-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
1402 lines (1322 loc) · 78.4 KB
/
Copy pathindex.html
File metadata and controls
1402 lines (1322 loc) · 78.4 KB
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
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>AI as a scout, not a judge — PARSE & Southern Kurdish (projector)</title>
<!-- Open Graph / social link preview — uses assets/hero.png (1200x630).
To re-render the hero after tweaks, run Chrome headless against
assets/hero.template.html and overwrite assets/hero.png. -->
<meta property="og:type" content="website">
<meta property="og:title" content="AI as a scout, not a judge — PARSE & Southern Kurdish">
<meta property="og:description" content="A 15-minute talk on how AI assists a Southern Kurdish dialect phylogenetics project, and the PARSE workstation built for it.">
<meta property="og:image" content="https://ardeleanlucas.github.io/parse-ai-talk/assets/hero.png">
<meta property="og:image:width" content="1200">
<meta property="og:image:height" content="630">
<meta property="og:url" content="https://ardeleanlucas.github.io/parse-ai-talk/">
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:title" content="AI as a scout, not a judge — PARSE & Southern Kurdish">
<meta name="twitter:description" content="A 15-minute talk on how AI assists a Southern Kurdish dialect phylogenetics project, and the PARSE workstation built for it.">
<meta name="twitter:image" content="https://ardeleanlucas.github.io/parse-ai-talk/assets/hero.png">
<!-- Swiss Modern: Archivo (display) + Nunito (body). Both sans, high-x-height. -->
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Archivo:wght@500;600;800;900&family=Nunito:wght@400;600;700&display=swap" rel="stylesheet">
<style>
/* ===========================================
SWISS MODERN — PROJECTOR EDITION
Bauhaus aesthetic tuned for a lecture-hall screen.
- Pure black on near-white for maximum contrast
- Sans-serif throughout (Archivo for display, Nunito for body)
- Minimum body font ≈ 20px so the back row can read every word
- No italic body text; bold instead, italics reserved for citations
- 1-2px solid black rules instead of hairlines
=========================================== */
:root {
--bg-primary: #fafafa;
--bg-tint: #f0f0f0;
--ink: #000000;
--ink-soft: #1a1a1a;
--ink-mute: #595959;
--rule: #000000;
--accent: #ff3300;
--accent-soft: rgba(255, 51, 0, 0.10);
--font-display: 'Archivo', 'Helvetica Neue', sans-serif;
--font-body: 'Nunito', 'Helvetica Neue', sans-serif;
/* PROJECTOR TYPE SCALE
Floor values bumped ~30-50% vs. the desktop version,
so even on a small laptop the text feels readable from row 15. */
--title-size: clamp(2.75rem, 6.5vw, 5.5rem);
--h2-size: clamp(2rem, 4.2vw, 3.5rem);
--h3-size: clamp(1.4rem, 2.4vw, 1.875rem);
--body-size: clamp(1.25rem, 1.8vw, 1.625rem);
--small-size: clamp(1rem, 1.4vw, 1.2rem);
--eyebrow-size: clamp(0.85rem, 1.1vw, 1rem);
--slide-padding: clamp(1.75rem, 4vw, 3.5rem);
--content-gap: clamp(0.85rem, 1.8vw, 1.75rem);
--element-gap: clamp(0.4rem, 0.9vw, 0.9rem);
--ease-out-expo: cubic-bezier(0.16, 1, 0.3, 1);
--duration-normal: 0.5s;
}
* { margin: 0; padding: 0; box-sizing: border-box; }
html, body { height: 100%; overflow-x: hidden; }
html {
scroll-snap-type: y mandatory;
scroll-behavior: smooth;
}
body {
font-family: var(--font-body);
font-size: var(--body-size);
color: var(--ink);
background: var(--bg-primary);
line-height: 1.45;
-webkit-font-smoothing: antialiased;
font-weight: 400;
}
/* ===========================================
SLIDE CONTAINER
=========================================== */
.slide {
width: 100vw;
height: 100vh;
height: 100dvh;
padding: var(--slide-padding);
scroll-snap-align: start;
display: flex;
flex-direction: column;
justify-content: center;
position: relative;
overflow: hidden;
background: var(--bg-primary);
}
/* Folio + header — bigger than desktop deck so still readable from far */
.slide::before {
content: attr(data-folio);
position: absolute;
top: clamp(0.85rem, 1.6vw, 1.5rem);
left: var(--slide-padding);
font-family: var(--font-body);
font-size: var(--eyebrow-size);
font-weight: 700;
color: var(--ink);
letter-spacing: 0.18em;
text-transform: uppercase;
}
.slide::after {
content: 'ARDELEAN · PARSE · 2026';
position: absolute;
top: clamp(0.85rem, 1.6vw, 1.5rem);
right: var(--slide-padding);
font-family: var(--font-body);
font-size: var(--eyebrow-size);
font-weight: 700;
color: var(--ink);
letter-spacing: 0.18em;
text-transform: uppercase;
}
.slide.title-slide::before,
.slide.title-slide::after { display: none; }
.slide-content {
flex: 1;
display: flex;
flex-direction: column;
justify-content: center;
max-height: 100%;
overflow: hidden;
gap: var(--content-gap);
}
/* ===========================================
TYPOGRAPHY
Headings use Archivo 800 — chunky, high-contrast.
=========================================== */
h1, h2, h3 {
font-family: var(--font-display);
font-weight: 800;
line-height: 1.02;
letter-spacing: -0.02em;
color: var(--ink);
}
h1 { font-size: var(--title-size); }
h2 { font-size: var(--h2-size); }
h3 { font-size: var(--h3-size); font-weight: 800; letter-spacing: -0.01em; }
.eyebrow {
font-family: var(--font-body);
font-size: var(--eyebrow-size);
text-transform: uppercase;
letter-spacing: 0.22em;
color: var(--accent);
font-weight: 700;
}
/* The "lede" subtitle, used on title + manifesto slides */
.lede {
font-family: var(--font-body);
font-size: clamp(1.25rem, 2.2vw, 1.75rem);
font-weight: 400;
color: var(--ink-soft);
line-height: 1.35;
max-width: 32ch;
}
.rule {
width: clamp(3rem, 7vw, 5.5rem);
height: 4px;
background: var(--accent);
border: none;
margin: var(--element-gap) 0;
}
/* ===========================================
LISTS — bigger bullets, thicker leading marks
=========================================== */
ul.bullets {
list-style: none;
display: flex;
flex-direction: column;
gap: clamp(0.6rem, 1.3vw, 1.1rem);
}
ul.bullets li {
position: relative;
padding-left: clamp(1.5rem, 2vw, 2rem);
font-size: var(--body-size);
line-height: 1.45;
color: var(--ink-soft);
}
ul.bullets li::before {
content: '';
position: absolute;
left: 0;
top: 0.65em;
width: clamp(0.85rem, 1.3vw, 1.1rem);
height: 3px;
background: var(--accent);
}
ul.bullets li strong {
color: var(--ink);
font-weight: 800;
}
/* Avoid italic body text — bad on projectors at distance.
Inline emphasis becomes the accent color + bold instead. */
ul.bullets li em {
font-style: normal;
font-weight: 700;
color: var(--accent);
}
ul.checks {
list-style: none;
display: flex;
flex-direction: column;
gap: clamp(0.55rem, 1.2vw, 1rem);
}
ul.checks li {
display: grid;
grid-template-columns: clamp(1.5rem, 2vw, 1.85rem) 1fr;
gap: clamp(0.6rem, 1.1vw, 0.95rem);
font-size: var(--body-size);
line-height: 1.4;
color: var(--ink-soft);
}
ul.checks li::before {
content: '✓';
font-family: var(--font-display);
color: var(--accent);
font-weight: 900;
line-height: 1.25;
font-size: 1.1em;
}
ul.checks li.cross::before {
content: '✕';
color: var(--ink-mute);
}
ul.checks li em {
font-style: normal;
font-weight: 700;
color: var(--ink);
}
/* ===========================================
LAYOUT PATTERNS
=========================================== */
.two-col {
display: grid;
grid-template-columns: 1fr 1.05fr;
gap: clamp(1.5rem, 4vw, 3.5rem);
align-items: center;
}
.two-col .col-text { display: flex; flex-direction: column; gap: var(--content-gap); min-width: 0; }
.two-col .col-figure { display: flex; align-items: center; justify-content: center; min-width: 0; }
/* Figures — black hard border, no soft shadows.
Bauhaus restraint and high projector legibility. */
figure.shot {
max-width: 100%;
max-height: min(72vh, 640px);
border: 2px solid var(--ink);
background: #ffffff;
padding: clamp(0.3rem, 0.7vw, 0.6rem);
}
figure.shot img {
display: block;
max-width: 100%;
max-height: min(64vh, 560px);
object-fit: contain;
}
figure.shot figcaption {
font-size: var(--small-size);
color: var(--ink);
text-align: center;
padding-top: clamp(0.35rem, 0.8vw, 0.6rem);
font-weight: 600;
}
/* Image-pair slide */
.pair-grid {
display: grid;
grid-template-columns: 1fr 1fr;
gap: clamp(1rem, 2.5vw, 2rem);
align-items: stretch;
}
.pair-grid figure.shot { width: 100%; max-height: min(62vh, 520px); }
.pair-grid figure.shot img { max-height: min(55vh, 460px); }
/* ===========================================
TITLE SLIDE
=========================================== */
.slide.title-slide {
justify-content: center;
background:
linear-gradient(var(--ink) 0%, var(--ink) 100%) 0 0 / clamp(40px, 8vw, 100px) 4px no-repeat,
var(--bg-primary);
}
.title-slide .slide-content {
justify-content: center;
align-items: flex-start;
max-width: 22ch;
}
.title-slide h1 {
font-size: clamp(3rem, 8vw, 6.5rem);
line-height: 0.95;
margin-bottom: clamp(0.75rem, 1.6vw, 1.3rem);
color: var(--ink);
}
.title-slide h1 .accent { color: var(--accent); }
.title-slide .lede {
font-size: clamp(1.35rem, 2.6vw, 2rem);
margin-bottom: clamp(1.25rem, 2.8vw, 2.25rem);
max-width: 28ch;
}
.title-slide .byline {
font-size: var(--small-size);
color: var(--ink);
letter-spacing: 0.14em;
text-transform: uppercase;
font-weight: 700;
}
.title-slide .byline strong { color: var(--accent); font-weight: 800; }
/* Big red signature block in the bottom-right of the title slide */
/* QR to the live deck — sits in the bottom-right of the title slide.
White background and a hard black border give phone cameras a clean
target across a lecture hall (~6m at this size). */
.title-slide .title-qr {
position: absolute;
bottom: var(--slide-padding);
right: var(--slide-padding);
display: flex;
flex-direction: column;
align-items: center;
gap: clamp(0.35rem, 0.8vw, 0.65rem);
}
.title-slide .title-qr img {
width: clamp(220px, 22vw, 380px);
height: clamp(220px, 22vw, 380px);
display: block;
background: #ffffff;
padding: clamp(8px, 0.9vw, 14px);
border: 3px solid var(--ink);
object-fit: contain;
}
.title-slide .title-qr .qr-label {
font-family: var(--font-body);
font-size: var(--eyebrow-size);
font-weight: 700;
color: var(--ink);
text-transform: uppercase;
letter-spacing: 0.18em;
text-align: center;
}
/* Red decorative corner block — hidden by default, shown only on
mobile where the QR is suppressed (audience already on the device). */
.title-slide .corner-block {
display: none;
position: absolute;
bottom: var(--slide-padding);
right: var(--slide-padding);
background: var(--accent);
}
/* ===========================================
MANIFESTO SLIDE (Slide 4)
=========================================== */
.pull-quote {
font-family: var(--font-display);
font-weight: 800;
font-size: clamp(1.85rem, 4vw, 3rem);
line-height: 1.1;
color: var(--ink);
max-width: 26ch;
margin: 0 auto;
text-align: center;
letter-spacing: -0.015em;
}
.pull-quote .accent { color: var(--accent); }
.principles {
display: grid;
grid-template-columns: repeat(2, 1fr);
gap: clamp(1rem, 2.2vw, 1.85rem) clamp(1.5rem, 3vw, 2.5rem);
max-width: 64ch;
margin: var(--content-gap) auto 0;
}
.principles .principle {
border-top: 3px solid var(--ink);
padding-top: clamp(0.5rem, 1vw, 0.85rem);
}
.principles .principle .num {
font-family: var(--font-display);
color: var(--accent);
font-weight: 900;
font-size: clamp(0.95rem, 1.4vw, 1.15rem);
letter-spacing: 0.18em;
}
.principles .principle h3 {
font-size: clamp(1.1rem, 1.9vw, 1.45rem);
font-weight: 800;
margin: 0.15em 0 0.35em;
}
.principles .principle p {
font-size: var(--small-size);
color: var(--ink-soft);
line-height: 1.4;
}
/* ===========================================
PIPELINE DIAGRAM SLIDE
=========================================== */
.pipeline {
display: grid;
grid-template-columns: repeat(4, 1fr);
gap: clamp(0.6rem, 1.4vw, 1.2rem);
margin-top: var(--content-gap);
align-items: stretch;
}
.station {
background: #ffffff;
border: 2px solid var(--ink);
padding: clamp(0.85rem, 1.5vw, 1.25rem);
display: flex;
flex-direction: column;
gap: clamp(0.3rem, 0.7vw, 0.55rem);
position: relative;
}
.station .step {
font-family: var(--font-body);
font-size: var(--eyebrow-size);
letter-spacing: 0.2em;
color: var(--accent);
text-transform: uppercase;
font-weight: 800;
}
.station h3 {
font-size: clamp(1.1rem, 1.85vw, 1.45rem);
font-weight: 800;
line-height: 1.05;
}
.station p {
font-size: var(--small-size);
color: var(--ink-soft);
line-height: 1.35;
}
.station .who {
margin-top: auto;
font-size: var(--eyebrow-size);
font-weight: 800;
letter-spacing: 0.16em;
text-transform: uppercase;
color: var(--ink-mute);
padding-top: clamp(0.3rem, 0.7vw, 0.55rem);
border-top: 1px solid var(--rule);
}
.station .who.ai { color: var(--accent); }
.station .who.human { color: var(--ink); }
.station .who.compute { color: var(--ink); }
.review-band {
margin-top: clamp(0.75rem, 1.5vw, 1.25rem);
padding: clamp(0.7rem, 1.3vw, 1rem) clamp(1rem, 1.8vw, 1.4rem);
background: var(--ink);
color: var(--bg-primary);
font-family: var(--font-display);
font-weight: 700;
font-size: clamp(1.05rem, 1.7vw, 1.35rem);
text-align: center;
letter-spacing: -0.005em;
}
/* ===========================================
CLOSING SLIDE
=========================================== */
.closing {
max-width: 48ch;
margin: 0 auto;
}
.closing .takeaway {
font-family: var(--font-display);
font-size: clamp(1.75rem, 3.4vw, 2.65rem);
line-height: 1.15;
font-weight: 800;
color: var(--ink);
margin-bottom: var(--content-gap);
text-align: center;
letter-spacing: -0.015em;
}
.closing .takeaway .accent { color: var(--accent); }
.closing .body-text {
font-size: var(--body-size);
color: var(--ink-soft);
margin-bottom: var(--content-gap);
text-align: left;
line-height: 1.45;
}
.closing .links {
text-align: center;
font-size: var(--small-size);
color: var(--ink);
letter-spacing: 0.04em;
font-weight: 600;
}
.closing .links a {
color: var(--accent);
text-decoration: none;
border-bottom: 3px solid var(--accent);
font-weight: 800;
}
/* ===========================================
ANIMATIONS — fast, restrained
=========================================== */
.reveal {
opacity: 0;
transform: translateY(14px);
transition: opacity var(--duration-normal) var(--ease-out-expo),
transform var(--duration-normal) var(--ease-out-expo);
}
.slide.visible .reveal { opacity: 1; transform: translateY(0); }
.slide.visible .reveal:nth-child(1) { transition-delay: 0.04s; }
.slide.visible .reveal:nth-child(2) { transition-delay: 0.12s; }
.slide.visible .reveal:nth-child(3) { transition-delay: 0.20s; }
.slide.visible .reveal:nth-child(4) { transition-delay: 0.28s; }
.slide.visible .reveal:nth-child(5) { transition-delay: 0.36s; }
.slide.visible .reveal:nth-child(6) { transition-delay: 0.44s; }
/* ===========================================
NAV DOTS + PROGRESS BAR
=========================================== */
.progress-bar {
position: fixed;
top: 0; left: 0;
height: 4px;
background: var(--accent);
width: 0%;
z-index: 100;
transition: width 0.25s linear;
}
.nav-dots {
position: fixed;
right: clamp(0.85rem, 1.6vw, 1.75rem);
top: 50%;
transform: translateY(-50%);
display: flex;
flex-direction: column;
gap: clamp(0.55rem, 1vw, 0.85rem);
z-index: 50;
}
.nav-dots button {
width: 12px; height: 12px;
border-radius: 0;
border: 2px solid var(--ink);
background: transparent;
cursor: pointer;
padding: 0;
transition: background 0.25s, transform 0.25s;
}
.nav-dots button:hover { transform: scale(1.3); }
.nav-dots button.active {
background: var(--accent);
border-color: var(--accent);
}
.keyboard-hint {
position: fixed;
bottom: clamp(0.85rem, 1.6vw, 1.75rem);
right: clamp(0.85rem, 1.6vw, 1.75rem);
font-size: var(--eyebrow-size);
color: var(--ink);
letter-spacing: 0.12em;
text-transform: uppercase;
font-weight: 700;
z-index: 50;
pointer-events: none;
}
.keyboard-hint kbd {
font-family: var(--font-body);
background: var(--ink);
color: var(--bg-primary);
border: none;
padding: 0.15em 0.45em;
margin: 0 0.15em;
font-size: 0.95em;
font-weight: 700;
border-radius: 0;
}
/* ===========================================
PRESENTER NOTES OVERLAY
Press N to toggle.
=========================================== */
#presenter-notes {
position: fixed;
bottom: 0; left: 0; right: 0;
background: var(--ink);
color: var(--bg-primary);
padding: clamp(1.1rem, 2.2vw, 1.85rem) clamp(1.75rem, 4vw, 3.5rem);
max-height: 42vh;
overflow-y: auto;
z-index: 200;
transform: translateY(100%);
transition: transform 0.3s var(--ease-out-expo);
font-family: var(--font-body);
font-size: clamp(1rem, 1.3vw, 1.2rem);
line-height: 1.5;
border-top: 4px solid var(--accent);
}
#presenter-notes.open { transform: translateY(0); }
#presenter-notes .notes-label {
display: block;
font-size: 0.85em;
letter-spacing: 0.22em;
text-transform: uppercase;
color: var(--accent);
margin-bottom: 0.5em;
font-weight: 800;
}
#presenter-notes .notes-body {
max-width: 72ch;
color: rgba(250, 250, 250, 0.95);
}
/* ===========================================
BREAKPOINTS
Floor is high enough that even small windows stay readable.
=========================================== */
@media (max-height: 760px) {
:root {
--slide-padding: clamp(1.25rem, 3vw, 2.5rem);
--content-gap: clamp(0.65rem, 1.4vw, 1.2rem);
--title-size: clamp(2.25rem, 5.5vw, 4rem);
--h2-size: clamp(1.6rem, 3.5vw, 2.5rem);
--body-size: clamp(1.05rem, 1.55vw, 1.35rem);
}
figure.shot { max-height: min(60vh, 480px); }
figure.shot img { max-height: min(54vh, 420px); }
}
@media (max-height: 600px) {
:root {
--slide-padding: clamp(0.9rem, 2.5vw, 1.75rem);
--content-gap: clamp(0.45rem, 1.1vw, 0.85rem);
--body-size: clamp(0.95rem, 1.4vw, 1.15rem);
--title-size: clamp(1.75rem, 4.5vw, 2.75rem);
--h2-size: clamp(1.35rem, 3vw, 2rem);
}
.nav-dots, .keyboard-hint { display: none; }
figure.shot { max-height: 48vh; }
figure.shot img { max-height: 42vh; }
}
@media (max-width: 820px) {
.two-col, .pair-grid, .principles, .pipeline {
grid-template-columns: 1fr;
}
.pipeline { grid-template-columns: repeat(2, 1fr); }
.two-col .col-figure { order: -1; }
figure.shot, .pair-grid figure.shot {
max-height: 38vh;
}
figure.shot img, .pair-grid figure.shot img {
max-height: 32vh;
}
}
@media (max-width: 540px) {
:root { --title-size: clamp(1.6rem, 8vw, 2.5rem); }
.pipeline { grid-template-columns: 1fr; }
}
/* ===========================================
MOBILE / TOUCH — let the deck scroll as one document.
The "one slide = one viewport" rule is right for projectors
but wrong for tall narrow phone screens. Here we:
- Drop scroll-snap, let slides be as tall as their content
- Stack every multi-column layout
- Show image-first on text+figure slides
- Hide projector UI (nav dots, keyboard hint, presenter notes)
- Tune the type scale for phone-reading distance
Triggered by narrow width OR by a coarse pointer (touch).
=========================================== */
@media (max-width: 768px), (pointer: coarse) {
html {
scroll-snap-type: none;
}
body {
line-height: 1.5;
}
.slide {
height: auto;
min-height: 100vh;
min-height: 100dvh;
overflow: visible;
scroll-snap-align: none;
justify-content: flex-start;
padding-top: clamp(3rem, 8vw, 4rem);
padding-bottom: clamp(2.5rem, 7vw, 3.5rem);
border-bottom: 2px solid var(--ink);
}
.slide:last-of-type {
border-bottom: none;
}
.slide-content {
max-height: none;
overflow: visible;
justify-content: flex-start;
}
/* Folio chips still useful as section labels, just smaller */
.slide::before, .slide::after {
font-size: clamp(0.6rem, 2.6vw, 0.78rem);
top: clamp(0.7rem, 2.2vw, 1rem);
}
/* Every multi-column layout collapses to a single stack */
.two-col,
.pair-grid,
.principles,
.pipeline {
grid-template-columns: 1fr;
gap: clamp(0.85rem, 3vw, 1.5rem);
}
/* Show the screenshot above the bullets on text+figure slides */
.two-col .col-figure { order: -1; }
/* The honest ledger uses an inline grid-template-columns;
attribute selector + !important is needed to override it */
.slide-content > div[style*="grid-template-columns: 1fr 1fr"] {
grid-template-columns: 1fr !important;
gap: clamp(1rem, 4vw, 1.5rem) !important;
}
/* Figures sized for phone, never bigger than half the screen */
figure.shot {
max-height: none;
}
figure.shot img {
max-height: 55vh;
}
/* Mobile type scale — readable at arm's length on a phone */
:root {
--title-size: clamp(2rem, 9vw, 3.5rem);
--h2-size: clamp(1.6rem, 6.5vw, 2.4rem);
--h3-size: clamp(1.15rem, 4.5vw, 1.5rem);
--body-size: clamp(1rem, 4vw, 1.2rem);
--small-size: clamp(0.9rem, 3.5vw, 1.05rem);
--eyebrow-size: clamp(0.7rem, 2.8vw, 0.85rem);
--slide-padding: clamp(1.1rem, 5vw, 2rem);
--content-gap: clamp(0.85rem, 3vw, 1.4rem);
}
.pull-quote {
font-size: clamp(1.5rem, 6vw, 2.25rem);
max-width: 100%;
}
.closing .takeaway {
font-size: clamp(1.5rem, 6vw, 2.25rem);
}
.title-slide h1 {
font-size: clamp(2.5rem, 10vw, 4rem);
}
.title-slide .lede {
font-size: clamp(1.15rem, 4.5vw, 1.5rem);
}
/* On phones, the audience is already on the device — the QR
would only point them to the page they're already viewing.
Hide it and bring back the red decorative block instead. */
.title-slide .title-qr { display: none; }
.title-slide .corner-block {
display: block;
width: 56px;
height: 56px;
bottom: clamp(1rem, 4vw, 2rem);
right: clamp(1rem, 4vw, 2rem);
}
/* Hide projector / presenter UI */
.nav-dots { display: none; }
.keyboard-hint { display: none; }
#presenter-notes { display: none; }
}
@media (prefers-reduced-motion: reduce) {
*, *::before, *::after {
animation-duration: 0.01ms !important;
transition-duration: 0.15s !important;
}
html { scroll-behavior: auto; }
#presenter-notes { transition: none; }
}
</style>
</head>
<body>
<div class="progress-bar" id="progress"></div>
<nav class="nav-dots" id="dots" aria-label="Slide navigation"></nav>
<div class="keyboard-hint">
<kbd>←</kbd><kbd>→</kbd> NAV · <kbd>N</kbd> NOTES
</div>
<!-- ===========================================
SLIDE 1 — TITLE
=========================================== -->
<section class="slide title-slide" data-folio="" data-notes="Thank you for having me. I want to walk you through how artificial intelligence actually fits into my thesis — which is a phylogenetic classification of Southern Kurdish dialects — and the workstation I built so the AI stays useful without taking over the linguist's job.">
<div class="slide-content">
<p class="eyebrow reveal">A talk on computer-assisted fieldwork</p>
<h1 class="reveal">AI as a <span class="accent">scout</span>,<br>not a judge.</h1>
<p class="lede reveal">Building a Southern Kurdish dialect tree with PARSE.</p>
<p class="byline reveal"><strong>Lucas Ardelean</strong> · University of Bamberg · University of Zurich · 2026</p>
</div>
<aside class="title-qr" aria-label="Scan to open the presentation on your phone">
<img src="assets/qr-code.png" alt="QR code linking to https://truenorth49.github.io/parse-ai-talk/">
<span class="qr-label">Scan to follow along</span>
</aside>
<div class="corner-block" aria-hidden="true"></div>
</section>
<!-- ===========================================
SLIDE 2 — THE CLASSIFICATION PUZZLE
=========================================== -->
<section class="slide" data-folio="01 · THE PUZZLE" data-notes="Before I introduce anything technical, I want to be clear that this slide describes the research task. The Southern Kurdish dialect classification is what I am writing the thesis on; it is what brought me to Bamberg three years ago. Everything that follows — the AI pipeline, the workstation, the Bayesian inference — exists in service of answering this question. The tool did not come first. Southern Kurdish has been hard to classify for decades. MacKenzie put it in one bucket in 1961; Fattah split it into seven subgroups in 2000; more recent work suggests it may not even be a clean genetic group, but a transitional contact zone. The reason the field hasn't converged is not laziness — it's that different features point in different directions. Phonology groups varieties one way, morphology another. So the question I am asking is not 'what is the right tree', but 'how much support does any tree actually have, given the data?' That framing is what brings probability and computation into the picture.">
<div class="slide-content two-col">
<div class="col-text">
<p class="eyebrow reveal">The classification puzzle</p>
<h2 class="reveal">Sixty years.<br>No consensus.</h2>
<hr class="rule reveal">
<ul class="bullets reveal">
<li><strong>11 speakers, 6 varieties</strong> — Faili, Kalhori, Khanaqini, Qasri, Mandali, Sahana</li>
<li>Iraq–Iran borderlands — a <em>shatter zone</em> of tribal, political, and contact pressure</li>
<li>MacKenzie (1961), Fattah (2000), Belelli (2019), Mohammadirad — no agreement</li>
<li>Isoglosses cross-cut each other; no clean tree by traditional methods</li>
</ul>
</div>
<div class="col-figure reveal">
<figure class="shot">
<img src="assets/sk_map_final.png" alt="Map of Southern Kurdish speaker origins across the Iraq-Iran borderlands">
<figcaption>Speaker origins — Iraq–Iran borderlands</figcaption>
</figure>
</div>
</div>
</section>
<!-- ===========================================
SLIDE 3 — THREE PROBLEMS, ONE JOB
=========================================== -->
<section class="slide" data-folio="02 · THE WRONG SHORTCUT" data-notes="Anyone who has tried to drop a fieldwork recording into a commercial transcription service knows the result. For Southern Kurdish in particular, the contrasts that distinguish one dialect from another fall inside acoustic territory that standard speech recognition was never trained on. The voiceless uvular /q/ collapses to /k/, or sometimes gets deleted altogether. The dark lateral /ɫ/ — which is phonemic in Kurdish, not allophonic like in English — merges into plain /l/. The three-way stop contrast — aspirated, unaspirated, voiced — gets flattened to two, because European models read voice onset time the way English does, and unaspirated stops get heard as voiced. Even the front rounded vowels, when they appear, get unrounded toward /i/ or /e/. So if I let a system transcribe everything for me, the dialect signal I'm trying to measure would be the first thing erased. The second problem is borrowing: the basic vocabulary I need is full of Arabic and Persian loans, and a computer will happily call a borrowing a cognate, because phonetically it is one — it just isn't an inherited one. And the third reality will be familiar to anyone who does fieldwork: a session yields far more than any one question needs. Each speaker produced more than 530 words across the session, but only 85 of those are phylogenetically necessary for this study. The remaining 445 or so are not noise — they are valuable for other research, on phonology or syntax or sociolinguistic variation — they are just not what a phylogeny is asking. So the job for this analysis is not 'transcribe everything'; it is to find the right 85 inside hours of recording where the elicitation order doesn't always follow the wordlist.">
<div class="slide-content">
<p class="eyebrow reveal">Why a "press the button" AI does not work here</p>
<h2 class="reveal">Three problems. One job.</h2>
<hr class="rule reveal">
<ul class="bullets reveal">
<li><strong>Diagnostic contrasts get flattened.</strong> Uvular /q/ collapses to /k/. Dark /ɫ/ merges into plain /l/. Unaspirated stops get heard as voiced. The contrasts that distinguish one Southern Kurdish variety from another are exactly the ones European-trained models were never built to hear.</li>
<li><strong>Borrowing looks like cognacy.</strong> Basic vocabulary is shot through with Arabic and Persian loans. A machine treats <em>waqt</em> as shared inheritance — phonetically it is; historically it is not.</li>
<li><strong>Fieldwork is rich; this question needs a slice.</strong> 530+ words elicited per speaker; only <em>85 are phylogenetically necessary</em>. The other 445+ are valuable for other research — phonology, syntax, sociolinguistic variation — just not what a phylogeny is asking.</li>
<li><strong>The job is not "transcribe everything."</strong> It is to find the right 85, in the right order, with the right confidence.</li>
</ul>
</div>
</section>
<!-- ===========================================
SLIDE 4 — MANIFESTO
=========================================== -->
<section class="slide" data-folio="03 · THE PRINCIPLE" data-notes="The framing I work in is called Computer-Assisted Language Comparison. The idea is very old and very simple: use the machine to scale human expertise, not to substitute for it. So in my pipeline, AI is allowed to read the audio and produce candidate transcriptions — both orthographic and phonemic. Those candidates are saved. But they remain candidates until I explicitly accept them. Every annotation in the final dataset has been reviewed and approved by me. The workstation I'll show you in a moment was built specifically to make that boundary — between candidate and accepted — impossible to cross by accident.">
<div class="slide-content">
<p class="eyebrow reveal" style="text-align:center;">A guiding principle</p>
<p class="pull-quote reveal">The machine narrows the search.<br>The <span class="accent">linguist</span> makes the judgment.</p>
<div class="principles reveal">
<div class="principle">
<span class="num">I.</span>
<h3>AI scouts</h3>
<p>Locates likely regions in long recordings. Offers a candidate transcription.</p>
</div>
<div class="principle">
<span class="num">II.</span>
<h3>The linguist judges</h3>
<p>Every accept, edit, or rejection is an explicit human action.</p>
</div>
<div class="principle">
<span class="num">III.</span>
<h3>Saved as drafts</h3>
<p>Models write candidates, not commitments. Acceptance is an explicit human action.</p>
</div>
<div class="principle">
<span class="num">IV.</span>
<h3>The trail stays visible</h3>
<p>Computed result and human correction stored as separate layers. Auditable forever.</p>
</div>
</div>
</div>
</section>
<!-- ===========================================
SLIDE 5 — PIPELINE
=========================================== -->
<section class="slide" data-folio="04 · THE PIPELINE" data-notes="Here is the whole pipeline on one slide. Audio comes in on the left, and a probability-weighted tree of dialect relationships comes out on the right. There are three places where AI does work in the middle. First, finding where each of the 85 target words occurs in hours of recording. Second, offering a candidate IPA transcription as a second opinion against mine. Third, proposing groupings of forms that might be cognate. Underneath every one of those arrows is a human review step. The next three slides take each one in turn.">
<div class="slide-content">
<p class="eyebrow reveal">Where AI sits in the pipeline</p>
<h2 class="reveal">Four stations. One chain of human review.</h2>
<hr class="rule reveal">
<div class="pipeline reveal">
<div class="station">
<span class="step">Step 1</span>
<h3>Read</h3>
<p>Three models transcribe 2–5 h of audio per speaker — orthographic, Kurdish-script, and phoneme-level IPA.</p>
<span class="who ai">AI</span>
</div>
<div class="station">
<span class="step">Step 2</span>
<h3>Review</h3>
<p>Linguist checks quality, picks the best repetition, splits multi-word responses, re-runs noisy outputs.</p>
<span class="who human">Human</span>
</div>
<div class="station">
<span class="step">Step 3</span>
<h3>Locate</h3>
<p>Use verified anchors plus the transcript layers to find the right 85 target words in 530+.</p>
<span class="who ai">AI</span>
</div>
<div class="station">
<span class="step">Step 4</span>
<h3>Group cognates</h3>
<p>LexStat clusters forms using Levenshtein distance + sound correspondences learned from the data.</p>
<span class="who compute">Computational</span>
</div>
</div>
<div class="review-band reveal">All four outputs feed Bayesian inference → a probability-weighted distribution over candidate trees.</div>
</div>
</section>
<!-- ===========================================
SLIDE 6 — JOB 1: LOCATING
=========================================== -->
<section class="slide" data-folio="05 · JOB ONE" data-notes="Without help, finding the 85 phylogenetically useful items inside thirty hours of conversation — where the order isn't fixed, where the speaker is repeating things, where the translator is interjecting in Arabic or Sorani — is a week of scrubbing through a waveform per speaker. So the pipeline reads the audio three times in parallel. The first model is general Whisper speech-to-text, which gives me an orthographic transcription with word-level timestamps. The second is razhan/whisper-base-sdh — a Whisper variant another researcher fine-tuned specifically for Southern Kurdish — so its vocabulary and acoustic expectations are much closer to the data than a generic model. The third is Wav2Vec2 XLSR-53, which reads the same audio as phonemes rather than words, producing IPA independently of any orthographic decision. Three transcription layers, saved side-by-side, all visible in the workstation. What I do in the workstation is review candidate regions — time spans where the joint signal of these three layers, plus verified cross-speaker references, says a given target word probably lives. I'll come back to the cross-speaker piece in two slides. The result is that finding the right 85 inside 530-plus elicited items is a focused review task rather than a thirty-hour listening task.">
<div class="slide-content two-col">
<div class="col-text">
<p class="eyebrow reveal">Job 1 · Reading the audio</p>
<h2 class="reveal">Three reads<br>of the same audio.</h2>
<hr class="rule reveal">
<ul class="bullets reveal">
<li><strong>Speech-to-text.</strong> Whisper produces orthographic words with word-level timestamps.</li>
<li><strong>Kurdish-tuned.</strong> A Whisper variant fine-tuned for Southern Kurdish returns Kurdish-script transcriptions.</li>
<li><strong>IPA, independently.</strong> A phoneme-level model reads the same audio as IPA — not as words.</li>
<li><strong>Three layers, one waveform.</strong> Finding the right 85 of 530+ words joins these layers with cross-speaker reference data.</li>
</ul>
</div>
<div class="col-figure reveal">
<figure class="shot">
<img src="assets/user-guide-annotate-workstation.png" alt="PARSE Annotate workstation showing waveform with parallel transcription tiers and ranked candidate regions">
<figcaption>Annotate — three layers, one waveform</figcaption>
</figure>
</div>
</div>
</section>
<!-- ===========================================
SLIDE 7 — JOB 2: IPA SECOND OPINION
=========================================== -->
<section class="slide" data-folio="06 · JOB TWO" data-notes="Job 2 is where the linguist does the work AI can't. Once the three transcription layers from Job 1 are loaded into PARSE, I go through each lexeme and make several judgments. First, I check the quality of the IPA and orthographic transcriptions — is what the model heard plausible given the audio? Second, speakers typically produced each target word two to four times in succession, so I pick which repetition to keep as the canonical token — usually the second one, since the first often has hesitation and the third can introduce hyper-articulation. Third, sometimes the speaker gave more than one word for the same prompt — synonyms, near-synonyms, or false starts followed by the real form — so I create additional slots to capture all of them rather than collapse them. Fourth, when the IPA model produced something obviously noisy, I re-run it on that segment with different settings rather than accept a garbled output. And fifth, for ambiguous IPA judgments, I use PARSE's spectrogram view to verify — formants for vowels, voice-onset timing for stops. To be clear about what PARSE does and doesn't do here: my own hand-written field notes from the original elicitation sit separately, on paper. They are a sanity check I bring to the desk; PARSE does not store them and does not compare model output against them automatically. The cross-checking is something I do, not something the system does.">
<div class="slide-content two-col">
<div class="col-text">
<p class="eyebrow reveal">Job 2 · Review and decide</p>
<h2 class="reveal">The work AI<br>can't do.</h2>
<hr class="rule reveal">
<ul class="bullets reveal">
<li><strong>Check transcription quality.</strong> Is what the model heard plausible given the audio?</li>
<li><strong>Pick the repetition.</strong> Each lexeme was elicited 2–4 times; choose which to use.</li>
<li><strong>Split multi-word responses.</strong> Speakers sometimes gave synonyms or false starts — both belong, in their own slots.</li>
<li><strong>Re-run noisy IPA.</strong> If the model output came back garbled, run it again with different settings.</li>
<li><strong>Spectrogram check.</strong> For ambiguous IPA, verify against formants and voice-onset timing.</li>
</ul>
</div>
<div class="col-figure reveal">
<figure class="shot">
<img src="assets/transcription-lanes-row-smoke.png" alt="PARSE transcription lanes showing parallel IPA and orthographic tiers for one speaker">
<figcaption>Parallel IPA and orthography lanes for review</figcaption>
</figure>
</div>
</div>
</section>
<!-- ===========================================
SLIDE 8 — JOB 3: LEXICAL ANCHOR ALIGNMENT
=========================================== -->
<section class="slide" data-folio="07 · JOB THREE" data-notes="A useful property emerged once a few speakers were processed. Because each item is elicited in roughly the same order, and each item is repeated two to four times in succession, I could use the verified data as a map. If I already know where 'water' and 'fire' landed in this recording, then the missing 'sun' is probably right between them. The system formalises that intuition with a scoring formula — phonetic similarity counts most, then repetition, then positional plausibility. The system never assigns anything; it just hands me a top-ranked candidate to check. The honest version of this slide is: the first speaker is the slow one. By the eighth speaker, the work is two or three times faster, because there is so much verified material to compare against.">
<div class="slide-content two-col">
<div class="col-text">