Introduction.md 17.8 KB
Newer Older
Eckhart Arnold's avatar
Eckhart Arnold committed
1
2
3
4
5
6
7
Introduction to [DHParser](https://gitlab.lrz.de/badw-it/DHParser)
==================================================================

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

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

8

Eckhart Arnold's avatar
Eckhart Arnold committed
9
10
11
12
Why use domain specific languages in the humanities?
----------------------------------------------------

Suppose you are a literary scientist and you would like to edit a poem
13
like Heinrich Heine's "Lyrisches Intermezzo". Usually, the technology of
Eckhart Arnold's avatar
Eckhart Arnold committed
14
choice would be XML and you would use an XML-Editor to write
15
something like this:
Eckhart Arnold's avatar
Eckhart Arnold committed
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

    <?xml version="1.0" encoding="UTF-8" ?>
    <gedicht>
        <bibliographisches>
            <autor gnd="118548018">Heinrich Heine</autor>
            <werk href="http://www.deutschestextarchiv.de/book/show/heine_lieder_1827"
                  urn="nbn:de:kobv:b4-200905192211">
                Buch der Lieder
            </werk>
            <ort gnd="4023118-5">Hamburg</ort>
            <jahr>1927</jahr>
            <serie>Lyrisches Intermezzo</serie>
            <titel>IV.</titel>
        </bibliographisches>
        <text>
            <strophe>
                <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>
            </strophe>
            <strophe>
                <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>
            </strophe>
        </text>
    </gedicht>

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](http://www.tei-c.org/index.xml), yet...)
  Editing and revising XML-encoded text is a pain. Just ask the 
  literary scientists who have to work with it.
54
55

- The XML encoding, especially TEI-XML, is often not intuitive. Only
56
  experts understand it. Now, if you had the idea that your humanist
Eckhart Arnold's avatar
Eckhart Arnold committed
57
58
  friend, who is not into digital technologies, might help you with
  proof-reading, you better think about it again.
59

Eckhart Arnold's avatar
Eckhart Arnold committed
60
61
- There is an awful lot of typing to do: All those lengthy opening 
  and closing tags. This takes time...
62
  
Eckhart Arnold's avatar
Eckhart Arnold committed
63
64
65
66
- 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.

67
On the other hand, there are good reasons why XML is used in the
di68kap's avatar
di68kap committed
68
69
70
humanities: Important encoding standards like 
[TEI-XML](http://www.tei-c.org/index.xml) are defined in
XML. Its strict syntax and the possibility to check data against
71
72
73
74
75
76
77
78
79
80
81
82
83
schema help to detect 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 back-end-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:
Eckhart Arnold's avatar
Eckhart Arnold committed
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
 
    Heinrich Heine <gnd:118548018>,
    Buch der Lieder <urn:nbn:de:kobv:b4-200905192211>,
    Hamburg <gnd:4023118-5>, 1927.
    
        Lyrisches Intermezzo
    
                 IV.
    
    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.

103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
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 on 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. 
Eckhart Arnold's avatar
Eckhart Arnold committed
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

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](https://gitlab.lrz.de/badw-it/DHParser) and enter
the directory `examples/Tutorial` on the command line interface (shell). 
Just run `python LyrikCompiler_example.py` (you need to have installed
[Python](https://www.python.org/) Version 3.4 or higher on your computer).
The output will be something like this:

    <gedicht>
        <bibliographisches>
            <autor>
                <namenfolge>Heinrich Heine</namenfolge>
                <verknüpfung>gnd:118548018</verknüpfung>
            </autor>
            <werk>
                <wortfolge>Buch der Lieder</wortfolge>
                <verknüpfung>urn:nbn:de:kobv:b4-200905192211</verknüpfung>
            </werk>
            <ort>
                <wortfolge>Hamburg</wortfolge>
                <verknüpfung>gnd:4023118-5</verknüpfung>
            </ort>
            <jahr>1927</jahr>
        </bibliographisches>
        <serie>Lyrisches Intermezzo</serie>
        <titel>IV.</titel>
        <text>
            <strophe>
                <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>
            </strophe>
            <strophe>
                <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>
            </strophe>
        </text>
    </gedicht>

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 
Eckhart Arnold's avatar
Eckhart Arnold committed
172
173
174
[Extended-Backus-Naur-Form (EBNF)](https://en.wikipedia.org/wiki/Extended_Backus%E2%80%93Naur_form),
which is a well-established formalism for the structural description of 
formal languages in computer sciences. An excerpt of the EBNF-definition
Eckhart Arnold's avatar
Eckhart Arnold committed
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
of our domain-specific encoding for the poem looks like this. (We leave out
the meta-data here. See 
[`examples/Tutorial/Lyrik.ebnf`](https://gitlab.lrz.de/badw-it/DHParser/blob/master/examples/Tutorial/Lyrik.ebnf)
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              = !/./

194
Without going into too much detail here, let me just explain a few basics of 
Eckhart Arnold's avatar
Eckhart Arnold committed
195
196
197
198
199
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.

200
201
202
203
204
205
206
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
at 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.
Eckhart Arnold's avatar
Eckhart Arnold committed
207
208
209
210

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
Eckhart Arnold's avatar
Eckhart Arnold committed
211
obligatory, but the §-signs help the converter to produce more useful error
Eckhart Arnold's avatar
Eckhart Arnold committed
212
213
messages.)

214
215
216
217
218
219
220
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
Eckhart Arnold's avatar
Eckhart Arnold committed
221
222
223
224
225
226
227
228
229
230
231
232
233
234
["Graphdatenbanken für Historiker"](http://mittelalter.hypotheses.org/5995).


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-repository:

    $ git clone git@gitlab.lrz.de:badw-it/DHParser.git

235
Now, if you enter the repo, you'll find three subdirectories:
Eckhart Arnold's avatar
Eckhart Arnold committed
236
237
238
239
240

    DHParser
    examples
    test
    
241
242
243
244
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.
Eckhart Arnold's avatar
Eckhart Arnold committed
245
 
246
247
248
249
250
251
252
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 `recompile_grammar.py` 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. 
Eckhart Arnold's avatar
Eckhart Arnold committed
253

254
255
Any DHParser-Project needs such a script. The content of the script is
pretty self-explanatory:
Eckhart Arnold's avatar
Eckhart Arnold committed
256
257
258
259
260
261
262

    from DHParser.testing import recompile_grammar
    if not recompile_grammar('.', force=True):
        with open('Lyrik_ebnf_ERRORS.txt') as f:
            print(f.read())
        sys.exit(1)
        
263
264
265
266
267
268
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 `recompile_grammar.py` from the command line:
Eckhart Arnold's avatar
Eckhart Arnold committed
269
270
271

    $ python3 recompile_grammar.py
    
272
273
274
275
276
You'll find that `recompile_grammar.py` has generated a new script with
the name `LyrikCompiler.py`. 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:
Eckhart Arnold's avatar
Eckhart Arnold committed
277
278
279

    $ python3 LyrikCompiler.py Lyrisches_Intermezzo_IV.txt >result.xml
    
280
281
282
283
284
285
286
287
288
289
290
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:
Eckhart Arnold's avatar
Eckhart Arnold committed
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

    ...
    <vers>
        <ZEICHENFOLGE>
            <:RegExp>Wenn</:RegExp>
            <:Whitespace> </:Whitespace>
        </ZEICHENFOLGE>
        <ZEICHENFOLGE>
            <:RegExp>ich</:RegExp>
            <:Whitespace> </:Whitespace>
        </ZEICHENFOLGE>
        <ZEICHENFOLGE>
            <:RegExp>in</:RegExp>
            <:Whitespace> </:Whitespace>
        </ZEICHENFOLGE>
        <ZEICHENFOLGE>
            <:RegExp>deine</:RegExp>
            <:Whitespace> </:Whitespace>
        </ZEICHENFOLGE>
        <ZEICHENFOLGE>
            <:RegExp>Augen</:RegExp>
            <:Whitespace> </:Whitespace>
        </ZEICHENFOLGE>
        <ZEICHENFOLGE>
            <:RegExp>seh',</:RegExp>
        </ZEICHENFOLGE>
    </vers>
    ...

    
321
322
323
324
325
326
327
328
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)
Eckhart Arnold's avatar
Eckhart Arnold committed
329
330

2. Transformation of the CST into an "abstract syntax tree" (AST)
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357

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 is always 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 from, 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
`LyrikCompiler_example.py` in the directory where the
AST-transformations are already included. Running the script 
Eckhart Arnold's avatar
Eckhart Arnold committed
358
359
360

    $ python LyrikCompiler_example.py Lyrisches_Intermezzo_IV.txt

361
362
363
364
365
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 `LyrikCompiler_example.py`. If you
scrool down to the AST section, you'll see something like this:
Eckhart Arnold's avatar
Eckhart Arnold committed
366
367
368
369
370

    Lyrik_AST_transformation_table = {
        # AST Transformations for the Lyrik-grammar
        "+": remove_empty,
        "bibliographisches": [remove_parser('NZ'), remove_tokens],
371
        "autor, werk, untertitel, ort": [],
Eckhart Arnold's avatar
Eckhart Arnold committed
372
        "jahr": [reduce_single_child],
373
374
        "wortfolge": [flatten(is_one_of('WORT'), recursive=False), remove_last(is_whitespace), collapse],
        "namenfolge": [flatten(is_one_of('NAME'), recursive=False), remove_last(is_whitespace), collapse],
Eckhart Arnold's avatar
Eckhart Arnold committed
375
376
377
378
379
        "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,
380
        "zeile": [],
Eckhart Arnold's avatar
Eckhart Arnold committed
381
        "ZEICHENFOLGE, NZ, JAHRESZAHL": reduce_single_child,
382
        "WORT, NAME, LEERZEILE, ENDE": [],
Eckhart Arnold's avatar
Eckhart Arnold committed
383
384
385
386
387
        ":Whitespace": replace_content(lambda node : " "),
        ":Token, :RE": reduce_single_child,
        "*": replace_by_single_child
    }

388
389
390
391
392
As you can see, AST-transformations a specified declaratively (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. 
Eckhart Arnold's avatar
Eckhart Arnold committed
393
 
394
395
396
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.
Eckhart Arnold's avatar
Eckhart Arnold committed
397
398

*to be continued, stay tuned...*