Source file normalizers.ml
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
(** Text normalization module matching HuggingFace tokenizers. *)
type normalized_string = {
normalized : string;
original : string;
alignments : (int * int) array;
}
(** Type representing a normalized string with alignment information *)
(** Main normalizer type - abstract *)
type t =
| Bert of {
clean_text : bool;
handle_chinese_chars : bool;
strip_accents : bool option;
lowercase : bool;
}
| Strip of { left : bool; right : bool }
| StripAccents
| NFC
| NFD
| NFKC
| NFKD
| Lowercase
| Nmt
| Precompiled of bytes
| Replace of { pattern : string; replacement : string }
| Prepend of { prepend : string }
| ByteLevel of { add_prefix_space : bool; use_regex : bool }
| Sequence of t list
(** Helper functions for character checking *)
let is_whitespace c =
match c with
| '\t' | '\n' | '\r' -> true
| c -> Uucp.White.is_white_space (Uchar.of_char c)
let is_control c =
match c with
| '\t' | '\n' | '\r' -> false
| c ->
let u = Uchar.of_char c in
Uucp.Gc.general_category u = `Cc
|| Uucp.Gc.general_category u = `Cf
|| Uucp.Gc.general_category u = `Cn
|| Uucp.Gc.general_category u = `Co
let is_chinese_char c =
let code = Char.code c in
(code >= 0x4E00 && code <= 0x9FFF)
|| (code >= 0x3400 && code <= 0x4DBF)
|| (code >= 0x20000 && code <= 0x2A6DF)
|| (code >= 0x2A700 && code <= 0x2B73F)
|| (code >= 0x2B740 && code <= 0x2B81F)
|| (code >= 0x2B920 && code <= 0x2CEAF)
|| (code >= 0xF900 && code <= 0xFAFF)
|| (code >= 0x2F800 && code <= 0x2FA1F)
let is_combining_mark c =
let u = Uchar.of_char c in
Uucp.Gc.general_category u = `Mn
(** Transform operations on normalized strings *)
let transform_string str transformations =
let buf = Buffer.create (String.length str * 2) in
let alignments = ref [] in
let current_pos = ref 0 in
List.iter
(fun (c, offset) ->
Buffer.add_char buf c;
alignments := (!current_pos, !current_pos + offset) :: !alignments;
current_pos := !current_pos + 1)
transformations;
{
normalized = Buffer.contents buf;
original = str;
alignments = Array.of_list (List.rev !alignments);
}
(** Filter characters from a string *)
let filter_string str pred =
let buf = Buffer.create (String.length str) in
let alignments = ref [] in
let orig_pos = ref 0 in
String.iter
(fun c ->
if pred c then (
Buffer.add_char buf c;
alignments := (!orig_pos, !orig_pos + 1) :: !alignments);
incr orig_pos)
str;
{
normalized = Buffer.contents buf;
original = str;
alignments = Array.of_list (List.rev !alignments);
}
(** Map characters in a string *)
let map_string str f =
let buf = Buffer.create (String.length str) in
let alignments = ref [] in
let pos = ref 0 in
String.iter
(fun c ->
let c' = f c in
Buffer.add_char buf c';
alignments := (!pos, !pos + 1) :: !alignments;
incr pos)
str;
{
normalized = Buffer.contents buf;
original = str;
alignments = Array.of_list (List.rev !alignments);
}
(** Apply Unicode normalization *)
let apply_unicode_normalization form str =
let normalized = Unicode.normalize form str in
let len = String.length normalized in
let alignments = Array.init len (fun i -> (i, i + 1)) in
{ normalized; original = str; alignments }
(** BERT text cleaning *)
let do_clean_text ns =
let ns' =
filter_string ns.normalized (fun c ->
let code = Char.code c in
not (code = 0 || code = 0xfffd || is_control c))
in
map_string ns'.normalized (fun c -> if is_whitespace c then ' ' else c)
(** Handle Chinese characters *)
let do_handle_chinese_chars ns =
let transformations = ref [] in
String.iter
(fun c ->
if is_chinese_char c then
transformations := (' ', 0) :: (c, 1) :: (' ', 1) :: !transformations
else transformations := (c, 0) :: !transformations)
ns.normalized;
transform_string ns.original (List.rev !transformations)
(** Strip accents *)
let do_strip_accents ns =
let ns = apply_unicode_normalization Unicode.NFD ns.normalized in
filter_string ns.normalized (fun c -> not (is_combining_mark c))
(** Lowercase *)
let do_lowercase ns = map_string ns.normalized Char.lowercase_ascii
(** Strip whitespace *)
let strip_whitespace ns ~left ~right =
let s = ns.normalized in
let len = String.length s in
let start =
if left then
let rec find_start i =
if i >= len then len
else if is_whitespace s.[i] then find_start (i + 1)
else i
in
find_start 0
else 0
in
let stop =
if right then
let rec find_stop i =
if i < 0 then 0
else if is_whitespace s.[i] then find_stop (i - 1)
else i + 1
in
find_stop (len - 1)
else len
in
let normalized = String.sub s start (stop - start) in
let alignments = Array.sub ns.alignments start (stop - start) in
{ normalized; original = ns.original; alignments }
(** Apply regex replacement *)
let apply_replace ns ~pattern ~replacement =
let regex = Str.regexp pattern in
let normalized = Str.global_replace regex replacement ns.normalized in
let len = String.length normalized in
let alignments =
Array.init len (fun i -> (i, min i (String.length ns.original - 1)))
in
{ normalized; original = ns.original; alignments }
(** Prepend string *)
let prepend_string ns ~prepend =
if String.length ns.normalized = 0 then ns
else
let prepend_len = String.length prepend in
let normalized = prepend ^ ns.normalized in
let alignments =
Array.concat
[
Array.make prepend_len (0, 1);
Array.map
(fun (a, b) -> (a + prepend_len, b + prepend_len))
ns.alignments;
]
in
{ normalized; original = ns.original; alignments }
(** Byte-level encoding *)
let bytes_char =
let chars = ref [] in
for i = 0 to 255 do
let c =
if i < 0x21 || (i >= 0x7F && i <= 0xA0) || i = 0xAD then
Char.chr
(if i < 0x21 then i + 0x41
else if i >= 0x7F && i <= 0x9F then i - 0x7F + 0x41
else 0x41)
else Char.chr i
in
chars := (i, c) :: !chars
done;
List.rev !chars
let apply_byte_level ns ~add_prefix_space ~use_regex:_ =
let bytes_map = bytes_char in
let find_char b = List.assoc b bytes_map in
let s =
if
add_prefix_space
&& String.length ns.normalized > 0
&& ns.normalized.[0] <> ' '
then " " ^ ns.normalized
else ns.normalized
in
let transformations = ref [] in
String.iter
(fun c ->
let byte = Char.code c in
transformations := (find_char byte, 0) :: !transformations)
s;
transform_string ns.original (List.rev !transformations)
(** NMT normalization *)
let do_nmt ns =
let ns' =
filter_string ns.normalized (fun c ->
let code = Char.code c in
not
((code >= 0x0001 && code <= 0x0008)
|| code = 0x000B
|| (code >= 0x000E && code <= 0x001F)
|| code = 0x007F || code = 0x008F || code = 0x009F))
in
map_string ns'.normalized (fun c ->
match Char.code c with
| 0x0009 | 0x000A | 0x000C | 0x000D | 0x1680 | 0x2028 | 0x2029 | 0x2581
| 0xFEFF | 0xFFFD ->
' '
| 0x200B | 0x200C | 0x200D | 0x200E | 0x200F -> ' '
| _ -> c)
(** Main normalization function *)
let rec normalize_impl normalizer ns =
match normalizer with
| Bert { clean_text; handle_chinese_chars; strip_accents; lowercase } ->
let ns = if clean_text then do_clean_text ns else ns in
let ns =
if handle_chinese_chars then do_handle_chinese_chars ns else ns
in
let strip_accents_val =
match strip_accents with Some v -> v | None -> lowercase
in
let ns = if strip_accents_val then do_strip_accents ns else ns in
let ns = if lowercase then do_lowercase ns else ns in
ns
| Strip { left; right } -> strip_whitespace ns ~left ~right
| StripAccents -> do_strip_accents ns
| NFC -> apply_unicode_normalization Unicode.NFC ns.normalized
| NFD -> apply_unicode_normalization Unicode.NFD ns.normalized
| NFKC -> apply_unicode_normalization Unicode.NFKC ns.normalized
| NFKD -> apply_unicode_normalization Unicode.NFKD ns.normalized
| Lowercase -> do_lowercase ns
| Nmt -> do_nmt ns
| Precompiled _bytes ->
ns
| Replace { pattern; replacement } -> apply_replace ns ~pattern ~replacement
| Prepend { prepend } -> prepend_string ns ~prepend
| ByteLevel { add_prefix_space; use_regex } ->
apply_byte_level ns ~add_prefix_space ~use_regex
| Sequence normalizers ->
List.fold_left (fun ns n -> normalize_impl n ns) ns normalizers
(** Constructors *)
let bert ?(clean_text = true) ?(handle_chinese_chars = true)
?(strip_accents = None) ?(lowercase = true) () =
Bert { clean_text; handle_chinese_chars; strip_accents; lowercase }
let strip ?(left = false) ?(right = true) () = Strip { left; right }
let strip_accents () = StripAccents
let nfc () = NFC
let nfd () = NFD
let nfkc () = NFKC
let nfkd () = NFKD
let lowercase () = Lowercase
let nmt () = Nmt
let precompiled bytes = Precompiled bytes
let replace ~pattern ~replacement () = Replace { pattern; replacement }
let prepend ~prepend = Prepend { prepend }
let byte_level ?(add_prefix_space = false) ?(use_regex = false) () =
ByteLevel { add_prefix_space; use_regex }
let sequence normalizers = Sequence normalizers
(** Operations *)
let normalize t str =
let initial =
{
normalized = str;
original = str;
alignments = Array.init (String.length str) (fun i -> (i, i + 1));
}
in
normalize_impl t initial
let normalize_str t str =
let ns = normalize t str in
ns.normalized
(** Serialization *)
let rec to_json = function
| Bert { clean_text; handle_chinese_chars; strip_accents; lowercase } ->
`Assoc
[
("type", `String "Bert");
("clean_text", `Bool clean_text);
("handle_chinese_chars", `Bool handle_chinese_chars);
( "strip_accents",
match strip_accents with None -> `Null | Some b -> `Bool b );
("lowercase", `Bool lowercase);
]
| Strip { left; right } ->
`Assoc
[
("type", `String "Strip");
("strip_left", `Bool left);
("strip_right", `Bool right);
]
| StripAccents -> `Assoc [ ("type", `String "StripAccents") ]
| NFC -> `Assoc [ ("type", `String "NFC") ]
| NFD -> `Assoc [ ("type", `String "NFD") ]
| NFKC -> `Assoc [ ("type", `String "NFKC") ]
| NFKD -> `Assoc [ ("type", `String "NFKD") ]
| Lowercase -> `Assoc [ ("type", `String "Lowercase") ]
| Nmt -> `Assoc [ ("type", `String "Nmt") ]
| Precompiled bytes ->
`Assoc
[
("type", `String "Precompiled");
("data", `String (Bytes.to_string bytes));
]
| Replace { pattern; replacement } ->
`Assoc
[
("type", `String "Replace");
("pattern", `Assoc [ ("String", `String pattern) ]);
("content", `String replacement);
]
| Prepend { prepend } ->
`Assoc [ ("type", `String "Prepend"); ("prepend", `String prepend) ]
| ByteLevel { add_prefix_space; use_regex } ->
`Assoc
[
("type", `String "ByteLevel");
("add_prefix_space", `Bool add_prefix_space);
("use_regex", `Bool use_regex);
]
| Sequence normalizers ->
`Assoc
[
("type", `String "Sequence");
("normalizers", `List (List.map to_json normalizers));
]
let rec of_json = function
| `Assoc fields -> (
match List.assoc_opt "type" fields with
| Some (`String "Bert") ->
let get_bool name default =
match List.assoc_opt name fields with
| Some (`Bool b) -> b
| _ -> default
in
let strip_accents =
match List.assoc_opt "strip_accents" fields with
| Some (`Bool b) -> Some b
| Some `Null | None -> None
| _ -> None
in
Bert
{
clean_text = get_bool "clean_text" true;
handle_chinese_chars = get_bool "handle_chinese_chars" true;
strip_accents;
lowercase = get_bool "lowercase" true;
}
| Some (`String "Strip") ->
let get_bool name default =
match List.assoc_opt name fields with
| Some (`Bool b) -> b
| _ -> default
in
Strip
{
left = get_bool "strip_left" false;
right = get_bool "strip_right" true;
}
| Some (`String "StripAccents") -> StripAccents
| Some (`String "NFC") -> NFC
| Some (`String "NFD") -> NFD
| Some (`String "NFKC") -> NFKC
| Some (`String "NFKD") -> NFKD
| Some (`String "Lowercase") -> Lowercase
| Some (`String "Nmt") -> Nmt
| Some (`String "Precompiled") -> (
match List.assoc_opt "data" fields with
| Some (`String s) -> Precompiled (Bytes.of_string s)
| _ -> failwith "Invalid Precompiled normalizer")
| Some (`String "Replace") ->
let pattern =
match List.assoc_opt "pattern" fields with
| Some (`Assoc [ ("String", `String s) ]) -> s
| Some (`String s) -> s
| _ -> failwith "Invalid Replace pattern"
in
let replacement =
match List.assoc_opt "content" fields with
| Some (`String s) -> s
| _ -> ""
in
Replace { pattern; replacement }
| Some (`String "Prepend") ->
let prepend =
match List.assoc_opt "prepend" fields with
| Some (`String s) -> s
| _ -> failwith "Invalid Prepend configuration"
in
Prepend { prepend }
| Some (`String "ByteLevel") ->
let get_bool name default =
match List.assoc_opt name fields with
| Some (`Bool b) -> b
| _ -> default
in
ByteLevel
{
add_prefix_space = get_bool "add_prefix_space" false;
use_regex = get_bool "use_regex" false;
}
| Some (`String "Sequence") -> (
match List.assoc_opt "normalizers" fields with
| Some (`List normalizers) -> Sequence (List.map of_json normalizers)
| _ -> failwith "Invalid Sequence normalizer")
| _ -> failwith "Unknown normalizer type")
| _ -> failwith "Invalid normalizer JSON"