Starting from 2021-07-01, all LRZ GitLab users will be required to explicitly accept the GitLab Terms of Service. Please see the detailed information at and make sure that your projects conform to the requirements. 17.8 KB
Newer Older
Eckhart Arnold's avatar
Eckhart Arnold committed
1 2 3 4 5 6 7
Introduction to [DHParser](

*This is just an appetizer. Full documentation coming soon...*

Motto: **Computers enjoy XML, humans don't.**


Eckhart Arnold's avatar
Eckhart Arnold committed
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54
Why use domain specific languages in the humanities?

Suppose you are a literary scientist and you would like to edit a poem
like Heinrich Heine's "Lyrisches Intermezzo". Usually, the technology 
of choice would be XML and you would use an XML-Editor to write to
code something like this:

    <?xml version="1.0" encoding="UTF-8" ?>
            <autor gnd="118548018">Heinrich Heine</autor>
            <werk href=""
                Buch der Lieder
            <ort gnd="4023118-5">Hamburg</ort>
            <serie>Lyrisches Intermezzo</serie>
                <vers>Wenn ich in deine Augen seh',</vers>
                <vers>so schwindet all' mein Leid und Weh!</vers>
                <vers>Doch wenn ich küsse deinen Mund,</vers>
                <vers>so werd' ich ganz und gar gesund.</vers>
                <vers>Wenn ich mich lehn' an deine Brust,</vers>
                <vers>kommt's über mich wie Himmelslust,</vers>
                <vers>doch wenn du sprichst: Ich liebe dich!</vers>
                <vers>so muß ich weinen bitterlich.</vers>

Now, while you might think that this all works well enough, there are
a few drawbacks to this approach:

- The syntax is cumbersome and the encoding not very legible to humans
  working with it. (And I did not even use 
  [TEI-XML](, yet...)
  Editing and revising XML-encoded text is a pain. Just ask the 
  literary scientists who have to work with it.
- The XML encoding, especially TEI-XML, is often unintuitive. Only
  experts understand it. Now, if you had the idea that your humanist
Eckhart Arnold's avatar
Eckhart Arnold committed
56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190
  friend, who is not into digital technologies, might help you with
  proof-reading, you better think about it again.
- There is an awful lot of typing to do: All those lengthy opening 
  and closing tags. This takes time...
- While looking for a good XML-Editor, you find that there hardly exist
  any XML-Editors any more. (And for a reason, actually...) In 
  particular, there are no good open source XML-Editors.

One the other hand, there are good reasons why XML is used in the
humanities: Important encoding standards like TEI-XML are defined in 
XML. It's strict syntax and the possibility to check data against a
schema  help detecting and avoiding encoding errors. If the schema 
is well defined, it is unambiguous, and it is easy to parse for a 
computer. Most of these advantages, however, are on a technical level
and few of them are actually exclusive advantages of XML.

All in all this means, that while XML is a solid backend-technology,
it still is a pain to work with XML as a frontend-technology. This is
where DHParser comes in. It allows you to define your own domain
specific notation that is specifically tailored to your editing needs
and provides an infrastructure that - if you know a little 
Python-programming - makes it very easy to convert your annotated
text into an XML-encoding of your choice. With DHParser, the same poem
above can be simply encoded like this:
    Heinrich Heine <gnd:118548018>,
    Buch der Lieder <urn:nbn:de:kobv:b4-200905192211>,
    Hamburg <gnd:4023118-5>, 1927.
        Lyrisches Intermezzo
    Wenn ich in deine Augen seh',
    so schwindet all' mein Leid und Weh!
    Doch wenn ich küsse deinen Mund,
    so werd' ich ganz und gar gesund.
    Wenn ich mich lehn' an deine Brust,
    kommt's über mich wie Himmelslust,
    doch wenn du sprichst: Ich liebe dich!
    so muß ich weinen bitterlich.

Yes, that's right. It is as simple as that. Observe, how much 
more effacious a verse like "Wenn ich mich lehn' an deine Brust, / 
kommt's über mich wie Himmelslust," can be if it is not cluttered with
XML tags ;-) 

You might now wonder 
whether the second version really does encode the same information
as the XML version. How, for example, would the computer know for 
sure where a verse starts and ends or a stanza or what is 
title and what stanza? Well, for all these matters there exist 
conventions that poets have been using for several thousand years.
For example, a verse always starts and ends in one an the same 
line. There is always a gap between stanzas. And the title is always 
written above the poem and not in the middle of it. So, if there is
a title at all, we can be sure that what is written in the first 
line is the title and not a stanza. 

DHParser is able to exploit all those hints in order to gather much the
same information as was encoded in the XML-Version. Don't believe it?
You can try: Download DHParser from the 
[gitlab-repository]( and enter
the directory `examples/Tutorial` on the command line interface (shell). 
Just run `python` (you need to have installed
[Python]( Version 3.4 or higher on your computer).
The output will be something like this:

                <namenfolge>Heinrich Heine</namenfolge>
                <wortfolge>Buch der Lieder</wortfolge>
        <serie>Lyrisches Intermezzo</serie>
                <vers>Wenn ich in deine Augen seh',</vers>
                <vers>so schwindet all' mein Leid und Weh!</vers>
                <vers>Doch wenn ich küsse deinen Mund,</vers>
                <vers>so werd' ich ganz und gar gesund.</vers>
                <vers>Wenn ich mich lehn' an deine Brust,</vers>
                <vers>kommt's über mich wie Himmelslust,</vers>
                <vers>doch wenn du sprichst: Ich liebe dich!</vers>
                <vers>so muß ich weinen bitterlich.</vers>

Now, you might notice that this is not exactly the XML-encoding as shown
above. (Can you spot the differences?) But you will probably believe me
without further proof that it can easily be converted into the other
version and contains all the information that the other version contains.

How does DHParser achieve this? Well, there is the rub. In order to convert
the poem in the domain specific version into the XML-version, DHParser 
requires a structural description of the domain specific encoding. This
is a bit similar to a document type definition (DTD) in XML. This 
structural description uses a slightly enhanced version of the 
[Extended-Backus-Naur-Form (EBNF)]( 
that is a well established formalism for the structural description of 
formal languages in computer sciences. And excerpt of the EBNF-definition
of our domain-specific encoding for the poem looks like this. (We leave out
the meta-data here. See 
for the full EBNF):

    gedicht           = { LEERZEILE }+ [serie] §titel §text /\s*/ §ENDE
    serie             = !(titel vers NZ vers) { NZ zeile }+ { LEERZEILE }+   
    titel             = { NZ zeile}+ { LEERZEILE }+
    zeile             = { ZEICHENFOLGE }+
    text              = { strophe {LEERZEILE} }+
    strophe           = { NZ vers }+
    vers              = { ZEICHENFOLGE }+
    ZEICHENFOLGE      = /[^ \n<>]+/~
    NZ                = /\n/~
    LEERZEILE         = /\n[ \t]*(?=\n)/~
    ENDE              = !/./

Without going into too much detail here, let me just explain a few basics of 
Eckhart Arnold's avatar
Eckhart Arnold committed
192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345
this formal description: The slashes `/` enclose ordinary regular expressions.
Thus, `NZ` for ("Neue Zeile", German for: "new line") is defined as `/\n/~` which
is the newline-token `\n` in a regular expression, plus further horizontal 
whitespace (signified by the tilde `~`), if there is any.

The braces `{` `}` enclose items that can be repeated zero or more times; with
a `+` appended to the closing brace it means one or more times. Now, look at the
definition of `text` in the 6th line: `{ strophe {LEERZEILE} }+`. This reads
as follows: The text of the poem consists of a sequence of stanzas, each of which
is followed by a sequence of empty lines (German: "Leerzeilen"). If you now
look a the structural definition of a stanza, you find that it consists of a 
sequence of verses, each of which starts, i.e. is preceeded by a new line.

Can you figure out the rest? Hint: The angular brackets `[` and `]` mean that and
item is optional and the `§` sign means that it is obligatory. (Strictly speaking, 
the §-signs are not necessary, because an item that is not optional is always
obligatory, but the §-signs help the converter to produce the right error

This should be enough for an introduction to the purpose of DSLs in the humanities. 
It has shown the probably most important
use case of DHParser, i.e. as a frontend-technology form XML-encodings. Of course
it can just as well be used as a frontend for any other kind of structured data,
like SQL or graph-strcutured data. The latter is by the way is a very reasonable
alternative to XML for edition projects with a complex transmission history. 
See Andreas Kuczera's Blog-entry on 
["Graphdatenbanken für Historiker"](

Tutorial: First Steps with DHParser

Disclaimer: *You'll need to be able to use a shell and have some basic
knowledge of Python programming to be able to follow this section!*

In order to try the example above, you should fetch DHParsers from its

    $ git clone

Now, if the enter the repo, you'll find three subdirectories:

The directory `DHParser` contains the Python modules of the DHParser-package, `test` - as you can 
guess - contains the unit-tests for DHParser. Now, enter the `examples/Tutorial`-directory. 
Presently, most other examples are pretty rudimentary. So, don't worry about them.
In this directory, you'll find a simple EBNF Grammar for poetry in the file `Lyrik.ebnf`. Have a 
look at it. You'll find that is the same grammar (plus a few additions) that has been mentioned
just before. You'll also find a little script `` that is used to compile an
EBNF-Grammar into an executable Python-module that can be used to parse any piece of text that
this grammar is meant for; in this case poetry. 

Any DHParser-Project needs such a script. The content of the script is pretty self-explanatory:

    from DHParser.testing import recompile_grammar
    if not recompile_grammar('.', force=True):
        with open('Lyrik_ebnf_ERRORS.txt') as f:
The script simply (re-)compiles any EBNF grammar that it finds in the current directory. "Recompiling" 
means that DHParser notices if a grammar has already been compiled and overwrites only that part of
the generated file that contains the actual parser. All other parts - we will come to that later
what these are - can safely be edited by you. Now just run `` from the command line:

    $ python3
You'll find that `` has generated a new script with the name ``. This
script contains the Parser for the `Lyrik.ebnf`-grammar and some skeleton-code for a DSL->XML-Compiler 
(or rather, a DSL-whatever compiler), which you can later fill in. Now let's see how this script works:

    $ python3 Lyrisches_Intermezzo_IV.txt >result.xml
The file `Lyrisches_Intermezzo_IV.txt` contains the fourth part of Heinrich Heine's Lyrisches Intermezzo
encoded in our own human-readable poetry-DSL that has been shown above. 
Since we have redirected the output to `result.xml`, you'll find a new file with this name in the 
directory. If you look at it with an editor - preferably one that provides syntax-highlighting for
XML-files, you'll find that it look's pretty much like XML. However, this XML-code still looks 
much more obfuscated than in the Introduction before. If you look closely, you can nonetheless see
that the poem itself has faithfully been preserved. For example, if you scroll down a few lines,
you'll find the, hardly recognizable first verse of the poem:

            <:Whitespace> </:Whitespace>
            <:Whitespace> </:Whitespace>
            <:Whitespace> </:Whitespace>
            <:Whitespace> </:Whitespace>
            <:Whitespace> </:Whitespace>

How come it is so obfuscated, and where do all those pseudo-tags like `<:RegExp>` and
`<:Whitespace>` come from? Well, this is probably the right time to explain a bit about
parsing and compilation in general. Parsing and compilation of a text with DHParser
takes place in three strictly separated steps:

1. Parsing of the text and generation of the "concrete syntax tree" (CST)
2. Transformation of the CST into an "abstract syntax tree" (AST)
3. And, finally, compilation of the AST into valid XML, HTML, LaTeX or what you like.

DHParser automatically only generates a parser for the very first step. The other steps have
to be programmed by hand, though DHParser makes tries to make those parts as easy as possible.
What you have just seen in your editor is a Pseudo-XML-representation of the concrete syntax
tree. (The output of a parser always is a tree structure, just like XML.) 
It is called concrete syntax tree, because it contains all the syntactic details that
have been described in the `Lyrik.ebnf`-grammar; and the grammar needs to describe all those
details, because otherwise it would not be possible to parse the text. On the other hand
most of these details do not carry any important information. This is the reason why in the
second step the transformation into an abstract syntax tree that leaves out the unimportant
details. There is now general rule of how to derive abstract syntax trees from concrete syntax
trees, and there cannot be, because it depends on the particular domain of application which
details are important and which not. For poems these might be different than, say, for a
catalogue entry. Therefore the AST-transformation has to be specified for each grammar
separately, just as the grammar has to be specified for each application domain.

Before I'll explain how to specify an AST-transformation for DHParser, you may want to
know what difference it makes. There is a script `` in the directory
where the AST-transformations are already included. Running the script 

    $ python Lyrisches_Intermezzo_IV.txt

yields the fairly clean Pseudo-XML-representation of the DSL-encoded poem that we have seen above.
Just as a teaser, you might want to look up, how the AST-transformation is specified with DHParser.
For this purpose, you can have a look in file ``. If you scrool down to
the AST section, you'll see something like this:

    Lyrik_AST_transformation_table = {
        # AST Transformations for the Lyrik-grammar
        "+": remove_empty,
        "bibliographisches": [remove_parser('NZ'), remove_tokens],
        "autor, werk, untertitel, ort": [],
Eckhart Arnold's avatar
Eckhart Arnold committed
347 348 349 350 351 352 353 354
        "jahr": [reduce_single_child],
        "wortfolge": [flatten(has_name('WORT'), recursive=False), remove_last(is_whitespace), collapse],
        "namenfolge": [flatten(has_name('NAME'), recursive=False), remove_last(is_whitespace), collapse],
        "verknüpfung": [remove_tokens('<', '>'), reduce_single_child],
        "ziel": reduce_single_child,
        "gedicht, strophe, text": [flatten, remove_parser('LEERZEILE'), remove_parser('NZ')],
        "titel, serie": [flatten, remove_parser('LEERZEILE'), remove_parser('NZ'), collapse],
        "vers": collapse,
        "zeile": [],
Eckhart Arnold's avatar
Eckhart Arnold committed
        "ZEICHENFOLGE, NZ, JAHRESZAHL": reduce_single_child,
        "WORT, NAME, LEERZEILE, ENDE": [],
Eckhart Arnold's avatar
Eckhart Arnold committed
358 359 360 361 362 363 364 365 366 367 368 369 370 371
        ":Whitespace": replace_content(lambda node : " "),
        ":Token, :RE": reduce_single_child,
        "*": replace_by_single_child

As you can see, AST-transformations a specified in a declarative fashion (with the option to add
your own Python-programmed transformation rules). This keeps the specification of the AST-transformation
simple and concise. At the same, we avoid adding hints for the AST-transformation in the grammar 
specification, which would render the grammar less readable. 
Next, I am going to explain step by step, how a domain specific language for poems like Heine's 
Lyrisches Intermezzo can be designed, specified, compiled and tested.

*to be continued, stay tuned...*