E-J048-068_Friginal_Walker_Randall


EuroAmerican	
  Journal	
  of	
  Applied	
  Linguistics	
  and	
  Languages	
  
E-JournALL	
  Volume	
  1,	
  Issue	
  1,	
  November	
  2014,	
  pages	
  48-68	
  
ISSN	
  2376-905X	
  	
  
DOI	
  http://dx.doi.org/10.21283/2376905X.1.4
www.e-journall.org	
  

Copyright	
  ©	
  2014.	
  This	
  work	
  is	
  licensed	
  under	
  the	
  Creative	
  Commons	
  Attribution-­‐NonCommercial-­‐NoDerivs	
  3.0	
  Unported	
  License.

1	
  Contact:	
  efriginal@gsu.edu	
  

Exploring	
  mega-­‐corpora:	
  Google	
  Ngram	
  Viewer	
  and	
  the	
  Corpus	
  of	
  
Historical	
  American	
  English	
  

ERIC	
  FRIGINALa1,	
  MARSHA	
  WALKERb,	
  JANET	
  BETH	
  RANDALLc 	
  
aDepartment	
  of	
  Applied	
  Linguistics	
  and	
  ESL,	
  Georgia	
  State	
  University	
  

bLanguage	
  Institute,	
  Georgia	
  Institute	
  of	
  Technology	
  
cAmerican	
  Language	
  Institute,	
  New	
  York	
  University,	
  Tokyo	
  

Received	
  10	
  December	
  2013;	
  received	
  in	
  revised	
  form	
  17	
  May	
  2014;	
  accepted	
  8	
  August	
  2014	
  

ABSTRACT 
EN The creation of internet-based mega-corpora such as the Corpus of Contemporary American English (COCA), the Corpus of

Historical American English (COHA) (Davies, 2011a) and the Google Ngram Viewer (Cohen, 2010) signals a new phase in 
corpus-based research that provides both novice and expert researchers immediate access to a variety of online texts and 
time-coded data. This paper explores the applications of these corpora in the analysis of academic word lists, in particular, 
Coxhead’s (2000) Academic Word List (AWL). Coxhead (2011) has called for further research on the AWL with larger corpora, 
noting that learners’ use of academic vocabulary needs to address for the AWL to be useful in various contexts. Results show 
that words on the AWL are declining in overall frequency from 1990 to the present. Implications about the AWL and future 
directions in corpus-based research utilizing mega-corpora are discussed.  

Keywords:	
  GOOGLE N-GRAM VIEWER, CORPUS OF HISTORICAL AMERICAN ENGLISH, MEGA-CORPORA, TREND STUDIES.

ES La creación de megacorpus basados en Internet, tales como el Corpus of Contemporary American English (COCA), el Corpus
of Historical American English (COHA) (Davies, 2011a) y el Visor de Ngramas de Google (Cohen, 2010), anuncian una nueva 
fase en la investigación basada en corpus, pues proporcionan, tanto a investigadores noveles como a expertos, un acceso 
inmediato a una gran diversidad de textos online y datos codificados con time-code. Este artículo explora las aplicaciones de 
estos corpus en el análisis de listas de vocabulario académico, en particular, Coxhead’s (2000) Academic Word List (AWL). 
Coxhead (2011) hizo patente la necesidad de seguir investigando las aplicaciones del AWL con corpus más amplios, al 
apuntar a que el uso de vocabulario académico por parte de los aprendices necesita ser considerado para que el AWL sea útil 
en diferentes contextos. Los resultados muestran que la frecuencia de uso general de las palabras contenidas en el AWL está 
disminuyendo desde 1990. Asimismo, se tratan los efectos de esta tendencia en el AWL y las futuras líneas de investigación 
de estudios que utilizan megacorpus. 

Palabras clave: VISOR DE NGRAMAS DE GOOGLE, CORPUS HISTÓRICO DE INGLÉS AMERICANO, MEGACORPUS, ESTUDIOS DE 

TENDENCIAS 

IT La creazione di mega-corpora basati su internet, come il Corpus of Contemporary American English (COCA), il Corpus of
Historical American English (COHA) (Davies, 2011a) e il Google Ngram Viewer (Cohen, 2010), inaugura una nuova fase della 
ricerca basata su corpora che permette a ricercatori e ricercatrici junior e senior di accedere a un’ampia gamma di testi on-line 
e dati con codifica time-code. L’articolo esplora le applicazioni di questi corpora nell’analisi di glossari (word lists) accademici, 
in particolare, l’Academic Word List (AWL) di Coxhead (2000). Coxhead (2011) ha sollecitato ulteriori ricerche sull’AWL tramite 
corpora di estensione maggiore, sottolineando la necessità di considerare l’uso del lessico accademico da parte degli 
apprendenti affinché l’AWL sia utile in molteplici contesti. I risultati mostrano che dal 1990 a oggi la frequenza generale dei 
lemmi dell’AWL è diminuita. Seguono considerazioni sulle implicazioni riguardanti l’AWL e i futuri orientamenti della ricerca 
basata sull’uso di mega-corpora. 

Parole-chiave: GOOGLE N-GRAM VIEWER, CORPUS OF HISTORICAL AMERICAN ENGLISH, MEGA-CORPORA, STUDI DI TENDENZA.



FRIGINAL,	
  WALKER,	
  &	
  RANDALL	
  
	
  
	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
   49	
  

1.	
  Introduction	
  
Over	
  the	
  past	
  several	
  years,	
  research	
  in	
  corpus	
  linguistics	
  has	
  produced	
  generalizable	
  and	
  accurate	
  

linguistic	
   information	
   that	
   has	
   been	
   useful	
   in	
   analyzing	
   variation	
   in	
   language	
   (Biber,	
   Conrad,	
   &	
   Reppen,	
  
1998).	
   Enabled	
   by	
   major	
   advancements	
   in	
   computational	
   technology,	
   software	
   design,	
   and	
   the	
   internet,	
  
corpus-­‐based	
   approaches	
   have	
   continued	
   to	
   explore	
   specialized	
   corpora	
   and	
   extensive	
   applications	
   of	
  
quantitative	
   data	
   which	
   would	
   otherwise	
   be	
   infeasible	
   in	
   traditional	
   research	
   methodologies.	
   As	
   a	
   result,	
  
researchers	
  using	
  corpora	
  are	
  able	
  to	
  pursue	
  a	
  dynamic	
  set	
  of	
  research	
  questions	
  that	
  often	
  result	
  in	
  radically	
  
different	
  perspectives	
  on	
  language	
  variation	
  and	
  use	
  from	
  those	
  taken	
  in	
  previous	
  research	
  (Biber,	
  Reppen,	
  &	
  
Friginal,	
   2010).	
   In	
   addition,	
   the	
   number	
   of	
   corpora	
   freely	
   distributed	
   online	
   has	
   increased	
   tremendously,	
  
allowing	
   both	
   novice	
   and	
   expert	
   researchers	
   the	
   opportunity	
   to	
   obtain	
   relevant	
   linguistic	
   frequency	
   and	
  
distributional	
  data	
  immediately.	
  Searchable	
  corpora	
  and	
  databases	
  with	
  built-­‐in	
  interfaces	
  that	
  provide	
  part-­‐
of-­‐speech	
   (POS)	
   tags,	
   demographic	
   information	
   of	
   writers	
   and	
   speakers	
   of	
   texts,	
   and	
   register-­‐specific	
  
comparative	
  charts	
  can	
  now	
  easily	
  be	
  retrieved	
  online.	
  This	
  opportunity	
  has	
  made	
  the	
  domain	
  of	
  web-­‐based,	
  
corpus	
   research	
   user-­‐friendly	
   and	
   accessible.	
   Developments	
   such	
   as	
   these	
   have	
   redefined	
   the	
   nature	
   of	
  
linguistic	
   research	
  with	
   corpora,	
   which	
   has	
   traditionally	
   been	
   associated	
   only	
   with	
   a	
   specialized	
   group	
   of	
  
linguists	
  in	
  academic	
  settings.	
  	
  	
  	
  	
  	
  	
  

Biber,	
  Conrad,	
  and	
  Reppen	
  (1998)	
  define	
  “corpus”	
  and	
  “corpus	
  representativeness”	
  as:	
  
[…]	
  not	
  only	
  a	
  collection	
  of	
  texts.	
  Rather,	
  a	
  corpus	
  seeks	
  to	
  represent	
  a	
  language	
  or	
  some	
  part	
  
of	
  a	
  language.	
  The	
  appropriate	
  design	
  for	
  a	
  corpus	
  therefore	
  depends	
  upon	
  what	
  it	
  is	
  meant	
  
to	
   represent.	
   The	
   representativeness	
   of	
   a	
   corpus,	
   in	
   turn,	
   determines	
   the	
   kind	
  of	
   research	
  
questions	
   that	
   can	
  be	
   addressed	
  and	
   the	
   generalizability	
   of	
   the	
   results	
   of	
   the	
   research.	
   (p.	
  
246)	
  
	
  

This	
   definition	
   and	
   operationalization	
   of	
   a	
   corpus	
   and	
   its	
   inherent	
   representativeness	
   were	
  
especially	
  relevant	
  during	
  pre-­‐computer	
  corpus	
  linguistics	
  and	
  were	
  broadly	
  emphasized	
  in	
  earlier	
  periods	
  of	
  
computer-­‐based	
  corpus	
  design,	
  such	
  as	
  during	
  the	
  development	
  of	
  the	
  Brown	
  corpus	
  and	
  its	
  cognates,	
  known	
  
as	
   the	
   first	
   family	
  of	
  corpora	
  (Svartvik,	
  2007).	
  Researchers	
  have	
  debated	
  the	
  need	
  to	
  delineate	
  how	
  large	
  a	
  
corpus	
  must	
  be	
  in	
  order	
  to	
  accurately	
  represent	
  the	
  language	
  of	
  a	
  particular	
  population	
  of	
  writers/speakers	
  
or	
   the	
   distribution	
   of	
   certain	
   grammatical	
   features	
   (e.g.,	
   Biber,	
   1990,	
   1993;	
  McEnery,	
   Xiao,	
   &	
   Tono,	
   2006;	
  
Miller,	
  2011).	
  The	
  corpus-­‐driven	
  approach,	
  which	
  prescribes	
  formulating	
  an	
  overarching	
  research	
  question	
  as	
  
a	
   primary	
   step	
   before	
   initiating	
   the	
   collection	
   of	
   a	
   corpus,	
   follows	
   corpus	
   design	
   principles	
   and	
   standards	
  
based	
  on	
   the	
   concepts	
  of	
   representativeness	
   and	
  generalizability	
  of	
  data	
   and	
  anticipated	
   results	
   (McEnery,	
  
Xiao,	
  &	
  Tono,	
   2006).	
  At	
   the	
   same	
   time,	
   however,	
   texts	
   intended	
   to	
   represent	
   varieties	
   and	
   sub-­‐varieties	
   of	
  
languages	
   have	
   been	
   compiled	
   over	
   the	
   years,	
   producing	
   large-­‐scale	
   general	
   corpora,	
   such	
   as	
   the	
   British	
  
National	
  Corpus	
  (BNC)	
  and	
  the	
  American	
  National	
  Corpus	
  (ANC),	
  which	
  can	
  be	
  said	
  to	
  represent	
  two	
  national	
  
varieties	
   of	
   English,	
   and	
   the	
   International	
   Corpus	
   of	
   English	
   (ICE),	
   which	
   purportedly	
   represents	
  
international/global	
  varieties	
  of	
  educated	
  English	
  (Bautista,	
  2011;	
  Nelson,	
  1996).	
  These	
  three	
  corpora	
  will	
  be	
  
discussed	
  further	
  in	
  the	
  sections	
  below.	
  

The	
  successful	
  collection	
  of	
  internet-­‐based	
  mega-­‐corpora	
  (>300	
  million	
  words)	
  such	
  as	
  the	
  Corpus	
  of	
  
Contemporary	
  American	
   English	
   (COCA),	
   the	
   Corpus	
   of	
  Historical	
   American	
   English	
   (COHA)	
   (Davies	
   2005,	
  
2011a,	
  2011b),	
  and	
  the	
  Google	
  Ngram	
  Viewer	
  (Cohen,	
  2010;	
  Hotz,	
  2010;	
  Michel,	
  Shen,	
  Aiden,	
  Veres,	
  &	
  Gray,	
  
2011)	
   may	
   signal	
   a	
   new	
   phase	
   in	
   corpus	
   design	
   in	
   which	
   sampling	
   texts	
   in	
   order	
   to	
   achieve	
   statistical	
  
representativeness	
  will	
  no	
  longer	
  be	
  required.	
  Mega-­‐corpora	
  have	
  the	
  potential	
  to	
  cover	
  an	
  entire	
  population	
  
of	
  texts	
  in	
  a	
  particular	
  register	
  instead	
  of	
  only	
  presenting	
  a	
  sample,	
  as	
  in	
  the	
  current	
  efforts	
  of	
  Google	
  Books	
  
to	
  scan	
  and	
  compile	
  books	
  and	
  manuscripts	
  produced	
  since	
  the	
  1500s	
  across	
  many	
  languages	
  (Toor,	
  2010).	
  
The	
  clear	
  message	
  here	
  is	
  that	
  technology	
  will	
  eventually	
  enable	
  researchers	
  to	
  have	
  easy	
  access	
  to	
  most,	
   if	
  
not	
  all,	
  texts	
  in	
  a	
  register.	
  For	
  example,	
  access	
  to	
  all	
  internet-­‐based	
  texts	
  (e.g.,	
  social	
  media	
  language	
  through	
  
Facebook	
  and	
  Twitter	
  posts,	
  personal	
  blogs,	
  news	
  reports,	
  etc.)	
  could	
  be	
  entirely	
  and	
  automatically	
  obtained	
  
without	
  the	
  need	
  for	
  sampling	
  or	
  manual/selective	
  crawling.	
  	
  
	
  
1.1.	
  Analysis	
  of	
  large	
  electronic	
  corpora	
  

Studies	
   based	
   on	
   large	
   electronic	
   corpora	
   began	
   in	
   the	
   late-­‐1960s	
   with	
   the	
   Brown	
   Corpus,	
   a	
   one	
  
million	
  word	
   collection	
  of	
  published	
  written	
   texts	
   in	
  American	
  English,	
   compiled	
  by	
  Kucera	
   and	
  Francis.	
  A	
  
parallel	
  corpus	
  of	
  British	
  written	
  texts,	
  the	
  Lancaster-­‐Oslo-­‐Bergen	
  (LOB)	
  Corpus	
  was	
  published	
  subsequently.	
  



EXPLORING	
  MEGA-­‐CORPORA	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
  50	
  

Kucera	
  and	
  Francis	
  pioneered	
  extensive	
  work	
  on	
  word	
  frequencies	
  and	
  distributions	
  of	
  grammatical	
  part-­‐of-­‐
speech	
   categories	
   using	
   the	
   Brown	
   and	
   LOB	
   corpora	
   (Biber,	
   1995).	
   In	
   the	
   early	
   1980’s,	
   the	
   increasing	
  
availability	
  of	
  personal	
  computers,	
  electronic	
  corpora,	
  and	
  corpus	
  tools	
  such	
  as	
  concordancers,	
  taggers,	
  and	
  
parsers	
   facilitated	
   wide-­‐ranging	
   linguistic	
   analyses.	
   In	
   those	
   years,	
   descriptive	
   studies	
   featuring	
   the	
  
distributions	
  of	
  linguistic	
  features	
  (e.g.,	
  passives,	
  verbs,	
  formulaic	
  sequences,	
  nominalizations)	
  across	
  corpora	
  
were	
   conducted	
   and	
  published.	
  During	
   the	
   same	
  period,	
   dictionaries,	
   such	
   as	
   the	
   Collins	
   COBUILD	
  English	
  
Language	
   Dictionary	
   (1987)	
   and	
   the	
   Longman	
   Dictionary	
   of	
   Contemporary	
   English	
   (1987),	
   intended	
   for	
  
language	
   learners	
   and	
   based	
   on	
   the	
   analysis	
   of	
   large	
   electronic	
   corpora,	
   began	
   to	
   be	
   published	
   (Biber,	
  
Reppen,	
  &	
  Friginal,	
  2010).	
  	
  

From	
   the	
   late	
   1980’s,	
   Biber	
   developed	
   quantitative	
   studies	
   of	
   tagged	
   texts	
   using	
   advanced	
  
multivariate	
  statistical	
  techniques	
  (e.g.,	
  Factor	
  and	
  Cluster	
  Analyses)	
  and	
  focusing	
  on	
  the	
  concept	
  of	
  statistical	
  
co-­‐occurrence	
   of	
   linguistic	
   features	
   across	
   registers.	
   Biber’s	
   comparison	
   of	
   large	
   volumes	
   of	
   texts	
  
representing	
   sub-­‐registers	
   of	
   speech	
   and	
   writing	
   in	
   cross-­‐linguistic	
   settings	
   provided	
   new	
   approaches	
   in	
  
simultaneous	
   explorations	
   of	
   general	
   and	
   specialized	
   corpora	
   (Friginal,	
   2009).	
   The	
   Longman	
   Grammar	
   of	
  
Spoken	
  and	
  Written	
  English	
  (Biber,	
  Johansson,	
  Leech,	
  Conrad,	
  &	
  Finegan,	
  1999),	
  for	
  example,	
  provided	
  wide-­‐
ranging	
   distributional	
   comparisons	
   of	
   linguistic	
   features	
   from	
   spoken	
   and	
   written	
   British	
   and	
   American	
  
Englishes.	
   These	
   descriptive	
   data	
   on	
   lexico-­‐grammatical	
   variation	
   have	
   also	
   been	
   used	
   in	
   the	
   design	
   of	
  
classroom	
  activities	
  for	
  English	
  language	
  teaching,	
  especially	
  for	
  non-­‐native	
  speakers	
  of	
  English.	
  	
  	
  	
  

Exponential	
   developments	
   in	
   internet	
   technology	
   in	
   the	
   last	
   decade	
   have	
   positively	
   altered	
   the	
  
nature	
   of	
   corpus-­‐based	
   research.	
   Collecting	
   written	
   texts	
   such	
   as	
   newspaper	
   articles,	
   published	
   academic	
  
papers,	
  opinion	
  columns,	
  and	
  weblogs	
  became	
  easily	
  manageable,	
  as	
  more	
  and	
  more	
  types	
  of	
  specialized	
  texts	
  
have	
  been	
  uploaded	
   and	
  distributed	
  online	
   (Grieve,	
   Biber,	
   Friginal,	
  &	
  Nekrasova,	
   2010;	
  Herring	
  &	
  Paolillo,	
  
2006).	
  Although	
  there	
  are	
  still	
  many	
  challenges	
  in	
  the	
  collection	
  of	
  various	
  specialized	
  corpora,	
  especially	
  in	
  
the	
   domain	
   of	
   spoken	
   discourse,	
   the	
   current	
   set	
   of	
   freely	
   available	
   online	
   corpora	
   shows	
   how	
   corpus	
  
linguistics	
  has	
  progressed	
  from	
  the	
  1960s	
  to	
  the	
  present	
  (Preiss,	
  Coonce,	
  &	
  Baker,	
  2009).	
  The	
   internet	
  as	
  a	
  
corpus	
   (Crystal,	
   2006)	
  not	
  only	
   covers	
  English	
   texts	
  but	
   also	
  other	
   languages	
  used	
  online	
  by	
  an	
   increasing	
  
number	
   of	
   users.	
   Automated	
   transcriptions	
   of	
   online	
   audio	
   and	
   video	
   clips	
   and	
   translation	
   services	
   (e.g.,	
  
Dragon	
  Dictation	
   software,	
   Google	
   Translate	
   from	
  Google	
   Labs)	
   have	
   also	
   been	
   introduced	
   and	
   developed,	
  
showing	
  promising	
  applications	
   that	
  could	
  help	
   in	
   the	
  collection	
  of	
  spoken	
  texts.	
  These	
  online	
   tools	
  greatly	
  
contribute	
   to	
   how	
   corpora	
   are	
   now	
   collected	
   and	
   may	
   necessitate	
   new	
   models	
   in	
   corpus	
   design	
   and	
  
compilation.	
  	
  	
  	
  	
  
	
  
1.2.	
  Development	
  of	
  English	
  electronic	
  mega-­‐corpora	
  

Since	
   its	
   release	
   in	
   the	
   early	
   1990s,	
   the	
   100	
  million	
  word	
  British	
  National	
   Corpus	
   (BNC,	
   including	
  
recent	
  XML	
  version)	
  has	
  been	
  used	
   in	
  more	
  corpus-­‐related	
  studies	
   than	
  any	
  other	
  English-­‐language	
  corpus	
  
(Davies,	
   2009).	
   The	
   BNC,	
   with	
   its	
   web-­‐based	
   interface,	
   BNCweb,	
   (http://bncweb.info/)	
   provides	
   a	
   client-­‐
program	
   for	
   searching	
   and	
   retrieving	
   lexico-­‐syntactic	
   and	
   textual	
   data	
   from	
   available	
   metadata	
   of	
  
writers’/speakers’	
  demographic	
   information	
  and	
  registers.	
  The	
  BNC	
  served	
  as	
  the	
  model	
  for	
  the	
  creation	
  of	
  
the	
   American	
   English	
   parallel	
   corpus,	
   ANC	
   (http://americannationalcorpus.org/),	
   which	
   has	
   released	
   two	
  
versions	
  of	
  component	
  sub-­‐corpora	
  totaling	
  over	
  22	
  million	
  words.	
  Work	
  on	
  the	
  ANC	
  is	
  has	
  progressed	
  slowly	
  
and	
   there	
   is	
   currently	
   no	
   user-­‐friendly	
   client	
   program	
   available	
   for	
   automatic	
   database	
   online	
   searches.	
  
However,	
  the	
  ANC	
  has	
  freely	
  downloadable	
  texts	
  totaling	
  over	
  15	
  million	
  words	
  and	
  the	
  annotated	
  versions	
  of	
  
the	
  corpus	
  also	
  include	
  grammatically-­‐tagged	
  data	
  and	
  other	
  XML	
  annotations.	
  	
  

Davies	
  (2009,	
  p.	
  59)	
  pointed	
  out	
  that,	
  valuable	
  as	
  it	
  is,	
  the	
  BNC	
  “is	
  beginning	
  to	
  show	
  its	
  age	
  in	
  some	
  
respects.”	
  He	
  noted	
  that	
  while	
  files	
  and	
  demographic	
  information	
  on	
  some	
  registers	
  have	
  been	
  corrected	
  or	
  
updated,	
  substantive	
  additions	
  to	
  the	
  corpus	
  have	
  been	
  very	
  limited	
  since	
  1993.	
  Because	
  there	
  is	
  no	
  currently	
  
planned	
   expansion	
   to	
   the	
   BNC,	
   it	
   may	
   become	
   increasingly	
   out	
   of	
   date	
   with	
   respect	
   to	
   recent	
   changes	
   in	
  
English	
  and	
  the	
  need	
  to	
  represent	
  additional	
  registers	
  such	
  as	
  texting/SMS	
  language	
  and	
  the	
  online	
  language	
  
of	
  social	
  media	
  and	
  blogging.	
  Similarly,	
   the	
  ANC’s	
  collection	
  of	
   texts	
  reflects	
  work	
  done	
   from	
  early-­‐	
   to	
  mid-­‐
2000s	
  with	
  no	
  continuing	
  contributions	
  that	
  allow	
  for	
  diachronic	
  comparisons	
  across	
  registers.	
  	
  

The	
  Corpus	
  of	
  Contemporary	
  American	
  English	
  (COCA),	
  developed	
  and	
  published	
  online	
  by	
  Davies,	
  
was	
  released	
   in	
  early	
  2008.	
   It	
   covers	
  a	
  diverse	
  collection	
  of	
  American	
  English	
   texts	
   totaling	
  more	
   than	
  385	
  
million	
   words	
   from	
   1990–2008	
   (20	
  million	
   words	
   each	
   year)	
   across	
   registers	
   grouped	
   into	
   the	
   following	
  
categories:	
  spoken,	
  fiction,	
  popular	
  magazines,	
  newspapers,	
  and	
  academic	
  journals.	
  Since	
  its	
  release,	
  COCA’s	
  
online	
  interface	
  has	
  been	
  widely	
  used	
  by	
  researchers,	
  teachers,	
  and	
  students	
  for	
  various	
  purposes,	
  including	
  



FRIGINAL,	
  WALKER,	
  &	
  RANDALL	
  
	
  
	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
   51	
  

producing	
   materials	
   for	
   teaching	
   collocations,	
   lexico-­‐syntactic	
   features	
   of	
   English,	
   and	
   diachronic	
   word	
  
frequency	
   changes	
   across	
   registers.	
   COCA	
   is	
   comparable	
   to	
   the	
   BNC/ANC	
   in	
   terms	
   of	
   text	
   types,	
   though	
   it	
  
deviates	
   from	
  them	
  in	
   the	
   types	
  of	
  spoken	
  data	
   it	
   includes:	
  COCA’s	
  spoken	
  texts	
   (20%	
  of	
   the	
  corpus)	
  come	
  
mainly	
   from	
  television	
  news	
  and	
  interview	
  programs,	
  rather	
  than	
  from	
  the	
  types	
  of	
  conversation	
  data	
  (e.g.,	
  
face	
   to	
   face	
   conversation,	
   service	
   encounters,	
   and	
   telephone	
   interactions)	
   available	
   in	
   the	
   BNC	
   or	
   other	
  
corpora	
   such	
   as	
   the	
   Longman	
   Corpus.	
   Davies	
   (2009)	
   maintains,	
   however,	
   that	
   COCA’s	
   overall	
   balanced	
  
composition	
  means	
  that	
  researchers	
  can	
  compare	
  data	
  across	
  registers	
  and	
  achieve	
  relatively	
  accurate	
  results	
  
that	
  show	
  patterns	
  of	
  change	
  in	
  the	
  language	
  from	
  the	
  1990s	
  to	
  the	
  present.	
  

In	
  addition	
   to	
   these	
   three	
  corpora,	
   the	
   International	
  Corpus	
  of	
  English	
   (ICE)	
  has	
  also	
  been	
  created	
  
and	
   made	
   available	
   since	
   the	
   mid-­‐1990s.	
   The	
   ICE	
   consists	
   of	
   one	
   million	
   words	
   per	
   spoken	
   and	
   written	
  
variety	
  of	
  English	
  produced	
  by	
  each	
  of	
  over	
  20	
  research	
  teams	
  worldwide	
  (http://ice-­‐corpora.net/ice/).	
  For	
  
most	
   participating	
   countries,	
   the	
   ICE	
   project	
   served	
   as	
   the	
   first	
   systematic	
   investigation	
   of	
   the	
   national,	
  
“educated”	
  English	
  variety	
   (Nelson,	
  1996).	
  Each	
   component	
   corpus	
   follows	
  a	
   common	
  corpus	
  design	
  and	
  a	
  
similar	
   scheme	
   for	
   grammatical	
   annotation,	
   as	
   the	
   ICE	
  was	
   primarily	
   intended	
   for	
   comparative	
   studies	
   of	
  
emerging	
  Englishes	
   all	
   over	
   the	
  world.	
   For	
   example,	
  Asian	
   varieties	
   of	
   English	
   available	
   for	
   free	
  download	
  
from	
   the	
   ICE	
   website	
   include	
   sub-­‐corpora	
   from	
   countries/territories	
   such	
   as	
   Hong	
   Kong,	
   the	
   Philippines,	
  
India,	
   and	
   Singapore,	
  where	
   English	
   has	
   been	
   used	
   extensively	
   as	
   the	
   language	
   of	
   business	
   and	
   education.	
  
Written	
   registers	
   in	
   the	
   ICE	
   range	
   from	
   student	
  writing,	
   novels	
   and	
   stories,	
   social	
   and	
   business	
   letters,	
   to	
  
published,	
   professional	
   texts.	
   The	
   spoken	
   data	
   component	
   features	
   face-­‐to-­‐face	
   conversations,	
   broadcast	
  
speech,	
   spontaneous	
   commentaries,	
   telephone	
   calls,	
   parliamentary	
   debates,	
   and	
   legal	
   cross-­‐examinations,	
  
among	
  other	
  groups	
  of	
  texts.	
  

Finally,	
   and	
   as	
   the	
   primary	
   focus	
   of	
   this	
   paper,	
   the	
   internet	
   has	
   paved	
   the	
  way	
   for	
   Google	
  Ngram	
  
Viewer	
   and	
   COHA,	
   two	
   large	
   online	
   databases	
   of,	
   specialized	
   categories	
   of	
   texts,	
   released	
   in	
   2010,	
   which	
  
reflect	
  current	
  academic	
  and	
  corporate	
  efforts	
  to	
  freely	
  distribute	
  corpus-­‐based	
  data	
  to	
  the	
  public.	
  These	
  two	
  
mega-­‐corpora	
  are	
  both	
  designed	
  to	
  address	
  diachronic	
  analyses,	
  but	
  can	
  also	
  be	
  used	
  for	
  synchronic	
  studies	
  
due	
  to	
  the	
  size	
  of	
  data	
  representing	
  recent	
  published	
  texts	
  (Davies,	
  2011a).	
  Both	
  of	
  these	
  applications	
  aim	
  at	
  
continuing	
   to	
   periodically	
   add	
   texts	
   to	
   their	
   databases.	
   A	
   number	
   of	
   studies	
   exploring	
   these	
   corpora’s	
  
potential	
   usefulness	
   in	
   language	
   analysis	
   and	
   corpus	
   linguistics	
   were	
   initiated	
   in	
   the	
   months	
   after	
   they	
  
became	
   publicly	
   available;	
   a	
   review	
   of	
   three	
   recent	
   conferences,	
   including	
   the	
   2011	
   and	
   2013	
   American	
  
Association	
  for	
  Corpus	
  Linguistics	
  Conference,	
  shows	
  that	
  over	
  25	
  papers	
  were	
  presented	
  using	
  either	
  COHA	
  
or	
  Google	
  Ngram	
  Viewer.	
  

	
  
2.	
  Exploring	
  mega-­‐corpora	
  
	
  
2.1.	
  Google	
  Ngram	
  Viewer	
  

Google	
  Ngram	
  Viewer	
   is	
   comprised	
   of	
   over	
   5.2	
  million	
   books	
   from	
  Google	
   Books,	
   a	
   subdivision	
   of	
  
Google	
  Inc.	
  that	
  has	
  conducted	
  an	
  extensive	
  scanning	
  of	
  published	
  manuscripts	
  in	
  order	
  to	
  create	
  a	
  database	
  
of	
  electronic	
  or	
  digitized	
  texts.	
  The	
  number	
  of	
  currently	
  scanned	
  books	
  comprises	
  approximately	
  4%	
  of	
  all	
  the	
  
books	
   ever	
   written	
   in	
   English	
   (Bohannon,	
   2011).	
   This	
   mega-­‐corpus	
   contains	
   over	
   500	
   billion	
   words;	
   the	
  
majority	
  of	
  them	
  are	
  in	
  English	
  (361	
  billion).	
  Other	
  available	
  languages	
  in	
  the	
  corpus	
  include	
  French,	
  Spanish,	
  
German,	
   Chinese,	
   Russian,	
   and	
  Hebrew	
   (Bohannon,	
   2011;	
   Keuleers,	
   Brysbaert,	
   &	
  New,	
   2011;	
  Michel	
   et	
   al.,	
  
2011).	
   The	
   Ngram	
   Viewer	
   contains	
   data	
   that	
   span	
   from	
   1550-­‐2008,	
   although	
   texts	
   before	
   the	
   1800s	
   are	
  
extremely	
   limited	
  and	
  often	
   there	
   are	
  only	
   a	
   few	
  books	
  per	
   year.	
   From	
   the	
  1800s,	
   the	
   corpus	
  grows	
   to	
  98	
  
million	
  words	
  per	
  year;	
  by	
  the	
  1900s,	
  it	
  reaches	
  1.8	
  billion,	
  and	
  11	
  billion	
  per	
  year	
  by	
  the	
  2000s	
  (Michel	
  et	
  al.,	
  
2011).	
  Hence,	
  texts	
  collected	
  for	
  Google	
  Ngram	
  Viewer	
  represent	
  the	
  largest	
  corpus	
  to	
  date.	
  	
  

Google	
  Ngram	
  Viewer	
  is	
  composed	
  of	
  raw	
  data	
  that	
  is	
  encoded	
  by	
  the	
  number	
  of	
  n-­‐grams,	
  adjacent	
  
sequences	
  of	
  n	
  items	
  from	
  a	
  text.	
  The	
  n-­‐gram	
  was	
  developed	
  as	
  a	
  response	
  to	
  the	
  concerns	
  of	
  Peter	
  Norvig,	
  
the	
   head	
   of	
   research	
   at	
   Google	
   Labs,	
   about	
   developing	
   the	
   online	
   viewer	
   interface	
   due	
   to	
   many	
   pending	
  
lawsuits	
  about	
  Google’s	
  book	
  digitizing	
  initiative	
  (Bohannon,	
  2010).	
  In	
  order	
  to	
  avoid	
  further	
  problems	
  with	
  
publishers	
  and	
  copyright	
  owners	
  of	
  published	
  materials,	
  the	
  Ngram	
  Viewer	
  makes	
  use	
  of	
  texts	
  converted	
  into	
  
various	
  n-­‐grams	
  so	
  that	
  data	
  from	
  the	
  whole	
  corpus	
  cannot	
  be	
  downloaded	
  as	
  complete	
  books,	
  and	
  in	
  effect,	
  
“cannot	
   be	
   read	
   by	
   a	
   human”	
   (Michel	
   et	
   al.,	
   2011,	
   p.176).	
   The	
   process	
   of	
   converting	
   Google	
   Books	
   into	
   n-­‐
grams	
  has	
  been	
  tedious,	
  and,	
  as	
  result,	
  just	
  5.2	
  million	
  of	
  the	
  15	
  million	
  Google	
  books	
  have	
  been	
  converted	
  so	
  
far.	
  Google’s	
  conversion	
  procedure	
  is	
  as	
  follows:	
  



EXPLORING	
  MEGA-­‐CORPORA	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
  52	
  

	
  
-­‐ Raw	
   texts	
   extracted	
   from	
   different	
   sources	
   are	
   pre-­‐processed	
   by	
   tokenizing	
  white-­‐spaces;	
   upper	
  
case	
  letters	
  are	
  converted	
  to	
  lower	
  case.	
  	
  

-­‐ Numbers	
   are	
   retained,	
   and	
   no	
   stemming/inflection	
   are	
   performed.	
   The	
   n-­‐gram	
   LMs	
   (Language	
  
Models)	
   are	
   word-­‐based	
   backoff	
   models,	
   where	
   the	
   n-­‐gram	
   probabilities	
   are	
   estimated	
   using	
  
Maximum	
   Likelihood	
   Estimation	
   (MLE)	
  with	
   smoothing	
   (Gao, Nguyen,	
   Li,	
   Thrasher,	
   Li,	
   &	
  Wang,	
  
2010).	
  	
  

-­‐ The	
  viewer	
  is	
  composed	
  of	
  n-­‐grams,	
  from	
  1-­‐gram	
  to	
  5-­‐grams.	
  N-­‐grams	
  represent	
  how	
  many	
  words	
  
are	
  in	
  a	
  lexical	
  bundle:	
  1-­‐gram	
  =	
  1	
  word,	
  while	
  5-­‐grams	
  =	
  5	
  words	
  (Bohannon,	
  2011).	
  Once	
  the	
  n-­‐
grams	
   are	
   saved	
   as	
   raw	
   data	
   in	
   Excel	
   files,	
   the	
   n-­‐grams	
   can	
   then	
   be	
   searched	
   using	
   the	
   online	
  
viewer.	
  	
  

-­‐ Search	
   results	
   are	
   displayed	
   by	
   frequency	
   for	
   each	
   n-­‐gram	
   determined	
   per	
   year	
   following	
   the	
  
formula:	
  total	
  number	
  of	
  instances	
  of	
  the	
  n-­‐gram	
  /	
  total	
  number	
  of	
  words	
  in	
  that	
  year.	
  

	
  
It	
   is	
   important	
   to	
   note	
   that	
   Google	
   Ngram	
   Viewer	
  was	
   not	
   created	
  with	
   research	
   in	
   linguistics	
   or	
  

corpus	
   linguistics	
   as	
   its	
   primary	
   application	
   (Michel	
   et	
   al.,	
   2011).	
   The	
   developers	
   of	
   the	
   viewer	
  wanted	
   to	
  
create	
   a	
   new	
   approach	
   to	
   humanities	
   research	
   that	
   they	
   coined	
   Culturomics.	
   Culturomics	
  
(www.culturomics.org/home)	
   is	
   a	
  way	
   to	
   quantify	
   culture	
   by	
   analyzing	
   the	
   growth,	
   change,	
   and	
  decline	
   of	
  
published	
  words	
   over	
   centuries.	
   This	
   would	
  make	
   it	
   possible	
   to	
   rigorously	
   study	
   the	
   evolution	
   of	
   culture	
  
using	
  distributional,	
  quantitative	
  data	
  on	
  a	
  grand	
  scale	
  (Bohannon,	
  2010).	
  In	
  an	
  effort	
  to	
  prove	
  the	
  adequacy	
  
of	
   and	
   to	
   provide	
   clear	
   impetus	
   for	
   Google	
   Ngram	
   Viewer,	
   Google-­‐affiliated	
   researchers	
   have	
   conducted	
   a	
  
series	
   of	
   studies	
   to	
   validate	
   the	
   usefulness	
   and	
   various	
   applications	
   of	
   their	
   new	
   program.	
   One	
   study,	
   for	
  
example,	
   showed	
   that	
   over	
   50%	
   of	
   the	
   words	
   in	
   the	
   n-­‐gram	
   database	
   do	
   not	
   appear	
   in	
   any	
   published	
  
dictionary	
  (Bohannon	
  ,	
  2010).	
  In	
  addition,	
  it	
  is	
  argued	
  that	
  patterns	
  and	
  cultural	
  influences	
  of	
  words	
  could	
  be	
  
clearly	
   established	
   and	
   tracked	
   across	
   timeframes.	
   The	
   use	
   of	
   Google	
   Ngram	
   Viewer	
   and	
   Culturomics,	
  
therefore,	
  contributes	
  academic	
  and	
  technical	
  value	
  to	
  the	
  study	
  of	
  culture,	
  making	
  it	
  arguably	
  a	
  new	
  cultural	
  
tool	
  that	
  has	
  several	
  possibilities.	
  Figure	
  1	
  displays	
  the	
  default	
  search	
  screen	
  of	
  Google	
  Ngram	
  Viewer	
  as	
  of	
  
September	
  2014.	
  
	
  

	
  
Figure 1. Default search screen of Google Ngram Viewer (http://books.google.com/ngrams) 

 
2.2. Corpus of Historical American English (COHA) 

Like	
  the	
  Google	
  Ngram	
  Viewer,	
  the	
  Corpus	
  of	
  Historical	
  American	
  English	
  was	
  debuted	
  in	
  2010	
  with	
  
limited	
  functionality.	
   Information	
  available	
   from	
  COHA	
  has	
  been	
  continually	
  updated	
  since	
  then	
  with	
  major	
  
upgrades	
   in	
   its	
   data	
   and	
   search	
   features,	
   including	
   interesting	
   information	
   on	
   visualization	
   techniques	
   as	
  
applied	
   to	
   COHA’s	
   diachronic	
   distributional	
   data	
   (Hilpert,	
   2011).	
   COHA	
   was	
   developed	
   by	
   Mark	
   Davies,	
  
Professor	
   of	
   Corpus	
   Linguistics,	
   from	
   Brigham	
   Young	
   University.	
   As	
   noted	
   earlier,	
   Davies	
   also	
   created	
   the	
  



FRIGINAL,	
  WALKER,	
  &	
  RANDALL	
  
	
  
	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
   53	
  

Corpus	
  of	
  Contemporary	
  American	
  English	
  and	
  many	
  other	
  interactive	
  online	
  databases	
  for	
  corpus	
  analyses,	
  
such	
  as	
  the	
  Corpus	
  do	
  Português,	
  Corpus	
  del	
  Español,	
  and	
  Corpus	
  of	
  LDS	
  General	
  Conference	
  Talks,	
  which	
  can	
  
be	
   accessed	
   from	
  his	
   personal	
  website	
   (http://davies-­‐linguistics.byu.edu/personal/).	
   COHA	
  was	
   funded	
   by	
  
the	
  National	
  Endowment	
  for	
  the	
  Humanities	
  as	
  part	
  of	
  its	
  "We	
  the	
  People"	
  initiative	
  (Davies,	
  2011a).	
  	
  

COHA	
  has	
  just	
  over	
  400	
  million	
  words,	
  compared	
  to	
  Google	
  Ngram	
  Viewer’s	
  500	
  billion	
  words.	
  Davies	
  
argues,	
  however,	
  that	
  this	
  substantial	
  difference	
  in	
  corpus	
  size	
  does	
  not	
  necessary	
  affect	
  reliability	
  of	
  results	
  
when	
   these	
   two	
   corpora	
   are	
   used	
   and	
   compared	
   across	
   a	
   range	
   of	
   linguistic	
   distributions.	
  When	
   searches	
  
were	
   performed	
   between	
   COHA	
   and	
   Google	
   Ngram	
   Viewer,	
   similar	
   results	
   were	
   found	
   once	
   data	
   were	
  
normalized	
   (Davies,	
  2011a,	
  2011b).	
  To	
  date,	
  COHA’s	
  data	
  cover	
   the	
  period	
   from	
  1810	
  to	
  2010,	
  and	
  specific	
  
registers	
  such	
  as	
  fiction,	
  magazine,	
  non-­‐fiction,	
  and	
  newspaper	
  texts.	
  Following	
  the	
  functionalities	
  established	
  
by	
  COCA,	
  COHA	
  also	
  provides	
  POS	
  lists,	
  collocates,	
  list	
  and	
  chart	
  options,	
  and	
  various	
  sorting	
  and	
  comparison	
  
options.	
  Figure	
  2	
  displays	
  the	
  default	
  search	
  screen	
  of	
  COHA.	
  	
  

	
  

	
  
Figure 2. Default search screen of COHA (http://corpus.byu.edu/coha/) 

	
  
According	
   to	
   Davies,	
   COHA	
   can	
   do	
   exactly	
   what	
   Google	
   Ngram	
   Viewer	
   does,	
   with	
   the	
   addition	
   of	
  

specific	
   features	
   that	
   directly	
   apply	
   to	
   linguistics	
   and	
   corpus-­‐based	
   research.	
   This	
   includes	
   the	
   ability	
   to	
  
employ	
  concordancing	
  tools,	
  display	
  data	
  in	
  context,	
  and	
  limit	
  results	
  by	
  register,	
  characteristics	
  not	
  available	
  
on	
  Google	
  Ngram	
  Viewer.	
  In	
  addition,	
  COHA	
  allows	
  more	
  powerful	
  searches,	
  for	
  example,	
  for	
  synonyms	
  and	
  
word	
   comparisons	
   across	
   historical	
   periods	
   -­‐	
   which	
   may	
   produce	
   insightful	
   analyses	
   about	
   cultural	
   and	
  
societal	
  shifts	
  (Davies,	
  2011a).	
  The	
  types	
  of	
  detailed	
  COHA	
  searches	
  include:	
  

-­‐ lexis,	
  through	
  mass	
  comparison	
  between	
  historical	
  periods,	
  
-­‐ morphology,	
  via	
  wildcards,	
  
-­‐ syntax,	
  from	
  POS-­‐tagged	
  data	
  included	
  in	
  the	
  program,	
  and	
  
-­‐ semantics,	
  via	
  collocates,	
  synonyms,	
  and	
  customized	
  lists.	
  

	
  
2.3.	
  Davies’	
  Google	
  Books	
  Interface	
  (2011c)	
  

In	
  mid-­‐2011,	
  Davies	
  launched	
  an	
  early	
  version	
  of	
  an	
  interface	
  for	
  the	
  Google	
  Ngram	
  Viewer	
  database	
  
that	
   allows	
   a	
  more	
   extensive	
   search	
   of	
   the	
  massive	
   Google	
   collection	
   than	
   the	
   basic	
  Ngram	
  Viewer.	
  While	
  
Davies	
  points	
  out	
  that	
  his	
  interface	
  is	
  not	
  an	
  official	
  product	
  of	
  Google	
  or	
  Google	
  Books	
  (Davies,	
  2011c),	
  this	
  
search	
  platform	
  is	
  very	
  similar	
  to	
  the	
  standard	
  COCA/COHA	
  structure	
  providing	
  options	
  for	
  wildcard,	
  lemma,	
  
part	
   of	
   speech,	
   synonyms,	
   and	
   collocates	
   searches.	
   This	
   interface	
   is	
   more	
   advanced	
   than	
   Google’s	
   online	
  
viewer,	
  with	
  presentations	
  of	
  frequency	
  data	
  that	
  go	
  beyond	
  simple	
  line	
  graphs	
  or	
  figures.	
  In	
  addition,	
  Davies’	
  
interface	
  allows	
  users	
   to	
   copy	
   the	
  data	
   to	
  other	
  applications	
   for	
   further	
  analysis	
   (Davies,	
  2011c),	
   a	
   feature	
  
that	
   is	
   not	
   currently	
   possible	
   in	
   the	
  Google	
   interface.	
   The	
  Davies	
   version	
   is	
   based	
   on	
   155	
   billion	
  words	
   of	
  
American	
   English	
   from	
   1810-­‐2009,	
  with	
   future	
   plans	
   to	
   integrate	
   other	
   Google	
   Books	
   collections	
   into	
   this	
  
new	
  interface,	
  including	
  texts	
  from	
  British	
  and	
  American	
  English	
  texts	
  from	
  the	
  1500s-­‐1700s,	
  as	
  well	
  as	
  texts	
  
from	
  Spanish,	
  German,	
  and	
  French	
  (a	
  grant	
  application	
  to	
  support	
  this	
  endeavor	
  was	
  submitted	
  in	
  mid-­‐2011).	
  
Davies’	
   interface	
   also	
   provides	
   a	
   guided	
   tour	
   of	
   the	
   site	
   (http://googlebooks.byu.edu/)	
   showing	
   its	
   major	
  



EXPLORING	
  MEGA-­‐CORPORA	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
  54	
  

features	
  and	
  an	
  intuitive	
  search	
  form.	
  A	
  click	
  for	
  each	
  search	
  query	
  automatically	
  fills	
  in	
  the	
  form	
  and	
  displays	
  
the	
  results	
  obtained	
  from	
  actual	
  American	
  English	
  books	
  collected	
  by	
  Google	
  Books.	
  	
  
	
  
2.4.	
  Recent	
  linguistic	
  analysis	
  of	
  data	
  from	
  COHA	
  and	
  Google	
  Ngram	
  Viewer	
  

One	
  way	
  to	
  explore	
  mega-­‐corpora	
  such	
  as	
  Google	
  Ngram	
  Viewer	
  and	
  COHA	
  is	
  to	
  conduct	
  diachronic	
  
and	
  trend	
  studies	
  comparing	
  the	
  distribution	
  of	
  various	
  linguistic	
  features	
  across	
  specific	
  time	
  periods.	
  Davies	
  
(2011a),	
   for	
   example,	
   presents	
   frequency	
   changes	
   in	
   lexical	
   items	
   (e.g.,	
   fellow,	
   teenager,	
   sublime,	
   global	
  
warming,	
   steamship)	
   used	
   in	
   published	
   texts	
   from	
   1810	
   to	
   the	
   present.	
   Hilpert	
   (2011)	
   uses	
   visualization	
  
techniques	
   using	
   Google	
   Visualization	
   (or	
   Google	
   Code)	
   and	
   the	
   statistical/graphical	
   software,	
   R,	
   to	
   show	
  
language	
   change	
   through	
   COHA.	
   Sociolinguistic	
   time	
   series	
   studies	
   on	
   semantic	
   shifts	
   (semantic	
   changes)	
  
across	
  its	
  designated	
  registers	
  can	
  also	
  be	
  conducted	
  using	
  COHA,	
  although	
  this	
  could	
  be	
  improved	
  if	
  corpora	
  
were	
  to	
  include	
  demographic	
  information	
  about	
  speakers/writers.	
  While	
  more	
  generic	
  trending	
  information	
  
is	
  provided	
  by	
  Google	
  Ngram	
  Viewer	
  from	
  published	
  books	
  and	
  manuscripts,	
  such	
  a	
  dataset	
  also	
  allows	
  for	
  
immediate	
   macro-­‐level	
   snapshots	
   of	
   quantitative	
   language	
   patterns	
   across	
   time,	
   from	
   the	
   1800s	
   to	
   the	
  
present.	
  	
  

Specialized	
   linguistic	
   work	
   from	
   Google	
   Ngram	
   Viewer	
   has	
   to	
   date	
   focused	
   on	
   cultural	
   trends	
  
throughout	
  history,	
  seen	
  through	
  the	
  books	
  that	
  have	
  been	
  scanned	
  by	
  Google.	
  The	
  Culturomics	
  team	
  looks	
  
quantitatively	
  at	
  linguistic	
  and	
  cultural	
  phenomena	
  reflected	
  in	
  the	
  language	
  of	
  these	
  texts.	
  This	
  reporting	
  of	
  
linguistic	
   trends	
   is	
   directly	
   related	
   to	
   the	
   fields	
   of	
   “lexicography,	
   the	
   evolution	
   of	
   grammar,	
   collective	
  
memory,	
  the	
  adoption	
  of	
  technology,	
  the	
  pursuit	
  of	
  fame,	
  censorship,	
  and	
  historical	
  epidemiology”	
  (Michel	
  et	
  
al.,	
   2011,	
   p.176).	
   Some	
   lexical	
   and	
   culturally-­‐specific	
   data	
   intended	
   for	
   popular	
   reading	
   obtained	
   from	
   the	
  
Google	
  Ngram	
  Viewer	
  and	
  Culturomics	
  include	
  the	
  following	
  (from	
  Cohen,	
  2010):	
  

-­‐ women	
  in	
  comparison	
  with	
  men	
  is	
  rarely	
  mentioned	
  until	
  the	
  early	
  1970s,	
  when	
  feminism	
  gained	
  a	
  
foothold.	
  The	
  two	
  topic	
  lines	
  eventually	
  cross	
  paths	
  about	
  1986.	
  	
  

-­‐ Mickey	
  Mouse	
  and	
  Marilyn	
  Monroe	
  do	
  not	
  get	
  nearly	
  as	
  much	
  attention	
  in	
  books	
  as	
  Jimmy	
  Carter.	
  
-­‐ There	
  are	
  many	
  more	
  references	
  in	
  English	
  than	
  in	
  Chinese	
  to	
  Tiananmen	
  Square	
  after	
  1989.	
  
-­‐ Grilling	
  has	
  been	
  used	
  frequently	
  from	
  the	
  late	
  1990s	
  and	
  outpaced	
  roasting	
  and	
  frying	
  in	
  2004.	
  

	
  
Michel	
   et	
   al.	
   (2011)	
   also	
  measured	
   the	
   endurance	
   of	
   fame	
   from	
   the	
   1800s,	
   reporting	
   that	
  written	
  

references	
  to	
  popular	
  celebrities	
  faded	
  twice	
  as	
  quickly	
  in	
  the	
  mid-­‐20th	
  century	
  as	
  they	
  did	
  in	
  the	
  early	
  19th.	
  
They	
  also	
  found	
  technological	
  advances	
  and	
  inventions	
  (e.g.,	
  telephone,	
  radio)	
  took,	
  on	
  average,	
  66	
  years	
  to	
  
be	
  adopted	
  by	
  the	
  larger	
  culture	
  in	
  the	
  early	
  1800s,	
  and	
  only	
  27	
  years	
  between	
  1880	
  and	
  1920.	
  They	
  tracked	
  
the	
  way	
  irregular	
  English	
  verbs	
  that	
  did	
  not	
  add	
  -­‐ed	
  at	
  the	
  end	
  for	
  past	
  tense	
  (i.e.,	
  learnt)	
  evolved	
  to	
  conform	
  
to	
  the	
  common	
  pattern	
  (learned).	
  And	
  finally,	
  as	
  an	
  application	
  of	
  quantitative,	
  lexico-­‐syntactic	
  data,	
  Michel	
  et	
  
al.	
  estimated	
  that	
  the	
  English	
  lexicon	
  has	
  grown	
  by	
  about	
  70	
  percent	
  to	
  more	
  than	
  a	
  million	
  words	
  in	
  the	
  last	
  
50	
   years	
  with	
  direct	
   implications	
   to	
   lexicographic	
   changes	
   and	
  updating	
  dictionaries	
   by	
  pinpointing	
  newly	
  
popular	
  words	
  and	
  obsolete	
  ones	
  (Cohen,	
  2010).	
  

	
  
3.	
  More	
  research	
  with	
  mega-­‐corpora	
  
	
  
3.1.	
  Focus	
  of	
  the	
  present	
  case	
  study	
  

There	
  are,	
  however,	
  possible	
  lines	
  of	
  research	
  using	
  mega-­‐corpora	
  that	
  move	
  beyond	
  culture-­‐based	
  
analyses	
  and	
  diachronic	
  studies.	
  The	
  remainder	
  of	
  this	
  paper	
  is	
  dedicated	
  to	
  an	
  exploratory	
  case	
  study	
  that	
  
shows	
   the	
   applications	
   of	
   Google	
   Ngram	
   Viewer	
   and	
   COHA	
   in	
   the	
   analysis	
   of	
   academic	
   word	
   lists,	
   in	
  
particular,	
  Coxhead’s	
  (2000)	
  Academic	
  Word	
  List	
  (AWL).	
  The	
  goal	
  is	
  not	
  to	
  compare	
  these	
  mega-­‐corpora,	
  but	
  
to	
  add	
  to	
  the	
  growing	
  number	
  of	
  studies	
  that	
  explore	
  the	
  applications	
  of	
  new	
  data	
  sources	
  that	
  may	
  serve	
  as	
  
models	
  for	
  more	
  specialized	
  analyses	
  of	
  COHA	
  and	
  Google	
  Ngram	
  Viewer.	
  	
  

Word	
   lists	
   are	
   a	
   conventional	
   concept	
   common	
   in	
   corpus-­‐based	
   studies	
   with	
   emphasis	
   on	
  
pedagogical	
  applications	
  of	
  corpus	
  data	
  especially	
  in	
  the	
  teaching	
  of	
  academic	
  writing	
  in	
  English.	
  The	
  study	
  of	
  
vocabulary	
  use	
  predates	
  other	
  areas	
  of	
  corpus	
  investigation	
  (Biber,	
  Conrad,	
  &	
  Reppen,	
  1998)	
  and	
  vocabulary	
  
teaching	
   materials	
   developed	
   from	
   corpora	
   have	
   been	
   commonly	
   used	
   in	
   many	
   writing	
   classrooms	
   and	
  
incorporated	
  in	
  dictionaries	
  and	
  textbooks	
  (McEnery,	
  Xiao,	
  &	
  Tono,	
  2006).	
  The	
  AWL	
  has	
  been	
  widely	
  used	
  in	
  
English	
  for	
  Academic	
  Purposes	
  (EAP)	
  and	
  the	
  teaching	
  of	
  second	
  language	
  writing	
  in	
  English	
  (Reppen,	
  2010).	
  
Coxhead	
  (2000,	
  2011)	
  has	
  called	
  for	
  further	
  research	
  on	
  the	
  AWL	
  with	
  larger	
  corpora,	
  noting	
  that	
  there	
  is	
  a	
  



FRIGINAL,	
  WALKER,	
  &	
  RANDALL	
  
	
  
	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
   55	
  

continuing	
  need	
  to	
  address	
  learners’	
  use	
  of	
  academic	
  vocabulary	
  for	
  the	
  AWL	
  to	
  be	
  useful	
  in	
  various	
  contexts	
  
over	
   a	
  decade	
   after	
   its	
   publication.	
  Google	
  Ngram	
  Viewer	
   and	
  COHA	
   could	
  provide	
   this	
   opportunity	
   for	
   an	
  
update,	
  as	
   these	
  mega-­‐corpora	
  have	
   the	
  volume	
  of	
  words	
  and	
  a	
  variety	
  of	
   registers	
  needed	
   to	
  check	
  actual	
  
patterns	
  of	
  lexical	
  distributions.	
  	
  

	
  
3.2.	
  The	
  Academic	
  Word	
  List	
  

The	
   AWL	
   was	
   created	
   by	
   Averil	
   Coxhead,	
   School	
   of	
   Linguistics	
   and	
   Applied	
   Language	
   Studies	
   at	
  
Victoria	
  University	
  of	
  Wellington,	
  as	
  a	
  rationalized,	
  more	
  specialized	
  extension	
  of,	
  or	
  perhaps	
  response	
  to,	
  the	
  
General	
  Service	
  List	
  (GSL).	
  The	
  GSL	
  is	
  a	
  list	
  of	
  over	
  2000	
  words	
  that	
  was	
  developed	
  by	
  Michael	
  West	
  in	
  1953	
  
and	
   updated	
   by	
   John	
   Bauman	
   and	
   Brent	
   Culligan	
   in	
   1995.	
   The	
   list	
   includes	
   the	
   most	
   frequent	
   words	
   of	
  
written	
   English	
   collected	
   primarily	
   for	
   English	
   language	
   learners	
   and	
   ESL	
   writing	
   teachers.	
   The	
   updated	
  
version	
   of	
   the	
   GSL	
   uses	
   frequencies	
   from	
   the	
   Brown	
   Corpus	
   (Bauman,	
   1995).	
  Words	
   from	
   the	
   AWL	
   came	
  
instead	
  from	
  an	
  “Academic	
  Corpus”	
  that	
  Coxhead	
  collected	
  herself.	
  The	
  corpus	
  contains	
  a	
  total	
  of	
  3.5	
  million	
  
words	
   with	
   texts	
   representing	
   multiple	
   academic	
   sources	
   such	
   as	
   journals	
   and	
   textbooks	
   published	
   from	
  
1993	
  to	
  1996.	
  Coxhead	
  also	
  included	
  texts	
  from	
  the	
  Brown	
  Corpus,	
  London/Oslo/Bergen	
  Corpus	
  (LOB),	
  and	
  
MicroConcord	
  Academic	
  Corpus.	
  The	
  four	
  predominant	
  academic	
  registers	
  include	
  Arts,	
  Commerce,	
  Law,	
  and	
  
Science,	
  and	
  encompass	
  28	
  different	
  sub-­‐registers	
  (Coxhead,	
  2000,	
  2002).	
  	
  

An	
  automated	
  search	
  of	
  the	
  Academic	
  Corpus	
  yielded	
  the	
  words	
  comprising	
  the	
  AWL,	
  which	
  consists	
  
of	
   570	
   word	
   families,	
   and	
   3,110	
   individual	
   words.	
   The	
   words	
   were	
   selected	
   based	
   on	
   their	
   specialized	
  
occurrence,	
  range,	
  and	
  frequency,	
  and	
  only	
  included	
  if	
  not	
  already	
  present	
  in	
  the	
  first	
  2,000	
  words	
  of	
  the	
  GSL.	
  
The	
  total	
  number	
  of	
  words	
  on	
  the	
  list	
  results	
  in	
  10%	
  of	
  all	
  the	
  words	
  in	
  academic	
  texts.	
  As	
  a	
  rule,	
  each	
  word	
  
had	
  to	
  be	
  used	
  at	
  least	
  100	
  times	
  in	
  the	
  Academic	
  Corpus.	
  Table	
  1	
  shows	
  sample	
  entries	
  from	
  the	
  AWL	
  with	
  
each	
  lemma	
  listed	
  on	
  the	
  left	
  and	
  all	
  of	
  its	
  lexemes	
  listed	
  underneath	
  each	
  lemma.	
  The	
  bolded	
  words	
  were	
  the	
  
most	
  frequent	
  on	
  the	
  sub-­‐list.	
  The	
  most	
  frequent	
  word	
  in	
  each	
  word	
  family	
  could	
  be	
  the	
  lemma	
  (like	
  the	
  word	
  
estimate	
  seen	
  below)	
  or	
  it	
  could	
  be	
  one	
  of	
  the	
  lexemes	
  (like	
  the	
  word	
  derived	
  also	
  below).	
  The	
  AWL	
  has	
  10	
  
sublists,	
  with	
  Sublist	
  1	
  containing	
  the	
  highest	
  frequency	
  words.	
  

	
  
Table 1 
Sample sublist from AWL 

 derive 
   derivation 
   derivations 
   derivative 
   derivatives 
   derived 
   derives 
   deriving 
  
distribute 
   distributed 
   distributing 
   distribution 
   distributional 

estimate 
   estimated 
   estimating 
   estimation 
   estimations 
   over-estimate 
   overestimate 
   overestimated 
   overestimates 
   overestimating 
   underestimate 
   underestimated 
   underestimates 
   underestimating 
  

function 
   functional 
   functionally 
   functioned 
   functioning 
   functions 
  
identify 
   identifiable 
   identification 
   identified 
   identifies 
   identifying 
   identities 
   identity 

	
  
As	
  noted	
  above,	
  Coxhead’s	
  primary	
  reason	
  for	
  creating	
  the	
  AWL	
  was	
  that	
  other	
  word	
  lists,	
  especially	
  

the	
  GSL,	
  were	
   limited	
   in	
   their	
   capacity	
   to	
  demonstrate	
  current	
   lexical	
  usage	
  across	
  academic	
  registers.	
  Yet,	
  
after	
  more	
  than	
  10	
  years,	
   it	
   is	
  possible	
  that	
  the	
  AWL	
  may	
  also	
  need	
  some	
  updating.	
  For	
  example,	
  Ming-­‐Tzu	
  
and	
   Nation	
   (2004)	
   completed	
   a	
   study	
   on	
   homographs	
  within	
   the	
   AWL	
   and	
   concluded	
   that	
   the	
   list	
   should	
  
include	
   a	
   wider	
   range	
   of	
   word	
   families	
   and	
   lemmas,	
   representing	
   a	
   range	
   of	
   academic	
   homographs.	
   In	
  
addition,	
  Nation	
  and	
  Waring	
  (1997)	
  also	
  argued	
  that	
  in	
  order	
  to	
  comprehend	
  a	
  text,	
  95%	
  to	
  98%	
  of	
  words	
  in	
  
the	
  text	
  must	
  be	
  fully	
  understood	
  and	
  acquired	
  by	
  learners.	
  Unfortunately,	
  as	
  Nation	
  and	
  Waring	
  found,	
  word	
  
lists	
  such	
  as	
  the	
  AWL	
  and	
  GSL	
  arguably	
  do	
  not	
  represent	
  at	
  least	
  95%	
  of	
  words	
  in	
  a	
  target	
  text.	
  



EXPLORING	
  MEGA-­‐CORPORA	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
  56	
  

3.3.	
  Data	
  collection	
  
The	
   overarching	
   research	
   question	
   of	
   this	
   exploratory	
   case	
   study	
   is:	
   How	
   are	
  words	
   on	
   the	
   AWL	
  

reflected	
  and	
  distributed	
  across	
  recent	
  time	
  frames	
  in	
  Google	
  Ngram	
  Viewer	
  and	
  COHA?	
  In	
  pursuing	
  this	
  line	
  
of	
  research,	
  it	
  may	
  also	
  be	
  possible	
  to	
  answer	
  whether	
  or	
  not	
  there	
  is	
  evidence	
  that	
  an	
  updated	
  word	
  list	
  is	
  
necessary.	
  The	
  words	
   analyzed	
   in	
   the	
   study	
  are	
   from	
  AWL	
  Sublist	
  1.	
  This	
   list	
   contains	
  60	
  words	
   and	
   their	
  
word	
   families.	
   The	
  most	
   frequent	
  word	
   from	
   each	
   lemma	
  was	
   selected	
   for	
   frequency	
   searches	
   in	
   the	
   two	
  
mega-­‐corpora	
  (see	
  Appendix	
  for	
  the	
  complete	
  list	
  of	
  60	
  words	
  used).	
  	
  

Google	
  Ngram	
  Viewer	
  was	
  searched	
  to	
  find	
  words	
  with	
  “increase	
  or	
  decrease”	
  counts	
  from	
  1998	
  to	
  
2008	
   that	
   are	
   greater	
   than	
   or	
   equal	
   to	
   .0050.	
   From	
   these	
   results,	
   10	
   words	
   were	
   chosen	
   to	
   be	
   further	
  
analyzed	
   qualitatively.	
   One	
   of	
   the	
   ten	
   words,	
   for	
   example,	
   was	
   labour,	
   which	
   was	
   found	
   to	
   have	
   a	
   rapid	
  
decrease	
  in	
  the	
  given	
  time	
  frame.	
  However,	
  since	
  labour	
  is	
  typically	
  British	
  English	
  and	
  not	
  American	
  English,	
  
it	
  was	
  excluded	
  from	
  this	
  case	
  study	
  (as	
  COHA	
  represents	
  only	
  American	
  English).	
  Because	
  of	
  that	
  exclusion,	
  
nine	
  words	
  were	
  ultimately	
   selected	
   for	
   distributional	
   analysis	
   using	
  Google	
  Ngram	
  Viewer:	
  analysis,	
   data,	
  
economic,	
  environment,	
  policy,	
  research,	
  section,	
  structure,	
  and	
  theory.	
  COHA,	
  from	
  the	
  same	
  time	
  frame,	
  was	
  
used	
   for	
   economic,	
   evidence,	
   individual,	
   major,	
   percent,	
   period,	
   principle,	
   section,	
   and	
   theory.	
   Words	
   with	
  
frequencies	
   with	
   a	
   difference	
   greater	
   than	
   or	
   equal	
   to	
   20	
   tokens	
   per	
  million	
   were	
   selected	
   for	
   additional	
  
analysis.	
   This	
   was	
   done	
   as	
   COHA	
   lists	
   actual	
   and	
   normalized	
   frequencies	
   compared	
   to	
   percentages	
   from	
  
Google	
  Ngram	
  Viewer.	
  

	
  
4.	
  Results	
  
	
  
4.1.	
  Macro-­‐level	
  distributions	
  of	
  15	
  AWL	
  words	
  from	
  Google	
  Ngram	
  Viewer	
  

Of	
  the	
  60	
  AWL	
  words	
  analyzed	
  in	
  this	
  study	
  for	
  both	
  platforms,	
  more	
  than	
  90%	
  are	
  shown	
  to	
  have	
  
declined	
  in	
  use	
  from	
  the	
  late-­‐1990s	
  to	
  the	
  present.	
  Fifteen	
  words	
  from	
  AWL	
  were	
  found	
  to	
  have	
  higher	
  levels	
  
of	
  differences	
  from	
  the	
  set	
  of	
  results	
  obtained	
  from	
  Google	
  Ngram	
  Viewer	
  and	
  COHA.	
  These	
  are:	
  (1)	
  analysis,	
  
(2)	
  data,	
  (3)	
  economic,	
  (4)	
  environment,	
  (5)	
  evidence,	
  (6)	
  individual,	
  (7)	
  major,	
  (8)	
  percent,	
  (9)	
  period,	
  (10)	
  
policy,	
   (11)	
   principle,	
   (12)	
   research,	
   (13)	
   section,	
   (14)	
   structure,	
   and	
   (15)	
   theory.	
   Figure	
   3	
   shows	
   Google	
  
Ngram	
  Viewer’s	
  word	
  chart	
  tracking	
  and	
  distributions	
  of	
  some	
  AWL	
  words	
  from	
  the	
  1900s	
  to	
  2000s.	
  	
  	
  
 

 
Figure 3. Sample distributional chart tracking of AWL words from Google Ngram Viewer, 1900s to 2000s 

 



FRIGINAL,	
  WALKER,	
  &	
  RANDALL	
  
	
  
	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
   57	
  

Specifically,	
   for	
   the	
   15	
   words	
   listed	
   above,	
   Figures	
   4	
   and	
   5	
   show	
   a	
   slow	
   decline	
   in	
   their	
   general	
  
frequencies	
  from	
  English	
  books	
  from	
  1998	
  to	
  2008	
  from	
  Google	
  Ngram	
  Viewer.	
  	
  
	
  

 
Figure 4. Frequency decline of words 1-7: analysis, data, economic, environment, evidence, individual, major from Google 
Ngram Viewer, 1998 to 2008. 

	
  

	
  
Figure 5. Frequency decline of words 8-15: percent, period, policy, principle, research, section, structure, theory from Google 
Ngram Viewer, 1998 to 2008 

	
  



EXPLORING	
  MEGA-­‐CORPORA	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
  58	
  

The	
  graphical	
  frequency	
  decline	
  of	
  these	
  common	
  AWL	
  words	
  as	
  illustrated	
  by	
  outputs	
  produced	
  by	
  
Google	
   Ngram	
   Viewer	
   suggests	
   that	
   there	
  may	
   be	
   a	
   need	
   to	
   update	
   AWL	
   and	
   similar	
  word	
   lists,	
   by	
   using	
  
mega-­‐corpora.	
  Academic	
  words	
  that	
  are	
  trending	
  higher	
  in	
  the	
  past	
  10	
  years	
  should	
  be	
  added.	
  A	
  quick	
  search	
  
for	
   words	
   such	
   as	
   software,	
   interface,	
   develop,	
   and	
   document	
   in	
   Google	
   Ngram	
   Viewer	
   shows	
   increasing	
  
frequencies	
  of	
  these	
  words	
  in	
  English	
  books.	
  Word	
  lists	
  heavily	
  rely	
  on	
  frequency	
  of	
  use,	
  which	
  could	
  be	
  more	
  
accurately	
  provided	
  by	
   larger	
  and	
  more	
  representative	
  databases.	
  While	
  English	
  books	
   from	
  Google	
  Ngram	
  
Viewer	
  clearly	
  do	
  not	
   represent	
   the	
  entirety	
  of	
  academic	
  writing	
  registers,	
   these	
   figures	
  give	
  a	
  macro-­‐level	
  
look	
   at	
   frequency	
   distributions	
   and	
   suggest	
   how	
   these	
   could	
   be	
   used	
   to	
   gauge	
   actual	
   usage	
   of	
   words	
   in	
  
published	
  texts	
  across	
  recent	
  time	
  periods.	
  This	
  macro-­‐level	
  data	
  could	
  then	
  be	
  used	
  to	
  further	
  track	
  specific	
  
lemmas	
   in	
   a	
   more	
   specialized	
   corpus	
   of	
   academic	
   writing	
   with	
   clearly	
   identified	
   sub-­‐registers	
   such	
   as	
  
disciplines,	
   text	
   types,	
   or	
   other	
   genres	
   (e.g.,	
   research	
   report,	
   freshmen	
   composition,	
   and	
   second	
   language	
  
writing).	
  	
  	
  

	
  
4.2.	
  Comparison	
  of	
  Google	
  Ngram	
  Viewer	
  and	
  COHA	
  results	
  

Together,	
  the	
  results	
  from	
  COHA	
  and	
  Google	
  Ngram	
  validate	
  an	
  overall	
  decline	
  of	
  frequencies	
  for	
  the	
  
AWL	
   Sublist	
   1	
   from	
   1998	
   to	
   2008	
   (from	
   1990	
   in	
   COHA).	
   Thirteen	
   of	
   the	
   15	
  words	
   from	
   COHA	
   produced	
  
trending	
  data	
  comparable	
  to	
  Google	
  Ngram	
  Viewer’s	
  output.	
  Discrepancies	
  are	
  found	
  in	
  the	
  tracking	
  numbers	
  
for	
  two	
  AWL	
  words:	
  research	
  and	
  data.	
  Research	
  shows	
  a	
  considerable	
  increase	
  in	
  COHA	
  from	
  164.16	
  tokens	
  
normalized	
  per	
  million	
  to	
  180.91	
  tokens	
  per	
  million.	
  This	
  result	
  is	
  in	
  contrast	
  to	
  what	
  Google	
  Ngram	
  Viewer	
  
found	
  for	
  research	
  as	
  shown	
  in	
  the	
  line	
  graph	
  (Figure	
  6)	
  below.	
  Data	
  in	
  COHA	
  has	
  a	
  very	
  slight	
  increase	
  from	
  
101.00	
  to	
  101.60	
  tokens	
  per	
  million	
  compared	
  to	
  Google	
  Ngram	
  Viewer’s	
  declining	
  line	
  graph.	
  

	
  

	
  
Figure 6. Frequency data for research, 1998 to 2008 from Google Ngram Viewer 

 
Although	
  COHA	
  and	
  Google	
  Ngram	
  Viewer	
  show	
  similar	
  overall	
   trends	
   for	
  most	
  of	
   the	
  words	
   from	
  

AWL	
  Sublist	
  1,	
  in	
  addition	
  to	
  the	
  discrepancies	
  noted	
  above,	
  there	
  are	
  also	
  observable	
  differences	
  in	
  the	
  micro	
  
distributions	
   of	
   the	
   15	
   words	
   examined	
   in	
   this	
   case	
   study.	
   These	
   differences	
   may	
   be	
   due	
   to	
   the	
   clear	
  
distinction	
  between	
   the	
   texts	
   comprising	
   these	
   two	
  mega-­‐corpora	
  or	
  possibly	
   from	
   the	
  major	
  difference	
   in	
  
total	
  word	
  counts.	
  Table	
  2	
  lists	
  words	
  with	
  the	
  greatest	
  difference	
  in	
  each	
  corpus	
  (all	
  declining	
  in	
  frequency),	
  
with	
  the	
  words	
  in	
  bold	
  indicating	
  that	
  the	
  word	
  appears	
  on	
  both	
  lists	
  and,	
  therefore,	
  exhibits	
  a	
  very	
  similar	
  
trend.	
  
	
  



FRIGINAL,	
  WALKER,	
  &	
  RANDALL	
  
	
  
	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
   59	
  

Table 2 
Comparison of micro differences between Google Ngram Viewer and COHA 

 Google Ngram Viewer COHA 
1 data section 
2 economic economic 
3 analysis percent 
4 policy major 
5 theory theory 
6 structure period 
7 environment evidence 
8 research individual 
9 section principle 

	
  
Interestingly,	
  data,	
  which	
  shows	
  a	
  slight	
   increase	
   in	
  COHA,	
  displays	
   the	
  biggest	
  decrease	
   in	
  Google	
  

Ngram	
  Viewer.	
  Section	
  has	
  the	
  biggest	
  decrease	
  in	
  COHA	
  -­‐	
  151	
  tokens	
  per	
  million	
  in	
  the	
  1990s	
  to	
  68	
  tokens	
  
per	
   million	
   in	
   the	
   2000s	
   –	
   while	
   it	
   has	
   the	
   lowest	
   decrease	
   in	
   Google	
   Ngram	
   Viewer.	
   Economic,	
   which	
  
demonstrates	
  the	
  second	
  biggest	
  decline	
  in	
  both	
  COHA	
  and	
  Google	
  Ngram	
  Viewer,	
  and	
  theory,	
  which	
  has	
  the	
  
fifth	
  highest	
  decrease	
  for	
  both	
  mega-­‐corpora,	
  are	
  the	
  other	
  words	
  in	
  common	
  between	
  the	
  two.	
  	
  
	
  
4.3.	
  Micro	
  comparison	
  of	
  section,	
  theory,	
  and	
  economic	
  lemmas/lexemes	
  across	
  corpora	
  

A	
  closer	
  look	
  at	
  the	
  lemmas	
  and	
  lexemes	
  of	
  the	
  three	
  words	
  that	
  showed	
  similar	
  waning	
  outcomes	
  in	
  
both	
  corpora	
  is	
  presented	
  below,	
  in	
  order	
  to	
  show	
  how	
  word	
  families	
  could	
  be	
  tracked	
  and	
  compared	
  further	
  
across	
   time	
  periods	
   and	
   possibly	
   sub-­‐registers.	
   Table	
   3	
   shows	
   the	
   lemmas	
   and	
   lexemes	
   from	
   the	
  AWL	
   for	
  
each	
  of	
  the	
  three	
  words.	
  

	
  
Table 3 
Comparison of micro differences between Google Ngram Viewer and COHA 

section theory economy 
sectioned theoretical economic 
sectioning theoretically economical 
sections theories economically 
 theorist economics 
 theorists economies 
  economist 
  economists 
  uneconomical 

	
  
Results	
  from	
  the	
  declining	
  lemma,	
  section,	
  show	
  that	
  only	
  section	
  and	
  sections	
  are	
  used	
  commonly	
  in	
  

COHA.	
  Both	
  sectioning	
  and	
  sectioned	
  did	
  not	
  clearly	
  register	
  in	
  the	
  graphical	
  outputs,	
  although	
  a	
  closer	
  look	
  at	
  
the	
  numbers	
   shows	
   that	
   the	
  use	
  of	
   sectioned	
   very	
   slightly	
   increased	
  during	
   this	
  period.	
  Figure	
  7	
   illustrates	
  
COHA’s	
  data	
  for	
  these	
  four	
  lexemes,	
  comparing	
  distributions	
  from	
  1990	
  and	
  2000.	
  

	
  



EXPLORING	
  MEGA-­‐CORPORA	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
  60	
  

	
  
Figure 7. COHA results for section 

	
  
The	
  distributions	
  of	
  theory	
  and	
  its	
  lexemes	
  from	
  Google	
  Ngram	
  Viewer	
  are	
  shown	
  in	
  Figure	
  8.	
  Theory	
  

is	
   the	
   most	
   frequently	
   used	
   word,	
   followed	
   by	
   theoretical	
   and	
   theories.	
   Consistent	
   with	
   the	
   overall	
   AWL	
  
findings,	
   these	
  words	
   also	
   demonstrate	
   a	
   general	
   decline.	
  Theoretically,	
   theorist,	
   and	
   theorists	
   exhibit	
   very	
  
minimal	
  frequencies	
  of	
  use	
  across	
  English	
  books	
  from	
  this	
  ten-­‐year	
  period.	
  
	
  

	
  
Figure 8. Google Ngram Viewer results for theory and its lexemes  

	
  
Figure	
  9	
  from	
  COHA	
  displays	
  a	
  comparatively	
  similar	
  result	
  to	
  data	
  from	
  Google	
  Ngram	
  Viewer.	
  Yet,	
  

one	
  of	
  COHA’s	
  advantages	
  over	
  Google	
  Ngram	
  Viewer	
  in	
  this	
  kind	
  of	
  micro	
  comparison	
  is	
  that	
  COHA	
  returns	
  
normalized	
   tokens	
   that	
   show	
   specific	
   and	
   easily	
   interpretable	
   numbers.	
   For	
   example,	
   the	
   lexeme	
   theorist	
  
increased	
   from	
   1.25	
   tokens	
   per	
   million	
   to	
   1.79	
   tokens	
   per	
   million	
   from	
   1990	
   to	
   2000,	
   which	
   was	
   not	
  



FRIGINAL,	
  WALKER,	
  &	
  RANDALL	
  
	
  
	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
   61	
  

observable	
  from	
  Ngram’s	
  figures	
  in	
  Google.	
  Overall,	
  theoretical,	
  theoretically,	
  theorist,	
  and	
  theorists	
  were	
  not	
  
common	
  in	
  the	
  two	
  corpora.	
  For	
  the	
  purpose	
  of	
  this	
  paper	
  -­‐	
  analyzing	
  AWL	
  -­‐	
  both	
  Google	
  Ngram	
  Viewer	
  and	
  
COHA	
   show	
   a	
   decline	
   in	
   the	
   frequently	
   employed	
  words	
   theory	
   and	
   theories,	
   while	
   less	
   common	
   lexemes	
  
stabilize	
  at	
  a	
  very	
  low	
  frequency	
  or	
  with	
  an	
  occasional,	
  very	
  slight	
  increase	
  in	
  use.	
  

	
  

	
  
Figure 9. COHA results for theory ad its lexemes 

	
  
Unlike	
  section	
  and	
  theory,	
  economic	
  is	
  not	
  a	
  lemma,	
  but	
  one	
  of	
  the	
  lexemes	
  of	
  economy.	
  Economic	
  and	
  

economy	
   are	
   the	
   two	
  most	
   commonly	
   used	
   words	
   in	
   this	
   word	
   family.	
   The	
   remaining	
   words,	
   economical,	
  
economically,	
  economics,	
  economies,	
  economist,	
  and	
  economists	
  are	
  not	
  as	
  common,	
  with	
  less	
  than	
  20	
  tokens	
  
per	
  million	
  (except	
  for	
  economics	
  in	
  1990)	
  in	
  COHA	
  (Figure	
  10).	
  	
  

	
  

 
Figure 10. COHA results for economic and its lexemes 

	
  



EXPLORING	
  MEGA-­‐CORPORA	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
  62	
  

All	
   lexemes	
   in	
   Figure	
   10	
   consistently	
   show	
   a	
   decline	
   (except	
   for	
   uneconomical	
   which	
   shows	
   no	
  
change	
  at	
   all,	
   but	
  did	
  not	
   register	
  bars	
   in	
   the	
   figure	
  because	
  of	
   very	
  minimal	
  normalized	
   tokens	
   found).	
   In	
  
sum,	
  Google	
  Ngram	
  Viewer	
  and	
  COHA	
  results	
  indicate	
  that	
  the	
  lemma	
  groups	
  of	
  section,	
  theory,	
  and	
  economic	
  
follow	
  very	
  similar	
  overall	
  tracking	
  trends.	
  The	
  most	
  frequent	
  lexeme	
  is	
  followed	
  by	
  another	
  that	
  is	
  somewhat	
  
frequently	
  used,	
  both	
  of	
  which	
  are	
  in	
  decline.	
  These	
  two	
  lexemes	
  are	
  then	
  followed	
  by	
  a	
  variety	
  of	
  much	
  less	
  
frequent	
  words	
  in	
  the	
  family	
  that	
  are	
  either	
  rarely	
  employed	
  or	
   in	
  decline.	
  The	
  COHA	
  tables	
  do	
  show	
  slight	
  
declines	
   in	
   less	
   frequent	
   theory	
   and	
   economy	
   lexemes,	
   while	
   for	
   a	
   couple	
   of	
   lexemes	
   (e.g.,	
   theorist,	
  
uneconomical),	
   a	
   slight	
   fall	
   or	
   slight	
   rise	
   is	
   also	
   observed.	
   Overall,	
   what	
   can	
   be	
   seen	
   from	
   these	
   micro-­‐
comparisons	
  explored	
  in	
  this	
  case	
  study	
  are	
  patterns	
  exhibiting	
  the	
  consistent	
  decline	
  in	
  use	
  of	
  the	
  words	
  on	
  
Sublist	
  1	
  of	
  the	
  Academic	
  Word	
  List.	
  	
  
	
  
5.	
  Conclusion	
  	
  

This	
   exploratory	
   case	
   study	
   suggests	
   that	
  many	
   of	
   the	
   commonly-­‐used	
  words	
   listed	
   as	
   part	
   of	
   the	
  
Academic	
  Word	
  List	
  (Coxhead,	
  2000)	
  are	
  declining	
  in	
  overall	
  frequency	
  of	
  use	
  when	
  investigated	
  on	
  a	
  macro	
  
level	
  using	
  mega-­‐corpora.	
  Both	
  Google	
  Ngram	
  Viewer	
  and	
  COHA	
  generally	
  confirm	
  this	
  trend.	
  Frequency	
  data	
  
from	
  these	
  tools	
  using	
  normalized	
  tokens	
  and	
  percentages	
  support	
  each	
  other	
  when	
  presented	
  through	
  visual	
  
plots	
  and	
  estimates,	
  and,	
  therefore,	
  further	
  attest	
  to	
  their	
  general	
  consistency.	
  It	
  is	
  reasonable	
  to	
  suggest	
  then	
  
that	
  Google	
  Ngram	
  Viewer	
  and	
  COHA	
  are	
  viable	
  resources	
  for	
  macro	
  linguistic	
  research	
  with	
  AWL	
  and	
  similar	
  
wordlists.	
  Further	
  microanalysis	
  of	
  word	
  distributions	
  indicates	
  that	
  less-­‐frequent	
  lexemes	
  have	
  maintained	
  a	
  
fairly	
   stable	
   rate	
   of	
   low	
   usage.	
   Even	
  with	
   very	
  minimal	
   tracking	
   of	
   normalized	
   tokens	
   and	
   percentages	
   of	
  
these	
   less-­‐frequent	
   lexemes,	
   both	
   Google	
   Ngram	
   Viewer	
   and	
   COHA	
   are	
   able	
   to	
   produce	
   important	
  
comparative	
  data	
  across	
  a	
  variety	
  of	
  words	
  useful	
  in	
  projecting	
  vocabulary	
  shifts	
  and	
  changes.	
  	
  	
  

These	
  results	
  of	
  an	
  overall	
  decreasing	
  rate	
  of	
  use	
  of	
  common	
  AWL	
  words	
  are	
  not	
  surprising.	
  Linguists	
  
have	
  long	
  noted	
  that	
  words	
  have	
  a	
  finite	
  life	
  span	
  (Crystal,	
  2006).	
  Michel	
  et	
  al.	
  (2011)	
  noted	
  that	
  word	
  usage	
  
is	
   typically	
   characterized	
   by	
   a	
   spike	
   at	
   some	
   point,	
   then	
   directly	
   followed	
   by	
   a	
   slow	
   decline.	
   Occasionally,	
  
words	
  revive	
  and	
  have	
  a	
  short	
  comeback	
  at	
  times,	
  again	
  followed	
  by	
  a	
  decline	
  measured	
  through	
  frequency	
  
distributions.	
  The	
  results	
  from	
  this	
  study	
  adhered	
  to	
  these	
  general	
  principles.	
   	
  

	
  
5.1.	
  Pedagogical	
  implications	
  

This	
  declining	
  trend	
  in	
  vocabulary	
  use,	
  when	
  measured	
  in	
  a	
  relatively	
  short	
  timeframe	
  (e.g.,	
  1990	
  to	
  
2008)	
  is	
  very	
  relevant	
  if	
  applied	
  to	
  standard	
  word	
  lists	
  used	
  for	
  language	
  teaching.	
  As	
  the	
  AWL	
  has	
  been	
  used	
  
as	
   a	
   reference	
   for	
   teachers	
   of	
   academic	
   writing	
   for	
   a	
   range	
   of	
   learners,	
   including	
   non-­‐native	
   speakers	
   of	
  
English,	
  accurate	
  distributions	
  representing	
  the	
  present	
  status	
  of	
  word	
  usage	
  in	
  specific	
  contexts	
   is	
  of	
  great	
  
importance.	
   Teachers	
   and	
   curriculum	
   developers,	
   then,	
   might	
   focus	
   on	
   these	
   distributions	
   to	
   match	
   how	
  
patterns	
   of	
   vocabulary	
   are	
   utilized	
   currently	
   outside	
   the	
   classroom.	
   In	
   fields	
   such	
   as	
   English	
   for	
   Specific	
  
Purposes	
  (ESP)	
  and	
  English	
  for	
  Academic	
  Purposes	
  (EAP),	
  these	
  are	
  very	
  important	
  arguments	
  for	
  pedagogy.	
  

Additionally,	
  accompanying	
  the	
  falling	
  frequencies	
  of	
  these	
  AWL	
  words	
  is	
  the	
  consistent	
  rise	
  of	
  new	
  
words.	
  Every	
  year,	
   there	
  are	
  approximately	
  8,500	
  new	
  words	
  added	
   to	
   the	
  English	
   language	
   (Michel	
   et	
   al.,	
  
2011).	
   The	
   constant	
   influx	
   of	
   new	
  words	
   gives	
   a	
  writer	
  more	
   options	
   in	
   presenting	
   ideas	
   and	
   structuring	
  
academic	
  arguments	
  or	
  explanations.	
  Such	
  spread	
  of	
  new	
  words	
  certainly	
  contributes	
  to	
  the	
  variation	
  in	
  word	
  
lists	
  and	
  may	
  cause	
  a	
  slow	
  decrease	
  in	
  overall	
  usage	
  of	
  the	
  words	
  similar	
  to	
  the	
  ones	
  on	
  the	
  AWL.	
  Coxhead’s	
  
AWL	
  appears	
   to	
  be	
  naturally	
  heading	
   in	
   this	
  direction,	
   and	
   there	
  may	
  be	
  a	
  need	
   for	
  a	
  new	
   list,	
   as	
  Coxhead	
  
herself	
   proposed	
   (Coxhead,	
   2011).	
   At	
   the	
   very	
   least,	
   the	
  AWL	
  needs	
   to	
   be	
   consistently	
   updated,	
   since	
   it	
   is	
  
used	
  by	
  many	
  teachers	
  for	
  reference	
  in	
  genre-­‐based	
  instruction,	
  especially	
  in	
  discipline-­‐specific	
  writing.	
  With	
  
the	
   availability	
   of	
   mega-­‐corpora	
   free	
   to	
   public	
   access,	
   there	
   is	
   no	
   reason	
   why	
   a	
   new,	
   expanded	
   word	
   list	
  
cannot	
  be	
  further	
  developed	
  by	
  various	
  groups	
  of	
  researchers.	
  	
  
 
5.2.	
  Comparison	
  between	
  Google	
  Ngram	
  Viewer	
  and	
  COHA	
  

Google	
  Ngram	
  Viewer	
   and	
  COHA	
  are	
   tools	
   that	
  will	
   continue	
   to	
   inspire	
   the	
  production	
  of	
   research	
  
studies	
  on	
   language	
  variation	
  and	
  use.	
  Although	
   this	
   study	
   found	
  minor	
  discrepancies	
   in	
   the	
  distributional	
  
data	
  of	
  some	
  words,	
  both	
  corpora	
  have	
  shown	
  consistently	
  similar	
  patterns	
  of	
  results,	
  especially	
  on	
  a	
  macro	
  
level,	
   even	
  with	
   the	
  obvious	
  difference	
   in	
   total	
  word	
  counts.	
   Statistically,	
  when	
   frequencies	
  are	
  normalized	
  
from	
  these	
  two	
  databases	
  and	
  the	
  investigated	
  linguistic	
  feature	
  is	
  not	
  extremely	
  rare,	
  the	
  large	
  word	
  count	
  
difference	
   has	
   very	
   limited	
   effect	
   (Biber,	
   1993).	
   COHA’s	
   present	
   structure	
   and	
   range	
   of	
   features,	
   however,	
  



FRIGINAL,	
  WALKER,	
  &	
  RANDALL	
  
	
  
	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
   63	
  

make	
   it	
  more	
   ideal	
   for	
   corpus-­‐based	
   research	
   than	
  Google	
  Ngram	
  Viewer.	
   This	
   database	
   is	
  monitored	
   and	
  
regularly	
   updated	
   and	
  Davies	
   provides	
   users	
   extensive	
   “Help”	
   features	
   and	
   a	
   bibliography	
   of	
   COCA/COHA	
  
published	
  studies	
  over	
  the	
  years.	
  COHA’s	
  clearly	
  defined	
  registers	
  and	
  search	
  features	
  such	
  as	
  collocates,	
  POS-­‐
tagged	
  data,	
  and	
  synonyms	
  are	
  grounded	
  in	
  linguistic	
  theory	
  and	
  are	
  well-­‐motivated	
  by	
  previous	
  research	
  in	
  
corpus	
   linguistics.	
   In	
   contrast,	
   Google	
   Ngram	
   Viewer’s	
   raw	
   database	
   has	
   very	
   limited	
   search	
   options	
   and	
  
interactive	
  features	
  at	
  present.	
  However,	
  while	
  impossible	
  to	
  read	
  as	
  individual	
  texts,	
  Google	
  allows	
  users	
  to	
  
download	
   n-­‐grams	
   which	
   could	
   then	
   be	
   processed	
   further	
   using	
   specialized	
   computer	
   programs.	
   COHA’s	
  
database	
   is	
  not	
   available	
   for	
   free	
  download	
   to	
  users.	
  A	
   sampling	
  of	
  Google’s	
   texts	
  of	
  published	
  books	
   from	
  
1500s	
  to	
  the	
  present	
  across	
  major	
  languages	
  would	
  represent	
  a	
  very	
  important,	
  globally	
  available	
  corpus	
  for	
  
extensive	
  linguistic	
  analysis.	
  	
  
	
  
5.3.	
  Directions	
  for	
  future	
  research	
  

Aside	
   from	
  an	
  exploration	
  of	
  AWL	
  word	
  distributions	
   in	
  mega-­‐corpora,	
  one	
  of	
   the	
  primary	
  goals	
  of	
  
this	
  study	
  was	
  to	
  contribute	
  to	
  the	
  growing	
  body	
  of	
  research	
  involving	
  the	
  use	
  of	
  Google	
  Ngram	
  Viewer	
  and	
  
COHA.	
   In	
   the	
   past	
   years,	
   several	
   diachronic	
   language	
   (vocabulary)	
   change	
   studies	
   used	
   these	
   tools	
   quite	
  
successfully,	
   but	
   have	
   mostly	
   focused	
   on	
   cultural	
   and	
   historical	
   shifts.	
   However,	
   systematic,	
   replicable	
  
research	
  methodologies	
  and	
  various	
  options	
  in	
  framing	
  research	
  questions	
  are	
  needed	
  to	
  serve	
  as	
  models	
  for	
  
how	
   these	
   tools	
   could	
   be	
   effectively	
   applied	
   in	
   the	
   fields	
   of	
   linguistics	
   and	
   corpus-­‐based	
   research.	
   More	
  
validation	
  through	
  similar	
  studies	
   is	
  necessary,	
   including	
  the	
  application	
  of	
  statistical	
  significance	
  testing	
  in	
  
examining	
  how	
  normalized	
  tokens	
  and	
  percentages	
  compare,	
  beyond	
  visual	
  representations	
  and	
  line	
  graphs.	
  
Additional	
   comparative	
   research	
  with	
   an	
   extensive	
   range	
   of	
   lexico/grammatical	
   features	
   and	
  demographic	
  
information	
  from	
  texts	
  will	
  also	
  have	
  to	
  be	
  explored	
  further.	
  	
  

Keuleers,	
  Brysbaert,	
   and	
  New	
  (2011)	
   suggest	
   that	
  Google	
  Ngram	
  Viewer	
  can	
  be	
  effectively	
  utilized	
  
for	
  many	
  types	
  of	
  studies	
  in	
  psycholinguistics.	
  There	
  are	
  also	
  potential	
  applications	
  of	
  Google	
  Ngram	
  Viewer	
  
for	
  manipulation	
  using	
  other	
  interfaces	
  that	
  aid	
  dictionary	
  research	
  and	
  linguistic	
  tagging	
  approaches	
  (Sekine	
  
&	
  Dalwani,	
  2010).	
  Like	
  Davies’	
  (2011)	
  attempt	
  with	
  the	
  Google	
  Books	
  interface,	
  other	
  searches	
  with	
  clearly-­‐
defined	
   linguistic	
   foci	
   could	
   be	
   completed	
   in	
   the	
   future.	
   The	
   use	
   of	
   Google’s	
   corpora	
   of	
   books	
   could	
   be	
  
valuable	
   in	
   a	
   variety	
   of	
   natural	
   language	
   processing	
   applications.	
   Results	
   of	
   comparative	
   and	
   contrastive	
  
analysis	
   of	
   linguistic	
   data	
   from	
   these	
   texts	
   across	
   time	
   periods	
  will	
   provide	
   fascinating	
   information	
   about	
  
human	
  language.	
  	
  	
  
	
  
	
  
	
  
References	
  
	
  
Bauman,	
  John	
  (1995).	
  Online.	
  About	
  the	
  General	
  Service	
  List.	
  Available	
  at:	
  http://jbauman.com/aboutgsl.html	
  

(accessed	
  March	
  2011).	
  
Bautista,	
  Lourdes	
  (2011).	
  Studies	
  in	
  Philippine	
  English:	
  Exploring	
  the	
  Philippine	
  component	
  of	
  the	
  International	
  

Corpus	
  of	
  English.	
  Manila,	
  Philippines:	
  De	
  La	
  Salle	
  University	
  Press.	
  
Biber,	
  Douglas	
  (1990).	
  Methodological	
  issues	
  regarding	
  corpus-­‐based	
  analysis	
  of	
  linguistic	
  variation.	
  Literary	
  

and	
  Linguistic	
  Computing,	
  5(4),	
  257-­‐269.	
  
Biber,	
  Douglas	
  (1993).	
  Representativeness	
  in	
  corpus	
  design.	
  Literary	
  and	
  Linguistic	
  Computing,	
  8(4),	
  243-­‐257.	
  
Biber,	
  Douglas	
  (1995).	
  Dimensions	
  of	
  register	
  variation:	
  A	
  cross-­‐linguistic	
  perspective.	
  Cambridge:	
  Cambridge	
  

University	
  Press.	
  
Biber,	
  Douglas,	
  Conrad,	
  Susan	
  &	
  Reppen,	
  Randi	
  (1998).	
  Corpus	
  linguistics:	
  Investigating	
  language	
  structure	
  

and	
  use.	
  Cambridge:	
  Cambridge	
  University	
  Press.	
  
Biber,	
  Douglas,	
  Johansson,	
  Stig,	
  Leech,	
  Geoffrey,	
  Conrad,	
  Susan	
  &	
  Finegan,	
  Edward	
  (1999).	
  Longman	
  grammar	
  

of	
  spoken	
  and	
  written	
  English.	
  Harlow,	
  Essex:	
  Pearson.	
  
Biber,	
  Douglas,	
  Reppen,	
  Randi	
  &	
  Friginal,	
  Eric	
  (2010).	
  Research	
  in	
  corpus	
  linguistics.	
  	
  In	
  Robert	
  Kaplan	
  (Ed.),	
  

The	
  Oxford	
  Handbook	
  of	
  Applied	
  Linguistics,	
  (pp.	
  548-­‐567).	
  Oxford:	
  Oxford	
  University	
  Press.	
  
Bohannon,	
  John	
  (2010).	
  Google	
  opens	
  books	
  to	
  new	
  cultural	
  studies.	
  Science,	
  330,	
  1600.	
  
Bohannon,	
  John	
  (2011).	
  Google	
  books,	
  Wikipedia,	
  and	
  the	
  future	
  of	
  culturomics.	
  Science,	
  331,	
  135.	
  
	
  
	
  



EXPLORING	
  MEGA-­‐CORPORA	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
  64	
  

Cohen,	
  Patricia	
  (2010,	
  December	
  16).	
  New	
  York	
  Times	
  online:	
  In	
  500	
  billion	
  words,	
  new	
  window	
  on	
  culture.	
  
Available	
  at:	
  http://www.nytimes.com/2010/12/17/books/17words.html?emc=eta1	
  (accessed	
  
February	
  2011).	
  

Coxhead,	
  Averil	
  (2000).	
  A	
  new	
  academic	
  word	
  list.	
  TESOL	
  Quarterly,	
  34(2),	
  213-­‐238.	
  
Coxhead,	
  Averil	
  (2002).	
  The	
  academic	
  wordlist:	
  A	
  corpus-­‐based	
  word	
  list	
  for	
  academic	
  purposes.	
  In	
  Bernhard	
  

Kettemann	
  &	
  Georg	
  Marko	
  (Eds.),	
  Teaching	
  and	
  learning	
  by	
  doing	
  corpus	
  analysis,	
  (pp.	
  73-­‐89).	
  
Amsterdam:	
  Rodopi.	
  

Coxhead,	
  Averil	
  (2011).	
  The	
  Academic	
  Word	
  List	
  10	
  years	
  on:	
  Research	
  and	
  teaching	
  implications.	
  TESOL	
  
Quarterly,	
  45(2),	
  355-­‐361.	
  

Crystal,	
  David	
  (2006).	
  Language	
  and	
  the	
  internet	
  (2nd	
  Ed.).	
  Cambridge:	
  Cambridge	
  University	
  Press.	
  
Davies,	
  Mark	
  (2005).	
  The	
  advantage	
  of	
  using	
  relational	
  databases	
  for	
  large	
  corpora:	
  Speed,	
  advanced	
  queries,	
  

and	
  unlimited	
  annotation.	
  International	
  Journal	
  of	
  Corpus	
  Linguistics,	
  10(3),	
  307–334.	
  
Davies,	
  Mark	
  (2009).	
  The	
  385+	
  million	
  word	
  Corpus	
  of	
  Contemporary	
  American	
  English	
  (1990–2008+):	
  

Design,	
  architecture,	
  and	
  linguistic	
  insights.	
  International	
  Journal	
  of	
  Corpus	
  Linguistics,	
  14(2),	
  159–
190.	
  

Davies,	
  Mark	
  (2011a).	
  Online.	
  The	
  Corpus	
  of	
  Historical	
  American	
  English	
  and	
  Google	
  Books/	
  culturomics.	
  
Available	
  at:	
  http://corpus.byu.edu/coha/compare-­‐googleBooks.asp	
  (accessed	
  August	
  2011).	
  

Davies,	
  Mark	
  (2011b).	
  Online.	
  N-­‐grams	
  and	
  word	
  frequency	
  data	
  from	
  the	
  Corpus	
  of	
  Historical	
  American	
  
English	
  (COHA).	
  Available	
  at:	
  http://www.ngrams.info	
  (accessed	
  December	
  2011).	
  	
  

Davies,	
  Mark	
  (2011c).	
  Online:	
  Google	
  Books	
  corpus.	
  Available	
  at:	
  http://googlebooks.byu.edu/	
  (accessed	
  
December	
  2011).	
  

Friginal,	
  Eric	
  (2009).	
  The	
  language	
  of	
  outsourced	
  call	
  centers.	
  Philadelphia:	
  John	
  Benjamins	
  Publishing.	
  
Gao,	
  Jianfeng,	
  Nguyen,	
  Patrick,	
  Li,	
  Xiaolong,	
  Thrasher,	
  Chris,	
  Li,	
  Mu	
  &	
  Wang,	
  Kuansan	
  (2010).	
  A	
  comparative	
  

study	
  of	
  Bing	
  web	
  n-­‐gram	
  language	
  models	
  for	
  web	
  search	
  and	
  natural	
  language	
  processing.	
  ACM	
  
SIGIR	
  Forum,	
  44(2),	
  59-­‐64.	
  

Grieve,	
  Jack,	
  Biber,	
  Douglas,	
  Friginal,	
  Eric	
  &	
  Nekrasova,	
  Tatiana	
  (2010).	
  Variation	
  among	
  blogs:	
  A	
  multi-­‐
dimensional	
  analysis.	
  In	
  Alexander	
  Mehler,	
  Serge	
  Sharoff	
  &	
  Marina	
  Santini	
  (Eds.),	
  Genres	
  on	
  the	
  web:	
  
Corpus	
  studies	
  and	
  computational	
  models,	
  (pp.	
  45-­‐71).	
  New	
  York:	
  Springer-­‐Verlag.	
  	
  

Herring,	
  Susan	
  &	
  Paolillo,	
  John	
  (2006).	
  Gender	
  and	
  genre	
  variation	
  in	
  weblogs.	
  Journal	
  of	
  Sociolinguistics,	
  
10(4),	
  439-­‐459.	
  

Hilpert,	
  Martin	
  (2011,	
  April).	
  Visualizing	
  language	
  change	
  with	
  diachronic	
  corpus	
  data:	
  Introducing	
  the	
  
flipbook	
  technique.	
  Paper	
  presented	
  at	
  the	
  Exploring	
  the	
  Boundaries	
  and	
  Applications	
  of	
  Corpus	
  
Linguistics	
  Conference,	
  Tuscaloosa,	
  AL.	
  

Hotz,	
  Robert	
  Lee	
  (2010,	
  December	
  17).	
  Wall	
  Street	
  Journal	
  online:	
  Word-­‐Wide	
  Web	
  launches	
  new	
  Google	
  
database	
  puts	
  centuries	
  of	
  cultural	
  trends	
  in	
  reach	
  of	
  linguists.	
  Available	
  at:	
  
http://online.wsj.com/article/SB10001424052748704073804576023741849922006.html?mod=W
SJ_article_related	
  (accessed	
  March	
  2011)	
  

Keuleers,	
  Emmanuel,	
  Brysbaert,	
  Mark	
  &	
  New,	
  Boris	
  (2011).	
  An	
  evaluation	
  of	
  the	
  Google	
  Books	
  ngrams	
  for	
  
psycholinguistic	
  research.	
  In	
  Kay-­‐Michael	
  Würzner	
  &	
  Edmund	
  Pohl	
  (Eds.),	
  Lexical	
  resources	
  in	
  
psycholinguistic	
  research	
  volume	
  3,	
  (pp.	
  23-­‐27).Potsdam,	
  Germany:	
  Universitätsverlag	
  Potsdam.	
  	
  

McEnery,	
  Tony,	
  Xiao,	
  Richard	
  &	
  Tono,	
  Yukio	
  (2006).	
  Corpus-­‐based	
  language	
  studies:	
  An	
  advanced	
  resource	
  
book.	
  New	
  York:	
  Routledge.	
  

Michel,	
  John-­‐Baptiste,	
  Shen,	
  Yuan	
  Kui,	
  Aiden,	
  Aviva,	
  Veres,	
  Adrian	
  &	
  Gray,	
  Matthew	
  (2011).	
  Quantitative	
  
analysis	
  of	
  culture	
  using	
  millions	
  of	
  digitized	
  books.	
  Science,	
  331,	
  176-­‐182.	
  	
  

Miller,	
  Donald	
  (2011).	
  Using	
  internal	
  criteria	
  to	
  assess	
  corpus	
  representativeness	
  for	
  lexical	
  studies.	
  Paper	
  
presented	
  at	
  the	
  2011	
  American	
  Association	
  for	
  Corpus	
  Linguistics	
  Conference,	
  Atlanta,	
  GA.	
  

Ming-­‐Tzu,	
  Kim	
  &	
  Nation,	
  Paul	
  (2004).	
  Word	
  meaning	
  in	
  academic	
  English:	
  Homography	
  in	
  the	
  Academic	
  
Word	
  List.	
  Applied	
  Linguistics,	
  25	
  (3),	
  291-­‐314.	
  

Nation,	
  Paul	
  &	
  Waring,	
  Robert	
  (1997).	
  ‘Vocabulary	
  size,	
  text	
  coverage,	
  and	
  word	
  lists’.	
  In	
  	
  
Norbert	
  Schmitt	
  &	
  Michael	
  McCarthy	
  (Eds.),	
  Vocabulary:	
  Description,	
  acquisition	
  and	
  pedagogy,	
  p.	
  6-­‐19.	
  

Cambridge:	
  Cambridge	
  University	
  Press.	
  
Nelson,	
  Gerard	
  (1996).	
  ‘The	
  design	
  of	
  the	
  corpus’.	
  In	
  Sidney	
  Greenbaum	
  (Ed.),	
  Comparing	
  English	
  worldwide:	
  

The	
  International	
  Corpus	
  of	
  English,	
  p.	
  27-­‐35.	
  Oxford:	
  Clarendon	
  Press.	
  
	
  
	
  



FRIGINAL,	
  WALKER,	
  &	
  RANDALL	
  
	
  
	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
   65	
  

Preiss,	
  Judita;	
  Coonce,	
  Andrew	
  &	
  Baker,	
  Brittany	
  (2009).	
  HMMs,	
  GRs,	
  and	
  n-­‐grams	
  as	
  lexical	
  substitution	
  
techniques	
  –	
  Are	
  they	
  portable	
  to	
  other	
  languages?	
  Paper	
  presented	
  at	
  the	
  International	
  Workshop:	
  
Natural	
  Language	
  Processing	
  Methods	
  and	
  Corpora	
  in	
  Translation,	
  Lexicography,	
  and	
  Language	
  
Learning,	
  Borovets,	
  Bulgaria.	
  

Reppen,	
  Randi	
  (2010).	
  Using	
  corpora	
  in	
  the	
  language	
  classroom.	
  New	
  York:	
  Cambridge	
  University	
  Press.	
  
Sekine,	
  Satoshi	
  &	
  Dalwani,	
  Kapil	
  (2010).	
  Ngram	
  search	
  engine	
  with	
  patterns	
  combining	
  token,	
  POS,	
  chunk	
  and	
  

NE	
  information.	
  LREC,	
  2010,	
  1-­‐5.	
  
Svartvik,	
  John	
  (	
  2007).	
  ‘Corpus	
  linguistics	
  25+	
  years	
  on’.	
  In	
  Roberta	
  Facchinetti	
  (Ed.),	
  Corpus	
  linguistics	
  25	
  

years	
  on,	
  p.11-­‐26.	
  Amsterdam:	
  Rodopi.	
  
Toor,	
  Amar	
  (2010).	
  Google	
  Ngram	
  Viewer	
  gives	
  new	
  historical	
  perspective	
  on	
  culture,	
  language.	
  Available	
  at:	
  

http://www.switched.com/2010/12/20/google-­‐books-­‐ngram-­‐viewer/	
  (accessed	
  March	
  2011).	
  
 



EXPLORING	
  MEGA-­‐CORPORA	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
  66	
  

Appendix	
  

Complete	
  List	
  of	
  60	
  Words	
  Used	
  in	
  the	
  Study	
  
	
  

Analysis Indicate 

Approach Individual 

Area Interpretation 

Assessment Involved 

Assume Issues 

Authority Labour 

Available Legal 

Benefit Legislation 

Concept Major 

Consistent Method 

Constitutional Occur 

Context Percent 

Contract Period 

Create Policy 

Data Principle 

Definition Procedure 

Derived Process 

Distribution Required 

Economic Research 

Environment Response 

Established Role 

Estimate Section 

Evidence Sector 

Export Significant 

Factors Similar 

Financial Source 

Formula Specific 

Function Structure 

Identified Theory 

Income Variables 

	
  
	
  
 
 



FRIGINAL,	
  WALKER,	
  &	
  RANDALL	
  
	
  
	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
   67	
  

Eric Friginal, Georgia State University 
efriginal@gsu.edu 
EN Eric Friginal specializes in technology and language teaching, applied corpus linguistics, cross-cultural communication, 

and discipline-specific writing. His primary research program focuses on the exploration of professional, spoken 
interaction; the acquisition of fluency in ESL; and the study of language, culture, and cross-cultural communication. He 
makes use of corpus and computational tools as well as qualitative and quantitative research approaches in analyzing 
and interpreting linguistic patterning from corpora. His present work aims to contribute linguistic data that could be used 
for materials production and the development of training curricula in language proficiency and task performance of ESL 
speakers. 

ES Eric Friginal está especializado en tecnología y enseñanza de lenguas, corpus lingüísticos aplicados, comunicación 
transcultural y escritura específica por disciplinas. Su principal programa de investigación se centra en el análisis de la 
interacción oral profesional, la adquisición de fluidez en inglés como segunda lengua y el estudio de lengua, cultura y 
comunicación transcultural. Utiliza corpus y herramientas informáticas, así como enfoques de investigación cualitativos y 
cuantitativos, en el análisis e interpretación del diseño lingüístico de diferentes corpus. Su trabajo actual pretende aportar 
datos lingüísticos que puedan ser utilizados para la producción de materiales didácticos y planes de estudio finalizados 
al dominio del idioma y la realización de tareas por hablantes de inglés como segunda lengua. 
 

IT Eric Friginal si è specializzato in tecnologia e insegnamento di lingue straniere, linguistica applicata dei corpora, 
comunicazione interculturale e linguaggi settoriali nella comunicazione scritta. Il suo principale interesse di ricerca si 
concentra sull’esplorazione della lingua parlata in ambito professionale, l’acquisizione della fluidità di espressione in 
inglese L2, nonché lo studio di cultura e linguaggio nella comunicazione interculturale. Allo scopo di analizzare e 
interpretare le configurazioni linguistiche presenti nei vari corpora fa ricorso a strumenti di linguistica computazionale e 
dei corpora, basandosi su una metodologia di ricerca quantitativa e qualitativa. Alla base del suo operato vi è la volontà di 
utilizzare dati linguistici nello sviluppo di materiali e programmi di formazione mirati a una conoscenza e una competenza 
attive dell’inglese L2. 
 

 
Marsha Walker, Georgia Institute of Technology 
marsharene@gmail.com 

EN Marsha Walker teaches various courses at the Language Institute (GaTech) focusing on the areas of student academic 
writing, reading, and grammar. She specializes in second language acquisition, language and literacy, classroom 
approaches, and corpus-based studies. Her research interests include the application of SLA theories, technology, and 
corpora in language teaching. She also has developed courses specific to her interests in literature and vocabulary, and 
she continually develops classroom teaching materials and instructional tools primarily intended for international students 
in U.S. universities. 
 

ES Marsha Walker es docente en varios cursos centrados en escritura académica para estudiantes, lectura y gramática en 
el Language Institute (GaTech). Está especializada en la adquisición de segundas lenguas (ASL), lenguaje y 
alfabetización, enfoques didácticos y estudios basados en corpus. Entre sus intereses se incluyen la aplicación de las 
teorías, tecnologías y corpus de ASL en la enseñanza de lenguas. Desarrolla también materiales didácticos en relación 
con su interés en la literatura y el léxico, y materiales y herramientas de enseñanza específicas para estudiantes 
internacionales en universidades estadounidenses. 
 

IT Marsha Walker insegna corsi di scrittura accademica, lettura e grammatica presso il Language Institute (GaTech). Si è 
specializzata nell’acquisizione di L2, nelle tecniche di lingua e alfabetizzazione, nei metodi di insegnamento in classe e 
nello studio dei corpora. Tra i suoi principali interessi di ricerca va ricordata l’applicazione di metodologie, tecnologie e 
corpus per l’acquisizione di una L2 nell’ambito dell’insegnamento di un lingua. Vasta è anche la sua produzione di 
materiali didattici legati al suo interesse per la letteratura e il lessico, e di materiali e strumenti educativi rivolti a studenti 
internazionali presenti nelle università statunitensi.  
 

 
 



EXPLORING	
  MEGA-­‐CORPORA	
  

E-­‐JournALL	
  1	
  (2014),	
  48-­‐68	
  68	
  

Janet Beth Randall, New York University, Tokyo 
randall.janet@gmail.com 
EN Janet Beth Randall specializes in ESL/EFL pedagogy and cross-cultural communication. Her work also consists of 

administration, curriculum development and pedagogically-oriented programs in English for Specific Purposes. Her 
current research and teaching interests include the use of corpus tools in content-based instruction and language 
program evaluation. 

ES Janet Beth Randall está especializada en pedagogía de segundas lenguas, lenguas extranjeras y comunicación 
intercultural. Asimismo, se dedica también a la administración de programas, diseño de currículum e inglés para fines 
profesionales. Sus ámbitos de investigación y enseñanza incluyen el uso de corpus en programas de instrucción 
basados en contenidos, y la evaluación de programas de estudio.  
 

IT Janet Beth Randall si è specializzata in didattica dell’inglese L2/LS e in comunicazione interculturale. Si dedica inoltre al 
coordinamento, alla creazione e allo sviluppo di programmi di studio e corsi di inglese per scopi specifici. I suoi ambiti di 
ricerca e insegnamento includono l’uso di corpus in programmi di studio basati su contenuti e la valutazione di programmi 
di lingua.