http://www.asks.jp/users/hiro/59059.html http://www.itmedia.co.jp/news/articles/0905/08/news021.html 最初読んだとき、違和感なく読めてしまったのですが、よくよく見てみると、そんなトリックがあったのですね。 さて、この「読めてしまう」がなぜよめてしまうのでしょうか？ 人間の言語モデルの単語パープレキシティは、約100ぐらいであると言われています。どういうことかというと、 人間が文章を読んでいるときに、次の単語を過去の文章から推測するのは 1/100 程度の 確率で正解するということです。 件のコピペですが、最初の文字は変わらないので、その正解率は平仮名の数(52)倍になります。 すなわち、52/100 =~ 0.5 実際には、最後の文字も変わらないし、 単語の長さが変わらないというもの、大きな
タグ
- すべて
- 2009 (36)
- 2010 (34)
- 2011 (20)
- 2ch (28)
- 3d (72)
- API (163)
- AR (106)
- Adobe (105)
- Ajax (229)
- Amazon (268)
- Bluetooth (17)
- CGM (54)
- CM (74)
- CRM (36)
- CSS (255)
- DNP (32)
- DNS (29)
- Disney (19)
- DoCoMo (79)
- English (281)
- Gmail (65)
- Google (1411)
- Groupon (73)
- HTML (178)
- IBM (44)
- IC (18)
- IE (79)
- Illustrator (17)
- Intel (30)
- JSON (27)
- Java (124)
- M&A (82)
- MIT (23)
- Mac (213)
- MacOS (17)
- MySQL (172)
- NTT (19)
- Nintendo (47)
- Oracle (17)
- PHP (240)
- Perl (200)
- Python (46)
- RSS (71)
- SEM (72)
- SEO (283)
- SF (32)
- SteveJobs (40)
- TOEIC (17)
- UI (306)
- USA (49)
- USB (28)
- Wiki (18)
- Winny (55)
- XML (34)
- academic (177)
- accounting (21)
- actionscript (63)
- ad (19)
- addon (122)
- adsense (21)
- advertisement (759)
- affiliate (50)
- agile (18)
- air (25)
- algorithm (154)
- ameba (29)
- america (93)
- analysis (91)
- analytics (65)
- android (295)
- animal (34)
- anime (59)
- apache (116)
- app (415)
- apple (556)
- application (46)
- architecture (115)
- art (131)
- article (72)
- astronomy (55)
- au (42)
- auction (21)
- audio (23)
- authentication (65)
- aws (33)
- biology (66)
- blog (347)
- book (1720)
- bookmark (29)
- bookmarklet (18)
- bot (22)
- brain (53)
- branding (88)
- browser (292)
- business (4114)
- c (29)
- ca (18)
- calendar (31)
- camera (56)
- car (66)
- career (239)
- ceo (23)
- cg (35)
- charcode (21)
- chat (22)
- cheatsheet (24)
- china (97)
- chrome (134)
- cinema (50)
- cloud (151)
- cms (34)
- code (20)
- coffee (18)
- color (50)
- column (52)
- comic (69)
- command (45)
- communication (456)
- community (69)
- company (367)
- computer (74)
- conference (62)
- contents (89)
- cook (54)
- cookie (24)
- cooking (30)
- cookpad (44)
- copyright (171)
- coupon (73)
- crawler (18)
- creative (86)
- crime (29)
- css3 (23)
- culture (268)
- data (136)
- database (33)
- db (221)
- debug (86)
- decentralize (34)
- dena (91)
- design (1100)
- desktop (18)
- developer (27)
- development (1340)
- device (21)
- disaster (90)
- display (58)
- document (148)
- dropbox (21)
- dvd (24)
- earthquake (63)
- ebook (209)
- ec (376)
- eclipse (24)
- eco (32)
- ecology (44)
- economics (197)
- economy (167)
- editor (51)
- education (304)
- emacs (23)
- employment (208)
- encode (24)
- encryption (22)
- energy (156)
- engineer (117)
- entertainment (34)
- entrepreneur (22)
- environment (48)
- essay (34)
- event (350)
- evernote (76)
- excel (32)
- experience (30)
- extension (135)
- facebook (538)
- fashion (165)
- feed (24)
- felica (37)
- filesystem (19)
- finance (376)
- firebug (23)
- firefox (316)
- flash (291)
- flex (20)
- flickr (24)
- flyer (22)
- font (56)
- food (339)
- format (24)
- foursquare (37)
- framework (108)
- free (80)
- freesoft (69)
- fun (20)
- funny (29)
- gadget (198)
- game (549)
- geek (18)
- generation (44)
- generator (46)
- geo (170)
- gigazine (85)
- gis (136)
- git (55)
- global (18)
- goods (52)
- gourmet (135)
- government (80)
- gps (61)
- graph (29)
- graphic (33)
- greasemonkey (49)
- gree (87)
- hack (20)
- hadoop (35)
- hardware (125)
- hatena (326)
- hdd (49)
- health (145)
- history (245)
- html5 (156)
- http (52)
- iOS (32)
- iPhone (974)
- iPod (26)
- iTunes (26)
- idc (32)
- idea (294)
- igo (29)
- image (214)
- industry (41)
- infrastructure (87)
- innovation (40)
- interface (123)
- interior (47)
- international (158)
- internet (325)
- interview (406)
- investment (93)
- ipad (195)
- ipo (32)
- it (303)
- item (205)
- itmedia (32)
- japan (513)
- japanese (63)
- javascript (848)
- job (64)
- journalism (31)
- jquery (140)
- kayac (19)
- kindle (34)
- knowledge (26)
- lab (24)
- language (125)
- law (169)
- learning (59)
- library (324)
- license (44)
- life (2437)
- lifehack (251)
- link (44)
- linux (291)
- literacy (47)
- literature (48)
- livedoor (82)
- local (113)
- location (19)
- love (57)
- magazine (73)
- mail (180)
- management (440)
- map (245)
- mapion (18)
- mapreduce (19)
- market (90)
- marketing (1241)
- mashup (40)
- material (61)
- math (111)
- mba (22)
- media (1011)
- medical (99)
- memcached (30)
- memo (45)
- micropayment (30)
- microsoft (274)
- military (31)
- mixi (185)
- mobile (1361)
- money (394)
- movie (191)
- music (193)
- nanapi (22)
- nature (35)
- neta (1057)
- network (266)
- news (72)
- nhk (23)
- nicovideo (107)
- nlp (39)
- novel (22)
- nuclear (66)
- office (140)
- os (125)
- oss (93)
- p2p (85)
- paper (46)
- password (26)
- patent (80)
- payment (87)
- pc (448)
- pdf (50)
- performance (219)
- person (242)
- phishing (38)
- photo (341)
- photoshop (33)
- physics (51)
- picture (22)
- plugin (29)
- podcast (23)
- politics (276)
- portal (21)
- presentation (205)
- printing (28)
- privacy (118)
- product (73)
- programming (1310)
- project (56)
- prototype.js (21)
- psychology (64)
- publishing (338)
- purchase (36)
- rails (119)
- rakuten (161)
- ranking (64)
- reading (82)
- recipe (47)
- recognition (27)
- recommendation (18)
- recruit (52)
- reference (98)
- report (47)
- research (335)
- retail (60)
- review (127)
- robot (30)
- ruby (209)
- salary (42)
- sbm (69)
- scalability (32)
- science (591)
- search (538)
- security (1128)
- server (385)
- service (751)
- share (25)
- shell (47)
- shopping (65)
- signage (78)
- siliconvalley (52)
- skype (24)
- slide (19)
- smartphone (121)
- sns (520)
- social (596)
- socialmedia (135)
- society (563)
- softbank (110)
- software (525)
- sony (72)
- space (29)
- spam (71)
- speech (81)
- speechrecognition (26)
- sports (39)
- sql (49)
- startup (411)
- statistics (115)
- stock (111)
- storage (130)
- story (34)
- strategy (153)
- study (364)
- subversion (25)
- summary (99)
- system (23)
- tax (46)
- technology (836)
- template (24)
- test (68)
- text (35)
- thinking (86)
- tips (2512)
- tokyo (67)
- tool (729)
- toppan (21)
- translation (37)
- travel (128)
- trend (19)
- trouble (85)
- tumblr (21)
- tuning (32)
- tutorial (169)
- tv (257)
- twitter (655)
- unicode (22)
- uniqlo (31)
- university (46)
- unix (49)
- usability (214)
- ux (51)
- venture (62)
- video (372)
- virus (44)
- vocaloid (20)
- vulnerability (62)
- web (4621)
- web2.0 (94)
- webapi (71)
- webdesign (361)
- webservice (1277)
- wifi (28)
- windows (345)
- wireless (79)
- word (26)
- work (818)
- world (46)
- writing (128)
- xss (39)
- yahoo (248)
- youtube (136)
- web (4621)
- business (4114)
- tips (2512)
- life (2437)
- book (1720)
- Google (1411)
- mobile (1361)
- development (1340)
- programming (1310)
- webservice (1277)
literatureとprogrammingに関するHeavyFeatherのブックマーク (1)
-
HeavyFeather 2009/05/13
パープレキシティの問題で、機械にも高い精度で元文章を推測可能。- nlp
- programming
- literature
- language
- algorithm
リンク -
1
公式Twitter
- @HatenaBookmark
リリース、障害情報などのサービスのお知らせ
- @hatebu
最新の人気エントリーの配信
処理を実行中です
キーボードショートカット一覧
j次のブックマーク
k前のブックマーク
lあとで読む
eコメント一覧を開く
oページを開く
設定を変更しましたx