[2010-03-04-1]ε#311. girl Ȥ褯 collocate ƻϲפǡȸζ (collocation) ¬ˡ (association measure) ˤϤĤμब뤳ȤѥؤǤϡLog-Likelihood Test ȤˤˡŪ褯ȤƤ뤬줾ηˡˤħΤǡʤ٤ʣˡΤ褤[2010-03-04-1]ƤȽʣʬ⤢뤬BNCweb ǼƤ7ηˡγơˤĤ Hoffmann et al. (149--58) Ȥʤ顤ħѤΥҥȤ
Ƽηˡϡ(a) (frequency of co-occurrence)(b) ͭ (significance of co-occurrence)(c) եȡ (effect-size) 1ġ뤤ʣȤ߹碌˴ŤƤ롥(b) ϡŪͭդǤȤγο٤ɽ魯ɸǤꡤζɽ魯ΤǤϤʤȤդɬפ롥(c) ϡѻ٤ȴ٤ȤδܤȤɸǤ롥
(1) Rank by frequency
ѻ붦٤ΤΤѤ롤ǤñľŪʼ١¾ηˡΤ褦ʣϤۤɤƤ餺ɸȤƤϺǤƤǽɵʤɤ̤뤳Ȥ¿̾ζʬϤˤѤʤ
(2) Log-likelihood
ͭѤ롥BNCweb ΥǥեȤηˡǡѥǹѤƤ롥ǽɵʤɤζˤƹ٤θȤζ䡤դ˶ˤ٤θ1, 2ʤɡˤȤζϤ롥٤ι⤤Ȥ߹碌˹ͿȤħꡤˤդפ롥
(3) Mutual information (MI)
եȡѤ롥ˤ褯ѤƤˡѤäƤ¿դפ롥ǽɵʤɤȤΤդ줿ŪӽƤϤ褤ȿ̡٤ζɽؤФ꤬㤷Фαƶ뤿ˡBNCweb Ǥ "Freq(node, collocate) at least" 10ʾꤹ뤳Ȥ侩롥ˤꡤ"conspicuous and intuitively appealing collocations involving words of intermediate frequency" (Hoffmann et al. 154) ⤭ĦȤʤ롥
(4) T-score
٤ȶͭθˡ٤1ʲ٤εʶɽˤĤƤ Rank by frequency Ȼ褦ʿ٤ι⤤ɽˤĤƤ϶ͭȿǤ롥ޤѻ٤٤ɬ⤯ʤ롥Log-likelihood ̤Ȥʤ뤳Ȥ¿٤ؤΥХϰضʤ롥ΡɤΤΤ1000礭ˡ̤ȯ뤳Ȥ롥
(5) Z-score
ͭȥեȡθˡ٤ζɽˤϥեȡŻ뤹뤬٤ζɽˤϤޤǥեȡ˴꤫ʤLog-likelihood MI ξħ褦ʡХμ줿ɸǤ롥MI Ʊͤˡ٤ζɽؤΥХߤΤǡ"Freq(node, collocate) at least" 5٤ꤹΤ褤Ȥ롥
(6) MI3
٤ȥեȡθˡMI ΤɽؤнŤ٤Ƥ롥ٶɽˤϥեȡٶɽˤ϶٤Ū褯ȿǤ롥POS ˤȤȤѤȸŪʣ줫ʤѸʤɤμФ˰Ϥȯ롥ΤȤƤϹٶɽؤΥХŪʶʬϤˤϸʤ
(7) Dice coefficient
MI3 Ʊͤˡ٤ȥեȡθˡMI3Ȱۤʤꡤٶɽˤ϶٤ٶɽˤϥեȡ褯ȿǤ졤ξԤڤؤޤʤΤħŪǤ롥ڤؤϡΡɤΤΤ٤ɽ٤10ܤۤɤǵȤ롥иŪˡZ-score Ȼ褦ʷ̤뤬Z-score ۤ٤˴ŤХʤ
ʾΤ褦¿ढäܰܤꤹ뤬Hoffmann et al. θˤСñηˡȤƤ Log-likelihood MI ǡηˡȤƤ Z-score Dice ȤΤȤǤ롥
͡ʷˡˤĤƤϡAssociation measures ȡ
Hoffmann, Sebastian, Stefan Evert, Nicholas Smith, David Lee, and Ylva Berglund Prytz. Corpus Linguistics with BNCweb : A Practical Guide. Frankfurt am Main: Peter Lang, 2008.
ѥؤߤޤʤʬʬˡϢϥ־뤳Ȥ¿ʤ褦ˤפ뤬դ˾¿ơȽǤ˺롥ƼʬΤǤʥޤȤƤȻפΤسΥԡɤˤĤƹԤʤ䤬Ǥ褯Ѥ BNC ˴ϢΤ濴ˡŪǤϤ뤬ĥ롥ޤȤϫϡŤ뼰ˤɤ뤫ɸΤۤΨŪȤˤʤĤĤ롦
1. BNC ե
BNCweb ̵Ͽ
BYU-BNC ̵Ͽ
BNC ( The British National Corpus )
2. BNC Υե
Quick Reference for Simple Query Syntax (PDF)
Reference Guide for the British National Corpus (XML Edition)
Reference Guide ܼ
* 6.5 Guidelines to the Wordclass Tagging
* The BNC Basic (C5) Tagset
* 9.8 Simplified Wordclass Tags
* 9.7 Contracted forms and multiwords
* 1 Design of the Corpus
* 9.6 Text and genre classification code
3. ѥϢ祵
David Lee ˤ Bookmarks for Corpus-based Linguists
* Corpora, Collections, Data Archives
* Software, Tools, Frequency Lists, etc.
* References, Papers, Journals
* Conferences & Project
4. hellog ε
#568. ѥȱѸ쥳ѥ: [2010-11-16-1]
#506. CoRD --- Ѹ˥ѥξ: [2010-09-15-1]
#308. Ѹκѱñꥹȡ: [2010-03-01-1]
ѥϢ: corpus
BNC Ϣ: bnc
COCA Ϣ: coca
5. ġ
Corpus Frequency Wizard
Paul Rayson's Log-likelihood Calculator
VassarStats
hellog Ρ#711. Log-Likelihood Tester CGI, Ver. 2: [2011-04-08-1]
Hoffmann, Sebastian, Stefan Evert, Nicholas Smith, David Lee, and Ylva Berglund Prytz. Corpus Linguistics with BNCweb : A Practical Guide. Frankfurt am Main: Peter Lang, 2008.
ε#913. BNC ˤä˽Ĵ ([2011-10-27-1]) Ǽꤢ Rayson et al. ǤϡüԤ̤ǤʤǯˤäѰۤĴƤ롥ǯȤäƤ⡤35̤ʾ夫Ǿ岼ʬ绨Ĥʬ̤ϤĤζ̣ͿƤ롥ʲϡχ2 ξ19̤ޤǤΰǤ (142--43)
| Rank | Under 35 | Over 35 | ||
| Word | χ2 | Word | χ2 | |
| 1 | mum | 1409.3 | yes | 2365.0 |
| 2 | fucking | 1184.6 | well | 1059.8 |
| 3 | my | 762.4 | mm | 895.2 |
| 4 | mummy | 755.2 | er | 773.8 |
| 5 | like | 745.2 | they | 682.2 |
| 6 | na as in wanna and gonna | 712.8 | said | 538.3 |
| 7 | goes | 606.6 | says | 443.1 |
| 8 | shit | 410.1 | were | 385.8 |
| 9 | dad | 403.7 | the | 352.2 |
| 10 | daddy | 380.1 | of | 314.6 |
| 11 | me | 371.9 | and | 224.7 |
| 12 | what | 357.3 | to | 211.2 |
| 13 | fuck | 330.1 | mean | 155.0 |
| 14 | wan as in wanna | 320.6 | he | 144.0 |
| 15 | really | 277.0 | but | 139.0 |
| 16 | okay | 257.0 | perhaps | 136.0 |
| 17 | cos | 254.4 | that | 131.3 |
| 18 | just | 251.8 | see | 122.1 |
| 19 | why | 240.0 | had | 118.3 |
ɸä Rayson et al. ʸɤBNC ǡŪʴʬव줿äդϿ֥ѥ4,552,555ˤоݤȤơä˽ǯҲŪϰ̤ˤ뺹餫ˤ褦ȤǤ롥װΤʤǡŪѰۤŪ˺Ǥ줿Τˤ뺹äȤȤʤΤǡܵǤϤη̤Ҳ𤷤
ޤʲ˵ͤβˤμɬפʤΤǡ˿ƤBNC ˼Ͽ줿äդϻִԤ2֤μʲä Walkman ˿ǤäǡΤǤꡤλִԤ73̾75̾Ǥ롥äо줹ִʳüԤˤĤƤ⡤Τۤ¿äơ֥ѥؤλΨǤСΤȤƽ⤯ʤ뤳ȤԻĤǤϤʤ
ƧޤǤ⡤ΤȤƽΤۤ褯äȤȤͤФѤ줿 word token ǤС1.00ȤȽ1.51äͭΨǤϡ1.00ȤȽ1.33ä˽βäǤΤۤ⤤ͭΨȤԸ椬뤬BNC Υ֥ѥǤϽƱΤβä¿äȤȤ嵭η̤طʤˤΤ⤷ʤˤ衤̣ͤǤ뤳Ȥϴְ㤤ʤ
ˡ٤äˤ˽Ƥߤ褦˽ٹ礤ι⤤ɤȴФˡϡȤƤ[2010-03-10-1], [2010-09-27-1], [2011-09-24-1]εǾҲ𤷤ΤƱˡǤ롥ѥȽѥ̤줾줫줿ɽͤ碌Ū˽ (χ2) ι⤤¤ؤФ褤ʲϡ25̤ޤǤΰǤ (136--37)
| Rank | Characteristically male | Characteristically female | ||
| Word | χ2 | Word | χ2 | |
| 1 | fucking | 1233.1 | she | 3109.7 |
| 2 | er | 945.4 | her | 965.4 |
| 3 | the | 698.0 | said | 872.0 |
| 4 | year | 310.3 | n't | 443.9 |
| 5 | aye | 291.8 | I | 357.9 |
| 6 | right | 276.0 | and | 245.3 |
| 7 | hundred | 251.1 | to | 198.6 |
| 8 | fuck | 239.0 | cos | 194.6 |
| 9 | is | 233.3 | oh | 170.2 |
| 10 | of | 203.6 | Christmas | 163.9 |
| 11 | two | 170.3 | thought | 159.7 |
| 12 | three | 168.2 | lovely | 140.3 |
| 13 | a | 151.6 | nice | 134.4 |
| 14 | four | 145.5 | mm | 133.8 |
| 15 | ah | 143.6 | had | 125.9 |
| 16 | no | 140.8 | did | 109.6 |
| 17 | number | 133.9 | going | 109.0 |
| 18 | quid | 124.2 | because | 105.0 |
| 19 | one | 123.6 | him | 99.2 |
| 20 | mate | 120.8 | really | 97.6 |
| 21 | which | 120.5 | school | 96.3 |
| 22 | okay | 119.9 | he | 90.4 |
| 23 | that | 114.2 | think | 88.8 |
| 24 | guy | 108.6 | home | 84.0 |
| 25 | da | 105.3 | me | 83.5 |
ʲˡѤ Log-Likelihood Tester, Ver. 2 ʸ褦ˡϥǡΥեޥåȤ䡤⡼ɤŬڤƤʤˤϥСǥ顼ǽΤա
| though | although | |
|---|---|---|
| Natural and pure sciences | 56.3 | 80.13 |
| Applied science | 37.36 | 68.31 |
| World affairs | 45.81 | 68.2 |
| Social science | 48.98 | 63.38 |
| Commerce and finance | 46.18 | 57.21 |
| Arts | 74.07 | 52.93 |
| Leisure | 45.85 | 49.46 |
| Belief and thought | 70.78 | 46.75 |
| Imaginative prose | 80.2 | 26.37 |
ε[2011-04-06-1]ǡthough although θˡκ˿줿Ʊǡ
4000Ķʤ The Longman Spoken and Written English Corpus (the LSWE Corpus) ȤѸʸˡBiber et al. (845--46) ǤϼΤ褦ˤ롥
Both of these subordinators [though and although] occur in all four registers [conversation, fiction, news, and academic prose], although the registers show different preferences of use. Conversation and fiction show a slightly greater use of though (concessive clauses are, however, uncommon in conversation generally). News shows no particular preference. In academic prose, although is about three times as frequent as though. Although seems to have a slightly more formal tone to it, fitting the style of academic prose . . . . The greater use of although by writers of academic prose may also result from an attempt to distinguish this subordinator from the common use of though as a linking adverbial in conversation . . . .
ޤƱ p. 842 ɽϡŪ though fiction ¿although academic prose ¿Ȥǧ롥ˤ뺹ƤȤη̤
Τ褦Ըơ BNC ( The British National Corpus ) ˤꤳΤƤߤ롥BNCweb ǡ{although/CONJ}, {though/CONJ} 줾측Written/Spoken, Text Domain, Sex of Author/Speaker, Perceived Level of Difficulty ʤ͡ʥѥǽиʬۤʬϤΩä̤ʲ˼ʿͥǡϤΥڡHTMLȡˡ
ޤWritten/Spoken κˤĤƤϡͽۤȤꡤξȤ Written ؤФ꤬㤷ʺ۷ though 0.66344 although 0.49770 ǡ餫˽դФˡLog-Likelihood Test Ǥϡp < 0.0001 Υ٥ǽդäդͭպΤ˼줿
ꡤäˤ뺹ⶽ̣դäդξǡalthough ͭպäλѤФäƤ롥though ˤĤƤϡ although ۤɸǤϤʤʤդǤ p < 0.05 ͭպˡ
ˡText Domain ̤٤ߤ롥9 Text Domain ̤ ( Natural and pure sciences, Applied science, World affairs, Social science, Commerce and finance, Arts, Leisure, Belief and thought, Imaginative prose ) 100νиɸಽͤǡξ Text Domain ٤ղΤʲοޤ

Text Domain ˤäξνи٤оŪʷ뤳Ȥ狼롥Ū sciences ( = academic prose ) although ΩImag(inative) Prose ( = fiction ) though ¿Log-Likelihood Test ǤϡText Domain ˤиκ p < 0.0001 ͭդǤ롥
ľŪˤԸη̤ͽۤȤǤϤ뤬although νˤؽѻʸǸѤȤ줿
Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad, and Edward Finegan. Longman Grammar of Spoken and Written English. Harlow: Pearson Education, 1999.
ε[2011-03-24-1] Log-Likelihood Test ˤˤ Rayson Log-likelihood calculator ѤФ褤ȽҤ٤ºݤθκݤ˺Ȥ⤦ưȻפäΤ CGI Ƥߤ٤ϤȻפȤꤢ
BNC_Male_Speakers BNC_Female_Speakers new 149 91 good 408 310 free 173 75 fresh 84 118 delicious 12 34 full 210 107 sure 532 328 clean 197 223 wonderful 270 258 special 177 82 crisp 10 16 fine 347 215 big 470 415 great 203 96 real 163 80 easy 326 157 bright 113 110 extra 347 203 safe 182 92 rich 120 45 #-------- corpus_size 4949938 3290569
˽֤ͭպä礭ΤϡбԤ֤ɤĤ֤줿 fresh, delicious, clean, wonderful, big ǡٿ˴ŤƷ줿 Diff_Co ( "Difference Coefficient" ֺ۷ ) ޥʥǤ뤳Ȥ顤ħŪʷƻȤȤˤʤ롥big ϰճʵ⤷̤Ǥ롥ФäͭպΤϲǼ easy rich Ǥ롥η̤Ϥɤ߹ळȤǤܺ٤Ĵ٤뤳ȤǤ롥ηƻȤϡüԤǤϤʤʹ̡ǯ𡤼ҲʤɤĴƤ⤪⤷ȱѤǤ롥
[2010-03-04-1]εǿ줿ѥؤǤϳƼˡѤ롥ĤˡΤʤǤ⡤ɽΥѥ֤٤Ӥꡤcollocation ٹ礤¬Τ˹ѤƤΤ Log-Likelihood Test ( LL Test, G Test, G2 Test ʤɤȤ˸ƤФ븡Ǥ롥ѥθ줿ʤΤǥΰۤʤ륳ѥ֤ǤӤǽǤꡤƱŪǰˤ褯ѤƤ2踡 ( Chi-Squared Test ) ⤤ĤǤ줿ˡɾƤꡤǶΥѥǤϹѤƤ롥㤨С2踡ϴ٤5꾯ʤȤٸȤѥ礭ΤȾΤӤȤ˿㤯ʤ뤬Log-Likelihood Test Ϥαƶˤ [ Rayson and Garside 2 ]
Log-Likelihood Test δŪʹͤϡѥȤˤɽδԤи١ʴ١ˤФͤȼºݤ˽и١ʴѻ١ˤκñʸȹͤۤɤ˶Ƥ뤫ɤȽꤹȤΤǤ롥ȤơΤ褦ʥǥBNC ( The British National Corpus ) äե֥ѥȽե֥ѥ̤ξ֥ѥ֤ f*ck Ȥ four-letter word ٤Ӥ롥BNCweb ꤳΥɤȡΤ褦ʷ̤줿
| Category | No. of words | No. of hits | Dispersion (over files) | Frequency per million words |
|---|---|---|---|---|
| Spoken | 10,409,858 | 579 | 63/908 | 55.62 |
| Written | 87,903,571 | 743 | 172/3,140 | 8.45 |
| total | 98,313,429 | 1,322 | 235/4,048 | 13.45 |
| Corpus 1 | Corpus 2 | Total | |
|---|---|---|---|
| Frequency of word | a | b | a+b |
| Frequency of other words | c-a | d-b | c+d-a-b |
| Total | c | d | c+d |
Powered by WinChalow1.0rc4 based on chalow