsequent \text and maths parts are not properly separated.

Equation replacements in English documents

The replacement collection of 'Parameters.math_repl_display' in file yalafi/ does not work well, if single letters are taken as replacements. For instance, 'V.' cannot be safely considered as end of a sentence. We now have chosen replacements as 'U-U-U' for German and English texts.

Furthermore, the English version of LanguageTool (like other proofreading tools) rarely detects mistakenly capital words inside of a sentence; they are probably considered as proper names. Therefore, a missing dot at the end of a displayed equation is hardly found. An experimental hack is provided by option --equation-punctuation of application script yalafi/shell/ described in section Example application.

Back to top

Command-line of pure filter

The LaTeX filter can be integrated in scripts, compare the examples in Tex2txt/

python -m yalafi [--nums file] [--repl file] [--defs file] [--pyth module]
                 [--extr macros] [--lang xy] [--ienc enc] [--unkn]

Without positional argument latexfile, standard input is read.

  • --nums file
    File for storing estimated original position numbers for each character of plain text. This can be used later to correct position figures in proofreader messages.
  • --repl file
    As option --replace in section Example application.
  • --defs file
    As option --define in section Example application.
  • --pyth module
    As option --python-defs in section Example application
  • --extr ma[,mb,...]
    As option --extract in section Example application.
  • --lang xy
    Language 'de' or 'en' (default: 'en', also taken in case of unknown language). Is used for adaptation of equation replacements, maths operator names, proof titles, and for handling of macros like '"='.
  • --ienc enc
    As option --encoding in section Example application.
  • --unkn
    As option --list-unknown in section Example application.

Back to top

Differences to Tex2txt

Invocation of python -m yalafi ... differs as follows from python ... (the script described in Tex2txt/

  • Macro definitions with \(re)newcommand in the LaTeX input are processed.
  • Macro arguments need not be delimited by {} braces or [] brackets.
  • Macros are expanded in the order they appear in the text.
  • Character position tracking for displayed equations is improved, see the example below.
  • Parameters like predefined LaTeX macros and environments are set in file yalafi/ You can modify them at run-time with script option '--pyth module'. The given Python module has to provide a function 'modify_parameters(parms)' receiving the parameter object 'parms', compare the example in
  • Option --defs expects a file containing macro definitions as LaTeX code.
  • Option --ienc is also effective for file from --defs.
  • Option --char (position tracking for single characters) is always activated.
  • Default language is English. It is also used for an unknown language.

YaLafi/yalafi/ is faster for input texts till about 30 Kilobytes, for larger files it can be slower than 'Tex2txt/ --char'. Run-time increases quasi linearly with file size. Due to token generation for each single “normal” character, memory usage may be substantial for long input texts.

Number of effective code lines (without blank and pure comment lines) is around 1050 for Tex2txt/ and 1350 for yalafi/*.py in total.


python -m --equation-punct all --output html test.tex > test.html

and input

For each $\epsilon > 0$, there is a $\delta > 0$ so that
\norm{y-x} < \delta \text{\quad implies\quad}
    \norm{A(y) - A(x)} < \epsilon, \label{lab}
Therefore, operator $A$ is continuous at point $x$.

we get

HTML report

Back to top

Remarks on implementation

Scanner / tokeniser

The scanner identifies token types defined in yalafi/

  • All “normal” characters yield an own token.
  • Many character combinations like '{', '\[' or '---' are recognised as “special tokens”.
  • Names of “normal” macros formed by a backslash and subsequent letters (method 'Parameters.macro_character()') result in a token, macros '\begin', '\end', '\item', and '\verb' are treated separately.
  • For space, we distinguish between character sequences that do or do not represent a paragraph break. In both cases, a single token is generated.
  • Comments starting with '%' consume the rest of the line and leading space on the next line, if it is not blank. A single token is generated.


The central method 'Parser.expand_sequence()' does not directly read from the scanner, but from an intermediate buffer that can take back tokens. On macro expansion, the parser simply pushes back all tokens generated by argument substitution. (Method 'Parser.expand_arguments()' collects tokens forming macro arguments and returns a list of replacement tokens that is eventually pushed back in the main loop.) The result is close to the “real” TeX behaviour, compare the tests in directory tests/.

A method important for simple implementation is 'Parser.arg_buffer()'. It creates a new buffer that subsequently returns tokens forming a macro argument (only a single token or all tokens enclosed in paired {} braces or [] brackets).

Parser for maths material

We follow the ideas described in section Handling of displayed equations, compare the tests in tests/ All unknown macros, which are not in the blacklist 'Parameters.math_ignore', are assumed to generate some “visible” output. Thus, it is not necessary to declare all the maths macros like \alpha and \sum.

Displayed equations are parsed as follows.

  • Equation environments are split into “lines” separated by '\\'.
  • Each “line” is split into “sections” delimited by '&'.
  • Each “section” is split into “maths parts” only consisting of maths material separated by intermediate \text{...} or \mbox{...} ('Parameters.math_text_macros').
  • Arguments of \text and \mbox are directly copied.
  • A “maths part” is substituted with a placeholder from rotating collection 'Parameters.math_repl_display', if it does not consist only of punctuation marks from 'Parameters.math_punctuation' or of operators from 'Parameters.math_operators'.
  • A leading maths operator is displayed using 'Parameters.math_op_text' (language-dependent), if the “maths part” is first in “section” and the “section” is not first on “line”.
  • Trailing interpunction of a “maths part” is appended to the placeholder.
  • If the “maths part” includes leading or trailing maths space from 'Parameters.math_space', then white space is prepended or appended to the replacement.
  • Replacements from 'Parameters.math_repl_display' are rotated
    • if a non-blank \text part is detected,
    • if a “maths part” starts with an operator and is first in “section”, but not on “line”
    • if a “maths part” only consists of an operator,
    • if a “maths part” includes trailing interpunction.

Back to top

Removal of unnecessary blank lines

In order to avoid creation of new blank lines by macros expanding to space or “nothing”, we include a token of type 'ActionToken' whenever expanding a macro. Method 'Parser.remove_pure_action_lines()' removes all lines only containing space and at least one such token. Initially empty lines are retained.

Back to top