-
Notifications
You must be signed in to change notification settings - Fork 0
/
presentation.html
547 lines (492 loc) · 21.7 KB
/
presentation.html
1
2
3
4
5
6
7
8
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
55
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
191
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
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
<html lang="en">
<head>
<meta charset="utf-8">
<title>A Free/Open-Source Morphological Transducer for Western Armenian</title>
<meta name="description" content="A Free/Open-Source Morphological Transducer for Western Armenian">
<meta name="author" content="Hovsep Dolatian and Daniel Swanson and Jonathan N. Washington">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<!-- , maximum-scale=1.0, user-scalable=no" -->
<link rel="stylesheet" href="./reveal.js/dist/reset.css">
<link rel="stylesheet" href="./reveal.js/dist/reveal.css">
<link rel="stylesheet" href="./reveal.js/dist/theme/beige.css" id="theme">
<!-- Code syntax highlighting -->
<link rel="stylesheet" href="./reveal.js/plugin/highlight/monokai.css" id="highlight-theme">
<!-- link rel="stylesheet" href="./reveal.js/lib/css/zenburn.css" -->
<!-- Printing and PDF exports -->
<script>
var link = document.createElement( 'link' );
link.rel = 'stylesheet';
link.type = 'text/css';
link.href = window.location.search.match( /print-pdf/gi ) ? './reveal.js/css/print/pdf.scss' : './reveal.js/css/print/paper.scss';
document.getElementsByTagName( 'head' )[0].appendChild( link );
</script>
<!--[if lt IE 9]>
<script src="lib/js/html5shiv.js"></script>
<![endif]-->
<link rel="stylesheet" href="jnw-overrides.css">
</head>
<body>
<div class="reveal">
<div class="slides">
<section id="title" data-state="no-toc-progress">
<style>.header1 header:after { content: "Header 1 Example"; }</style>
<h1><div style="margin-bottom: 0.5em !important ;">A Free/Open-Source Morphological Transducer for Western Armenian</div></h1>
<div style="padding-top:0in; padding-bottom: 0.25in;">
<!-- h3 -->
<div style="width: 100%; padding-bottom: 0.25in;">
<div style="width: 32%; float: left;">
<img style="width: 5.5em; height: 5.5em;" src="img/pic_HD.jpg"/><br />
Hossep Dolatian¹
</div>
<!-- /h3-->
<div style="width: 32%; float: left;">
<img style="width: 5.5em; height: 5.5em;" src="img/pic_DS.jpg"/><br />
Daniel Swanson²
</div>
<div style="width: 32%; float: left;">
<img style="width: 5.5em; height: 5.5em;" src="img/pic_JNW.jpg"/><br />
Jonathan N. Washington³
</div>
<div style="clear: both"></div>
</div>
<ol class="shrink">
<li>Stony Brook University, USA</li>
<li>Indiana University, USA</li>
<li>Swarthmore College, USA</li>
</ol>
</div>
<div style="background-color: white;">
<img class="noframe" src="img/sb_logo.jpg" style="height: 5em;"/>
<img class="noframe" src="img/iu_trident_web_crimson.svg" style="height: 5em;"/>
<img class="noframe" src="img/swarthmorelogowhite.svg" style="height: 5em;"/>
<img class="noframe" src="img/Apertium_logo_lightbg.svg" style="height: 5em;"/>
</div>
</section>
<section>
<h2>Overview</h2>
<div class="float-left">
<div class="box">
<p class="bbb">Our talk in a nutshell</p>
<p>We developed a prototype <span class="highlight">free/open source</span> morphological transducer for <span class="highlight"/>Western Armenian</span></p>
<!-- ul -->
<!-- li>Turkic non-finite verb forms often argued to be multifunctional</li -->
<!-- li>Commonly assumed multifunctionality of Turkic non-finite verb forms does not hold</li>
<ul>
<li>Conflates important distinctions</li>
<li>Too few categories</li>
</ul>
<li>Proposal for novel delineation</li>
</ul -->
</div>
</div>
<div class="float-right"><!-- style="margin-top: 1em;" --> <!-- #padding-top: 10pt | to make the top vertically align with the other box -->
<div class="box">
<p class="bbb">The next ~12 minutes:</p>
<ul>
<li>Background</li>
<ul>
<li>Western Armenian</li>
<li>Morphological tranducers</li>
</ul>
<li>Implementation and Challenges</li>
<li>Evaluation</li>
<li>Future work</li>
<li>Conclusion</li>
</ul>
</div>
</div>
</section>
<section>
<h2>Background</h2>
<h3>Western Armenian</h3>
<!-- <img src="img/Sakha.png" style="height: 10em;"><br /> -->
<ul>
<li>IE language</li>
<li>~1.6 million speakers</li>
<li>Spoken primarily in diaspora</li>
<li>Status varies from viable to endangered, based on location</li>
<li>Morphological typology
<ul>
<li> Primarily suffixing and agglutinative
<li> Multiple conjugation classes and declension classes
<li> Has root suppletion and stems
</ul> </li>
</ul>
</section>
<!--
<section data-background="img/map.svg" data-background-position="left center" data-background-size="contain">
<div style="background-color: rgba(0.3, 0.3, 0.3, 0.4); transparency 50%; position: absolute; top: 0px; right: -7em; margin: 0 0 0 0; padding: 0.5em 0.5em 0.5em 0; color: black">
<p>Might not need the map</p>
<ul style="list-style-type: none; text-shadow: 2px 2px 8px #FFFFFF;">
<li>List:</li>
<ul>
<li><span class="code">hye</span> - Eastern Armenian</li>
<li><span class="code"><b class="highlight-cyan" style="text-shadow: 2px 2px 8px #000000;">hyw</b></span> - <b class="highlight-cyan" style="text-shadow: 2px 2px 8px #000000;">Western Armenian</b></li>
</ul>
</ul>
</div>
</section>
-->
<section data-auto-animate>
<span data-id="backgroundFSTs"><h2>Background</h2>
<h3>Morphological tranducers</h3></span>
<ul>
<li>Twofold function: morphological analysis and generation<br />
<div class="center">
<span class="code" style="line-height: 1.5 !important;">սիրեցին <span class="highlight">↔</span> սիրել<tag>v</tag><tag>tv</tag><tag>past</tag><tag>pret</tag><tag>p3</tag><tag>pl</tag><tag>indc</tag></span><br />
<span class="center small ipa" style="line-height: 1.5 !important;">[siɾeʦin] ‘<em>they liked</em>’ <span class="highlight">↔</span> [siɾel] ‘<em>to like</em>’, verb, intransitive, past, preterite, 3rd person, plural, indicative</span>
</div>
</li>
<li>Implemented as finite state transducers (FST)</li>
<li>Compiled from <span class="highlight">hand-coded</span> lexical, morphotactic, and morphophonological generalisations</li>
<li>Only one development cycle</li>
</ul>
</section>
<section data-auto-animate>
<span data-id="backgroundFSTs"><h2>Background</h2>
<h3>Morphological tranducers</h3></span>
<ul>
<li><span class="highlight">many uses</span> in language technology and “downstream” tasks:</li>
<ul>
<li>can be repurposed as spell checkers</li>
<li>may be used in rule-based machine translation pipelines</li>
</ul>
<li>Some <span class="highlight">current uses</span> of this transducer:</li>
<ul>
<li>Verb paradigm generator <span class="small">(<a href="https://apertium.github.io/apertium-paradigmatrix/">apertium.github.io/apertium-paradigmatrix</a>)</span></li>
<li>Virtually-complete morphology helps find synchronic variation, and </li>
<li>Useful when writing a reference grammar</li>
<li>Looking into pedagogical uses</li>
</ul>
</ul>
</section>
<section>
<h2> Number of lemmas</h2>
<div class="float-left">
<table class="shrink nolines">
<tr><th></th><th>category</th><th>entries</th><th>tag</th></tr>
<tr class="addline-top"><th rowspan="4" class="rot90"><div>Core POS</div></th><td>Noun</td><td class="raln">39 006</td><td><tag>n</tag></td></tr>
<tr><td>Adjective</td><td class="raln">18 617</td><td><tag>adj</tag></td></tr>
<tr><td>Verb</td><td class="raln">7 441</td><td><tag>v</tag></td></tr>
<tr><td>Adverb</td><td class="raln">1 895</td><td><tag>adv</tag></td></tr>
<tr class="addline-top"><th rowspan="4" class="rot90"><div>Names</div></th><td>Given name</td><td class="raln">4 848</td><td><tag>np</tag><tag>ant</tag></td></tr>
<tr><td>Surname</td><td class="raln">2 052</td><td><tag>np</tag><tag>cog</tag></td></tr>
<tr><td>Location name</td><td class="raln">1 183</td><td><tag>np</tag><tag>top</tag></td></tr>
<tr><td>Other name</td><td class="raln">22</td><td><tag>np</tag><tag>al</tag></td></tr>
<tr class="addline-top"><th rowspan="7" class="rot90"><div>Function, other</div></th><td>Pronoun</td><td class="raln">415</td><td><tag>prn</tag></td></tr>
<tr><td>Adposition</td><td class="raln">130</td><td><tag>pr</tag>, <tag>post</tag></td></tr>
<tr><td>Abbreviation</td><td class="raln">81</td><td><tag>abbr</tag></td></tr>
<tr><td>Interjection</td><td class="raln">49</td><td><tag>ij</tag></td></tr>
<tr><td>Conjunction</td><td class="raln">48</td><td><tag>conj</tag></td></tr>
<tr><td>Numeral</td><td class="raln">41</td><td><tag>num</tag></td></tr>
<tr><td>Particle</td><td class="raln">9</td><td><tag>particle</tag></td></tr>
<tr class="addline-top addline-bottom"><td></td><th class="laln">Total</th><td class="raln">75 837</td><td></td></tr>
</table>
</div>
<!-- div style="float: center;">
<img src="img/lexicon.jpg"/><br />
</div -->
<div class="float-right">Recognizes/generates<br/><span class="highlight">millions</span> of<br/>inflected forms </div>
</section>
<!--
<section>
<h2>Existing transducers for comparison</h2>
<h3>Comparison of lexicon size</h3>
<div>hyw: >999 999 stems</div>
</section>
-->
<section>
<h2>Implementation</h2>
<ul class="nobullet">
<li>Two-level approach using <tt>lexd</tt>+HFST:</li>
<ul class="shrink">
<li>“diacritics” in lexical form, like <code><arch>Plն</arch></code></li>
<li>used in morpholophonology to select phonologically-conditioned allomorphs<br/>(e.g., of the plural and definite)</li>
</ul>
</ul>
<!-- The lexical form uses diacritics to later pick phonologically-conditioned allomorphs of the plural and definite -->
<div class="box" style="width: 100%">
<h4>orthographic form</h4>
<div style="margin-bottom: 0.5em !important;"><span class="code">ձեռքերէն</span></div><span class="ipa small">/ʦeɾkʰ-eɾ-e-n/<br/>‘<em>from the hands</em>’</span>
</div>
<p>↕ (<code class="highlight">twol</code>: phonological mappings)</p>
<div class="box" style="width: 100%">
<h4>morphological/lexical form</h4>
<span class="code">ձեռք><arch>Ninfl</arch>><arch>Plն</arch>եր>է><arch>defն</arch></span>
</div>
<p>↕ (<code>lexc</code>: lexicon + morphotactics)</p>
<div class="box" style="width: 100%">
<h4>analysis: <span class="highlight">lemma, POS, grammatical tags</span></h4>
<span class="code">ձեռք<tag>n</tag><tag>pl</tag><tag>abl</tag><tag>def</tag></span>
</div>
</section>
<section>
<h2>Challenges: Lexicon compilations</h3>
<div class="box">
<h3>Problem:</h3>
<ul>
<!-- li> For a compound with monosyllabic second stem, need to manually check if the semantics use plural -եր /-eɾ/ or -ներ /-neɾ/ </li -->
<!-- li>There are many Armenian dictionaries online, but essentially none of them provide data formats (text files) that allow for intensive use </li -->
<li>Many online Armenian dictionaries, but data formats don't allow for intensive use</li>
<li>Ideal: text files or accessible databases</li>
</ul>
</div>
<div class="box">
<h3>Solution: </h3>
<ul>
<li>Scraped XML files from the Kouyoumdjian dictionary from
<a href="nayiri.com/">Nayiri</a> </li>
<li>Extracted a 70K lexicon that we used to populate our analyser</li>
<li>Unfortunately Nayiri doesn't publicly offer their XML anymore :( </li>
<li>But you can use our adaptation of it from our <a href="https://github.com/apertium/apertium-hyw/tree/master/lexd_components">GitHub</a> :) </li>
</ul>
</div>
<aside class="notes">
<p class="shrink">What to read, in presenter notes</p>
</aside>
</section>
<section>
<h2>Challenges: compound plurals </h2>
<div class="box">
<h3>Problem:</h3>
<ul>
<!-- li> For a compound with monosyllabic second stem, need to manually check if the semantics use plural -եր /-eɾ/ or -ներ /-neɾ/ </li -->
<li>Compound with monosyllabic second stem:<br/>Is plural -եր /-eɾ/ or -ներ /-neɾ/?</li>
<ul>
<li>
<span class="code">դրամատուներ ↔ դրամատուն<tag>n</tag><tag>pl</tag></span>
<div class="ipa">[tʰəɾɑmɑdun-eɾ] ↔ [tʰəɾɑmɑdun] ‘<em>money-houses</em>’</div>
</li>
<li>
<span class="code">մեծատուններ ↔ մեծատուն<tag>n</tag><tag>pl</tag></span>
<div class="ipa"> [med͡zɑdun-neɾ] ↔ [med͡zɑdun] ‘<em>big-housed</em>’</div>
</li> </ul>
<li>Need to manually code based on semantics</li>
</ul>
</div>
<div class="box">
<h3>Solution: </h3>
<ul>
<li>Double-checked against EANC to get the right semantics</li>
</ul>
</div>
<aside class="notes">
<p class="shrink">What to read, in presenter notes</p>
</aside>
</section>
<section>
<h2>Challenges: infix punctuation </h2>
<div class="box">
<h3>Problem:</h3>
<ul> <li> Punctuation markers are added after vowel of stressed syllable of word<br />
<table class="nolines">
<tr>
<th>Final stress</th>
<th>Pre-schwa stress</th>
<th>Irregular stress</th>
</tr>
<tr>
<td>կապի՞կ</td>
<td>կապի՞կը</td>
<td>ո՞րքան</td>
</tr>
<tr>
<td class="ipa">[ɡɑˈbi<sup>?</sup>ɡ]</td>
<td class="ipa">[ɡɑˈbi<sup>?</sup>ɡə]</td>
<td class="ipa">[ˈvo<sup>?</sup>ɾkʰɑn]</td>
</tr>
<tr class="addline-bottom">
<td>‘<em>monkey?</em>’</td>
<td>‘<em>the monkey?</em>’</td>
<td>‘<em>how many?</em>’</td>
</tr>
</table> </li> </ul>
</div>
<div class="box">
<h3>Solution: </h3>
<ul>
<li>Have rules that add infix at end of word, and then “metathesise” (actually: delete and copy)</li>
<li>Add control characters at location of irregular stress</li>
</ul>
</div>
<aside class="notes">
<p class="shrink">What to read, in presenter notes</p>
</aside>
</section>
<section>
<h2>Evaluation: Coverage</h2>
<div class="float-left">
<div class="box">
<ul>
<li><span class="highlight">Naive coverage</span>: <em>the number of forms in a corpus that receive an analysis, regardless of whether or not the analysis is correct (e.g., in context)</em></li>
</ul>
</div>
</div>
<div class="float-right" style="margin-top: 1em">
<table>
<tr>
<th>Corpus</th>
<th>Tokens</th>
<th>Coverage</th>
</tr>
<tr>
<td>Newspapers</td>
<td class="raln">~16M</td>
<td class="raln">87.68%</td>
</tr>
<tr>
<td>Wikipedia</td>
<td class="raln">~2.59M</td>
<td class="raln">88.79%</td>
</tr>
<tr>
<td>New Testament</td>
<td class="raln">~190K</td>
<td class="raln">90.87%</td>
</tr>
</table>
<p class="shrink">Adequate coverage: robust morphology, but many missing stems</p>
</div>
</section>
<section>
<h2>Evaluation: Precision & Recall</h2>
<div>
<ul>
<li>created gold standard:</li>
<ul>
<li>1225 valid words of Western Armenian</li>
<li>randomly selected from Newspaper corpus</li>
<li>manually annotated output of transducer</li>
</ul>
</ul>
</div>
<div class="float-left">
<div class="box">
<ul class="nobullet">
<li style="margin-bottom: 1em;"><span class="highlight">precision</span>: <em>how many of the transducer-provided analyses for the tokens were correct<!-- percentage of analyses returned by the transducer that is correct --></em></li>
<li><span class="highlight">recall</span>: <em>how many of the correct analyses were retrieved from the transducer<!-- percentage of the correct analyses that is returned --></em></li>
</ul>
</div>
</div>
<div class="float-right">
<div>results:</div>
<table>
<tr>
<th>Tokens</th>
<th>Precision</th>
<th>Recall</th>
</tr>
<tr>
<td>1225</td>
<td class="raln">90.58%</td>
<td class="raln">74.82%</td>
</tr>
</table>
<p class="shrink">i.e., most forms returned by the transducer were deemed correct, but many correct analyses were not returned by the transducer (mostly due to low coverage)</p>
</div>
</section>
<section>
<h2>Future work</h2>
<ul>
<li>Expand <span class="highlight">lexicon</span> (add more lemmas)</li>
<li>Morphological/syntactic <span class="highlight">disambiguation</span></li>
<li>More language technology applications of transducer? (spell checkers, MT, etc.?)</li>
<li><em>Slowly</em> working on incorporating Eastern morphology so that the analyzer is <span class="highlight">bi-dialectal</span></li>
<ul class="nobullet">
<li class="arrow">Simple MT between different varieties of Armenian</li>
</ul>
<li> Incorporate an orthography-pronunciation convertor</li>
<li> Honestly, it's open-source, so you can do WHATEVER you want with it! </li>
</ul>
</section>
<section>
<h2>Conclusion</h2>
<ul>
<li>Expansion of language technology for Western Armenian</li>
<li>Robust transducer</li>
<ul>
<li>high coverage</li>
<li>high precision</li>
<li>moderate recall</li>
</ul>
<li>Lots of room for improvement!</li>
<li>Ready for use in</li>
<ul>
<li>language technology applications</li>
<li>downstream tasks</li>
</ul>
</ul>
<div class="box fragment">
<p class="bbb"><span class="cyrillic">Շնորհակա՛լ ենք</span>! / Thank you!</p>
<ul>
<li>Fork at <a href="https://github.com/apertium/apertium-hyx">github.com/apertium/apertium-hyw</a></li>
<li>Try out at <a href="https://beta.apertium.org/#analysis?aLang=hyx_hyw">beta.apertium.org</a></li>
</ul>
</div>
</section>
</div>
</div>
<!-- script src="./reveal.js/lib/js/head.min.js"></script -->
<script src="./reveal.js/dist/reveal.js"></script>
<script src="./reveal.js/plugin/notes/notes.js"></script>
<script src="./reveal.js/plugin/markdown/markdown.js"></script>
<script src="./reveal.js/plugin/highlight/highlight.js"></script>
<script src="./reveal.js/plugin/zoom/zoom.js"></script>
<script>
// Full list of configuration options available here:
// https://github.com/hakimel/reveal.js#configuration
Reveal.initialize({
controls: true,
progress: true,
center: true,
history: true,
hash: true,
// The "normal" size of the presentation, aspect ratio will be preserved
// when the presentation is scaled to fit different resolutions. Can be
// specified using percentage units.
width: 1280, //960,
//height: 720, //700,
height: 1024,
//width: 960,
//height: 700,
// Factor of the display size that should remain empty around the content
margin: 0.1,
// Bounds for smallest/largest possible scale to apply to content
minScale: 0.2,
maxScale: 1.0,
//theme: Reveal.getQueryHash().theme, // available themes are in /css/theme
transition: 'slide', // none/fade/slide/convex/concave/zoom
// transition: Reveal.getQueryHash().transition || 'fade', // default/cube/page/concave/zoom/linear/fade/none
// Shows the slide number using default formatting
//Reveal.configure({ slideNumber: true });
//slideNumber: true,
//if true (default), fragments are generated on separate pages of pdfs
//pdfSeparateFragments: false,
// Slide number formatting can be configured using these variables:
// "h.v": horizontal . vertical slide number (default)
// "h/v": horizontal / vertical slide number
// "c": flattened slide number
// "c/t": flattened slide number / total slides
//Reveal.configure({ slideNumber: 'c' });
slideNumber: 'c',
// Parallax scrolling
// parallaxBackgroundImage: 'https://s3.amazonaws.com/hakim-static/reveal-js/reveal-parallax-1.jpg',
// parallaxBackgroundSize: '2100px 900px',
// Optional reveal.js plugins
// More info https://github.com/hakimel/reveal.js#dependencies
/***
dependencies: [
{ src: './reveal.js/plugin/markdown/marked.js', condition: function() { return !!document.querySelector( '[data-markdown]' ); } },
{ src: './reveal.js/plugin/markdown/markdown.js', condition: function() { return !!document.querySelector( '[data-markdown]' ); } },
{ src: './reveal.js/plugin/highlight/highlight.js' },
{ src: './reveal.js/plugin/search/search.js', async: true },
{ src: './reveal.js/plugin/zoom-js/zoom.js', async: true },
{ src: './reveal.js/plugin/notes/notes.js', async: true }
]
***/
// Learn about plugins: https://revealjs.com/plugins/
plugins: [ RevealMarkdown, RevealZoom, RevealHighlight, RevealNotes ]
});
</script>
</body>
</html>