Discovering English with
Sketch Engine
a corpus approach to studying English
James Thomas
Dear Students,
On the following pages, you will find the first practical section of a book that I am writing for you,
among others. There is a long introductory section which precedes this which will be available
soon.
Working through this book has a number of goals. You should ...
1. learn a lot of language
2. learn a lot about language
3. learn how to extract linguistic information from linguistic data
4. be able to solve many language quandaries that you encounter in your work as teachers.
Register for Sketch Engine here: http://bit.ly/muni_rego. Although some parts of this software
can be used without registration, much of what you need to do requires registration. The
information on the registration pages needs to be read as you click through it. Choose a
memorable password.
The structure of the book is a step by step guide to using Sketch Engine. There are hundreds of
questions posed which need to be investigated online as your Sketch Engine skills expand. It is
therefore best worked through cover to cover with Sketch Engine open and a notebook at hand.
Every aspect of the software allows you to ask new questions or interpret data beyond the scope
of the given question. It is therefore expected that the reader will be constantly going from the
known to the unknown at many levels, and enjoying a rich learning experience.
Some KAA students are familiar with Sketch Engine, but this book is designed to take you in
different directions. Arguably, the most important thing you will gain from this experience is
learning what language questions can be asked.
Some parts of this book have been used in various training sessions and courses, but this is its
first outing as a continuous course. Therefore, any feedback you have will be greatly
appreciated. You can fill in a quick form here: http://bit.ly/DESKE_feedback115
Without further ado, let the fun begin!
James Thomas
Words and phrases
21
1. Words and phrases
Three levels of querying
Most of the language questions and discussions in this book can be addressed in most of the corpora
available in Sketch Engine. But as we are focussing on the core features of British English, DESKE
mainly uses the British National Corpus (Tree Tagger) for its discovery tasks.
Open it by clicking on its name in the table on the Home Page6
. It may be necessary to click the All
Corpora button to see it the first time you visit the page. Once a corpus has been used, it appears in
the top table of your Homepage, which shows the ten corpora you have most recently used. Clicking
on the name of a corpus, takes you to a Query form with three levels:
1. The top level is Query types where we search for words and phrases, sometimes like an
internet search. The top of the top is Simple Query, and we will explore this first.
2. The second level lets us filter a top level query within a Context. This means that we can
search for something in the context of other words and/or parts of speech, e.g. searching
for dictionary in the context of look and a preposition. See Question 156 p.103. Search for
kith in the context of kin as we saw in Error! Reference source not found..
3. The third type allows us to specify the Text Types in which the word and/or context items
occur, e.g. in spoken language, in academic texts, in legal texts, spoken by women.
Asking questions simply
A good research question is one that you can envisage finding an answer to – never just ask the question.
Always imagine the possible answers and how you would find them out. (Wray and Bloomer
2012:1)
Simple Query field
We will click on the Query types button
to toggle between the Simple Query and
the full form from p.63 onwards. When
you type a word or phrase into the Simple Query field and click Make Concordance, it searches the
corpus and returns page(s) of the search results.
Let's ask our first question.
6 https://the.sketchengine.co.uk/auth/corpora/
Discovering English with Sketch Engine
22
Question 1 Does whose only refer to people or does it also refer to non-living
things?
Before answering any of these questions, make a note of what your intuition tells
you.
Typewhose into Simple Query,click Make Concordance, and look at thenouns and
pronouns on the left of the node to see what it refers to. Here is a sample of five
lines.
Anyone who is under the impression that whose is limited to people only, as who is,
tends to use which or that instead of whose. This is wrong.
Question 2 Do we say people who or people that?
Simply type one into Simple Query and the other into another Simple Query in another
browser tab. Click between them to compare your findings.
Given that one of them is about ten times more frequent than the other, we'll take
that as the pattern of normal usage. For further information, look down the concordance
page at what follows people that – this is revealing.
Question 3 The following three expressions refer to the same thing: pass away,
bite the dust, kick the bucket. What do their various frequencies tell
us?
Do the numbers surprise you? Why might there be such a range of results? Observe
how the verbs conjugate in each one.
What meaning do the three share? Do you know other wordings for this concept?
Do we need so many?
Per million (HPM) is a number that appears beside the raw frequency of hits at the top of the concordance
page. It is most useful when comparing searches in different corpora and subcorpora. In
ukWaC, a corpus of over 1.5 billion words of English, while holidaying occurs 44 times, which is still
0.0 per million. In enTenTen, a much larger corpus, while holidaying occurs 287 times, and this is
still 0.0 per million. Expressing the number of hits per million is referred to as NORMALISATION. See
the Online Glossary.
Words and phrases
23
Question 4 Do problems lie? If so, what is meant by this?
It often happens that we meet items of language that surprise us, and asking the
thousands of native speakers whose language has been sampled in this corpus can
not only answer the question, but provide other quite useful usage information, ser-
endipitously.
In the Simple Query field type the two base forms (lemmas): problem lie. What data
is returned? What sorts of things precede and follow this pair of words? What information
do you now have?
Here are several other quirky collocations that may be equally revealing:
shed light, burning issue, fall pregnant, lose interest
Compare their HPMs in different corpora.
Question 5 Do people respond to something with the phrase that's a moot point to
mean that they don't believe what was just said?
Before doing anything, what does your intuition tell you?
Type moot point into Simple Query and be prepared for some
very clear patterns. Is it ever used in the plural? What does its
position relative to the full stops and commas tell us? What
about the verb? Is it in the past often enough to be called a
pattern? Is it used with modal verbs? Is there any reported
speech? Does the language appear to be mostly spoken or
written?
Question 6 Can you say handsome woman in English?
You can say anything. Try the question, Do you say handsome woman in English?
Students are routinely taught that handsome woman is not 'good' English, but
when you type this collocation into the Simple Query field, be prepared to find that
it does indeed occur. At issue is not the collocation, but the semantics of handsome.
An internet image search of various adjectives that describe men as physically attractive
yield quite different pages of thumbnails, e.g. handsome man, sexy man,
gorgeousman.To find more such adjectives, click on Thesaurusin the left panel and
enter handsome.
The features of handsome, although most common in men, also appear in a certain
type of woman, namely a handsome one. We revisit this question towards the end
of this book (p.184).
Remember that data has to beinterpreted before it becomes information, thus goingbeyond simple
frequency counts is an important step.
Moot Hall, Keswick UK
Discovering English with Sketch Engine
24
We are now going to work through one example. In the process, we observe some features of Sketch
Engine, start to see how language is the source of the data from which we draw conclusions about
patterns of normal usage and how these are sometimes exploited by language users. We also consider
the relevance of these things to language learning.
Question 7 Does anyone really say, while holidaying?
Perhaps a language teacher made it up to demonstrate something, or perhaps it
appeared in translation software. If you are ever suspicious of a construction, even
in your own writing, don't hesitate to check it in a corpus.
Remember that the BNC is more than 25 years old – it is sometimes fairer to ask,
Did people say …? rather than Do people say …?
Enter the phrase into Simple Query and click the Make Concordance button.
The data is very revealing. The screenshot below shows all five concordances of
while holidaying. Don't worry if yours looks different – in the section below, View
Options (p. 39), we learn how to customise the appearance of concordance pages.
How many times does while holidaying occur in the BNC? In what sort of texts? Is it
ever at the beginning of a sentence? What precedes it? What immediately follows
it?
Our search item occurs five times in the whole BNC of 100 million words. Five of anything is very
small as the 0.0 per million tells us. This means that for every million words in the corpus, the item
occurs this often. The word while occurs 54,791 times in the BNC which is 484.9 times per million
words. This throws some more light on the significance of while holidaying.
Would you expect the alternative construction, while on holiday(s) to occur more frequently? Try it.
Bear in mind that the first recorded use of holiday as a verb is as recent as 186915
, which makes it less
entrenched in the language than most vocabulary.
It would be fair to conclude that while holidaying is possible, but not probable.
Words and phrases
25
Given that there are millions if not billions of words written and spoken in English every day, the
paltry 100 million words of the BNC really is a tiny sample. Chances are that if it occurs five times in
the BNC, it occurs thousands of times in authentic usage every day16
.
The probability that while and holidaying would occur together (contiguously) would be the same
as the probability that any two words would co-occur if:
• all words were equally frequent in the language or the corpus, and if
• their parts of speech did not influence their likelihood of co-occurrence, e.g. adjective +
noun is probable in English whereas adjective + adverb is not. In fact, we will find (p.130)
that adjective + adverb is not rare7
– it is just that there is only one salient case when the
adverb post-modifies the adjective.
GDEX in the name of the grey bar across the top of any corpus search. It is the abbreviation for Good
Dictionary Example. As various operations are performed on search results, e.g. sorting, sampling,
filtering, the grey bar expands with breadcrumbs. You can therefore return the data to previous
states by clicking on them.
Returning to our concordance, the text to the left of the NODE (the search item), usually indicates
that something happened while holidaying; the text on the right usually says where, as can be expected
in unmarked word order (see below). In one case only, it starts a sentence.
When you click on the node, a wider context box pops up as in this screenshot. Click on the arrow
head at the top to close it.
7 ske.li/bnc_adj_adv_nodeforms (UK)
Discovering English with Sketch Engine
26
Click on the blue letter(s) and number(s) to the left of a concordance line to have this metadata pop
up. If there is a lot of metadata, it will scroll vertically.
Unmarked and marked features of language
Deviations from the normal, standard, typical way of doing something are said to be marked. There
must be a situation which allows for choice; the choice may be conscious or subconscious. Wearing
red shoes with your suit instead of the black ones you usually wear is a marked choice, and is likely
to have an effect especially on those familiar with your typical attire. What constrains your choice?
In English, unmarked word order revolves around Subject Verb Object. The verb and the object form
the predicate and this us something new about the subject.
Thus he met Catharine Hanson, a handsome woman of 29 and immediately led her to
his bedchamber.
Canano introduced me to two handsome women, one of whom was his mistress.
Subject Verb Complement is also standard since the verbs in this structure are copular verbs, which
are among the most frequently used, e.g. be, appear, seem, looks like.
Nona Thorp was a handsome woman but she had withdrawn into herself, …
Marked word order sees some restructuring that does more than provide new information about
the subject.
… by far the handsomest women we have seen in France were born and reared in Amer-
ica.
I thought what a very handsome woman she was.
Vocabulary choices often display degrees of markedness: the CONNOTATIONS of a word or phrase often
motivate choosing bright over clever, when describing a student. Troponyms also express degrees
of markedness, for example, in the series,
go  walk  stroll  amble
each one is more marked than the one before. Each one contains an additional element of meaning:
ifsomeone iswalking, it doesnot mean that theyare ambling, but if they are ambling,they aresurely
walking. Analogous issues arise when we explore hypernyms (p.110).
Words and phrases
27
Question 8 Would you expect each of the troponyms to be less frequent than the
one to the left of it?
Use Simple Query to observe the occurrences of these words. As they all appear in
various parts of speech, the results will be mixed.
go walk stroll amble
Frequencies
Hits pm
What are some of the reasons for this wide range of hits?
Question 9 Would you expect that these frequencies impact on the order in which
native speaker children acquire them (FLA), or learners learn them
(SLA)?
FLA questions can be investigated using the CHILDES corpus, which is among the
Sketch Engine corpora. Corpus study of SLA is mostly conducted via learner corpora.
See the Online Glossary for information.
The normal, standard, typical way of doing or saying something is obviously more
frequent than any of the marked variations on the linguistic menu. This means that
language learners have more exposure to them and are primed for unmarked forms
first.
These verbs of motion are followed up on p.72 when we have some more skills under
our belts.
Question 10 One swallow does not a summer make. What is this curious word or-
der?
Is this a proverb or an idiom? Where does it come from? Why is the verb at the end?
Under the influence of German? Which meaning of make does this structure realise?
Is this marked word order a pattern of normal usage?
The BNC does not have enough data for this archaism. Make a
Simple Query search in LEXCMI for does not a. The indefinite
article blocks the verb from appearing in its unmarked position.
There are 123 hits, 79 of which have the structure with
nounphrases: NP does nota NP make.Useyour browser'sFind:
make to see the prevalence and position of make on the concordance
page. The pattern is clear. Right sort them to see the lexical patterns as
well. For example,
Discovering English with Sketch Engine
28
However, chutzpah alone does not a good film make.
Of course, one battle does not a war make, …
The contracted form also exists8
, but many fewer than in the full form. For example,
One election loss doesn't a failure make.
Having a girlfriend doesn't a straight man make!
See p.33 for forming queries with apostrophes.
But why not: One swallow makes not a summer? What is the function of does in our
idiom: negation or emphasis? See the paradigm under Periphrasis (p.93) for more
on the roles of do.
In Shakespeare's day, the periphrastic do had not yet settled into its modern day
usage. Romeo and Juliet was written around 1595. See what they were saying:
JULIET O think'st thou we shall ever meet again? (do you think …)
ROMEO I doubt it not; and all these woes shall serve (I don't doubt it)
For sweet discourses in our time to come.
Many expressions in English also come from the King James Bible (1611) and retain
their echo of times past.
Man doth not live by bread alone. Deuteronomy 8:2-3, Matthew 4.4, Luke 4.4.
But does our idiom date from this period?
Let us consider this earlier form: Search LEXCMI for dothnot a.The seven examples
contain contemporary references, e.g. girlband, scouts. What does this suggest?
There is only one example of doth not a in the BNC, also with contemporary refer-
ences.
A: A little smoke and a couple of pills doth not a junkie make.
B: Yeth Shakeshpeare, you're right. It dothn't. Oth courth noth Speaker
B is clearly amused by what he or she takes as Shakespearean English. But
why all the 'th's?
For comparison, search for the unmarked form, does not make a, in LEXCMI. It
seems that many of the NP make + NPs are delexical verbs, e.g. make a profit, make
a contribution.
Does our target structure, NP does not a NP make, have its own underlying meaning?
Could it mean: indicate, constitute, justify the label of…, or … ?
8. The concordance is here: ske.li/lexcmi_doesnt_a_make (UK)
Words and phrases
29
Insearchingfor answers,we havefound manyquestions.Chesarà, sarà! Guideddiscoveryis journey.
One swallow does not a summer make is an underlying principle in pattern hunting: only when
many instances flock in are we able to declare something a pattern of normal usage.
Constrained options
The corollary of Kilgarriff's statement quoted on p.1, Language is never, ever, ever random, is that
every choice we make is constrained by a continuum of factors that range from the purely linguistic
grammatical conventions? to the pragmatic considerations that are embedded in the context of the
culture.
The sentence you have just read took much longer to write than to read. It is a long sentence containing
a number of infrequent words sometimes in standard constructions, sometimes not. As a
native speaker, such elements as makeachoice,factorsrangefrom AtoB,context ofculture,placed
no demands on my writing. Structuring the sentence with three that clauses, however, certainly
took some juggling and a string of choices were made to achieve this result.
Halliday's view is that “users select from options that arise in the environment of other options, and
the power of language resides in its organization as a huge network of interrelated choices“ (Systemic
functional grammar, 2013
Michael Lewis’ The Lexical Approach (1993) divides grammar for pedagogical purposes into Patterns,
Facts and Choices17
. Patterns are the regularities within a language such as morphological
forms (plural s, past –ed), unmarked word order (SVOMPT) and the gerund following prepositions.
Facts, on the other hand, relate to individual items. For example,
• the use of the subjunctive after certain words (mandative subjunctive)
• irregular forms of about 200 mostly high frequency verbs
• the comparative and superlative forms of some high frequency adjectives
• the irregular plurals of some mostly low frequency nouns
• some verbs are most frequently used in the passive
• words have different meanings which are mostly realised by their patterns18
• cognates might have quite a different meaning or use in other languages
All of these have to be learned as language facts.
Choice presents the greatest challenge for non-native speakers because selecting the most probable
form from possible forms involves weighing up the constraints on each one. There are more options
as we move up the Hierarchy of Language, as seen is this diagram.
Discovering English with Sketch Engine
30
This could be thought of as a fuzzy area of language, but in fact, as the options
arewhittled away through a process of elimination, only fewremain. After all,
everycommunicative act hasa perlocutionary force, such as persuading, convincing,
enlightening or getting someone to do something, and each of these
has its own lexico-grammatical realisations: never ever random. Perlocutionary
force is an aspect of Austin's (1911–1960) influential theory of speech acts,
a cornerstone of the field of pragmatics. Another choice we make is referring
to Shakespeare as Shakespeare, Will, the Bard, a man, he, the playwright?
Consider what motivates someone to choose one at the time of speaking or
writing. Jean Aitchison (1989) refers to the psychological trigger that performs
this choice in native speakers. Non-native speakers, however, draw on whatever criteria they
have acquired, through general exposure and direct study, and thus the choice is made more consciously.
A corpus query shows the choices that have been made.
Other Simple Query searches
We can now work through some more questions that can be asked and answered using Simple
Query only. Note that lemmas entered into the Simple Query field search for all the inflected and
conjugated forms that we saw in the Inflection side of the Morphology Taxonomy (p. 68), whereas
word forms you enter return that word form only. For
example, searching for be welcome returns the combinations
of all eight forms of BE plus all three forms of
WELCOME, whereas is welcome, searches for all forms of
is, of which there is only one, and all forms of welcome, as can be seen in this screenshot. To see the
frequency list for BE WELCOME, open this permalink9
.
Question 11 Is open dooran adjective + noun or a verb + noun? Or something else?
The minimal use of inflection in contemporary English guarantees the high proportion
of words that undergo conversion. Here we have an example of a pair of words
that look like a typical verb + noun combination, but which turn out to be used in a
variety of POS combinations.
As we saw with be welcome, Simple Query searches for all the word forms of all
lemmas regardless of part of speech.
There are 412 hits of open door (3.6 per million) in the BNC. It is mainly an adjective
+ noun as in Corp ex. 13. The following example shows a compound adjective describing
policy. We then see it is a verb and its object, conjugated and declined respectively.
This literal use of to open a door contrasts with its figurative use in the
last example. Interesting, isn't it?
9 ske.li/bnc_be_welcome
Words and phrases
31
To the right, a partially opened door led to a connecting office.
Harding, the man who implemented BNFL's open door policy has said goodbye.
… he would prowl the ward, opening doors, locking me in a bathroom.10
Our album opened doors for us all over the world.
These variations are problematic for error tagging software: be prepared for unexpected,
unreliable results.
Question 12 I’ve come across boldly go a few times and wonder if it is more than a
collocation.
Type the phrase into Simple Query. There are not very many hits, which tells us
something. There are a few words in the contexts which give a good clue as to the
source of this phrase.
Does GO conjugate or is the word form part of a fixed phrase?
It is interesting to find that GO is very rarely preceded by an adverb that describes
the going. Open this permalink11
and look at the list; it was generated using the Frequency
tool, which we study on p. 52. For more information about boldy go, see its
very own entry in Wikipedia19
.
Question 13 I’ve heard the phrase, Just when you thought it was safe to … and I
wonder how it might be used. Is it ironic?
Let’s search for just when you and see where it leads. In fact, the raw unsorted data
doesn’t easily lead us anywhere. Click on Sort Right in the left panel and scroll down
the page. Who said language wasn’t patterned?
Question 14 I’ve also heard in my point of view. I thought it was from...
In the BNC there is one example with in and 69 with from. Let that be a lesson to
you! A single instance cannot be pattern of normal usage.
Now try an internet search for in my point of view. What do you learn?
FormingSimple Querysearches issimilar to querying Google and Yahoo, etc, but theresults are very
different because the internet is a vast, random collection of documents of many types changing
moment by moment. This is in stark contrast to a corpus designed for language study analysis. Thus
the internet is able to say what is possible in English while a corpus search tells us what is probable.
This contrast appears in classroom language too: compare Can you say this in English? with Do you
say this in English? Let that be a lesson to you too!
10 See here for other examples of this pattern: ske.li/iq
11 ske.li/bnc_ly_go (FI)
Discovering English with Sketch Engine
32
Question 15 How is the word germane used?
Resist the temptation to look this word up in dictionary, at least until we have
worked through the steps below. In any case, there are questions to be asked that a
dictionary will not answer.
Search for germane in the BNC using Simple Query. It is indeed a rare word.
It is mostly followed by to. Search now for germane toin Simple Query12
. Is toa preposition
or an infinitive marker when used with germane? Or both?
Which verb precedes germane? Always, often, sometimes?
What sorts of things are the subjects?
By this stage, you have probably inferred the meaning and worked out the pattern
that germane is used in.
How would you express this notion in other words?
People's needs for precision and concision vary. These needs certainly influence the
choices we make when writing and speaking. It can take a long time to weigh up the
suitability pros and cons of particular expression. The author, Gustav Flaubert,
made such high demands on himself, or went to such great pains to find le mot
juste, that his overall output was much smaller than that of many writers of the period.
le mot juste means the most precise and appropriate wording for a specific
concept in a particular context, for which English does not have an expression. We
borrowed it from French. The absence of a word or phrase for a concept is referred
to as a lexical gap. This cannot be sought in a corpus!
Question 16 Does curiouser occur more frequently than more curious?
Is frequency an adequate criterion for determining patterns of normal usage?
As with boldly go, scanning the context provides a good hint as to how curiouser
slipped into the language and how it is used. Another example of this is never walk
alone, which may be sage advice in some circumstances, but the phrase is as culturally
loaded as curiouser and boldly go. What can you discover about this?
The formal aspect of language we observe in curiouser relates to what we know about the formation
of the comparative and superlative in English, as we explore on p. 103.
Question 17 Is Pacific Ocean written with this capitalisation? And is it always preceded
by the?
12 ske.li/bnc_germane_to (UK)
Words and phrases
33
And what about Czech Republic, East Timor, and other such compound country
names. Are they always used with the definite article? Can you determine a pattern
here?
Apostrophes in English
Welcome to punctuation, a right royal pain in corpus work, and none more so than the apostrophe.
In English, it is used for contractions e.g. they're, who'd. and possessives Megan 's, sufferer's. Behind
the scenes, corpora store them as two separate words.
To search for words with apostrophes, it is therefore necessary to separate the two parts, e.g. pianist
's,they 'd,hangover's.To complicate matters evenfurther,thefull negative contraction is separated
from the base word, e.g. do n't, could n't, wo n't.
Full stops, exclamation marks, question marks, colons and semicolons are treated as separate tokens
which is why the number of tokens in a corpus exceeds the number of words, as seen in the list
of corpora on the homepage.
Question 18 Does the phrase who'd have thought stand alone? Or is it a part of a
sentence? If so, how is it followed up?
Enter the search into Simple Query. Try it in various corpora.
Question 19 Is it the norm to use apostrophes with decades? Search for both 1970s
and 1970’s in the Simple Query field.
It is necessary to include the space: 1970 s, 1970 's. As with wrong spelling, other
instances of incorrect usage appear in corpora because they occur in the data that
was included. But as we are primarily
interested in identifying patterns of
normal usage, 1970's occurring 98
times vs. the 3,046 occurrences of
1970s unequivocally answers the question.
Question 20 What's the difference between …
If this is a chunk, we should be able to observe plenty of examples in corpora. It may
exist in both the contracted and full forms. Corp ex. 17 is a request for information,
whereas Corp ex. 18 is a formula used in English jokes.
What is the difference between information and data?
What's the difference between a prostitute and a cockerel?
Discovering English with Sketch Engine
34
Question 21 Is why followed by both can't and cannot?
For example, do we say Why can't I come too? and Why cannot I come too?
Once again, before answering, what do you expect the answer to be?
Operators available in Simple Query
Operators such as question marks and asterisks are used in searches for various substitutions. Unfortunately,
these work differently in other search fields as we shall see.
Question mark: d?g finds lemmas that begin with d, followed by one letter followed by g, Since Simple
Query is a lemma search, the results will include inflected forms, for example, dog,dogged, dags.
Click on Node Forms in the left panel to summarise the findings. z??k allows two letters between
them, while m???y three. ???n?tion, etcetera.
Question 22 Are wh- words followed by both can't and cannot?
What pattern emerges from this observation?
Question 23 Is the spelling of words with either –ise or -ize a difference between UK
and US English?
As the LEXCMI corpus has 100 million words of US English, the question mark is
very valuable in such "bi-lingual" corpora.
Select verbs that might have this spelling variation and search for them with the
question mark in place of the alternating letter, for example,
reali?e comprimi?e organi?e televi?e bowdleri?e epitomi?e ostraci?e
Corpora provide data that can indicate such tendencies, but it is necessary to read
further afield. Interestingly, this spelling difference is not related to a US-UK divide,
rather Cambridge prefers ise while Oxford prefers ize, making both acceptable in
UK English. (More on this at this link20
). There are some words which only have one
correct spelling. For example, summarise, prize.
Do these spelling tendencies extend to other parts of speech?
For example, search for televi?ion, reali?ation.
Question 24 Do you love me
We cannot use a question mark in a question as it functions as an 'operator'. If you
want to find out how people follow up a question, enter it without the punctuation.
Words and phrases
35
We study in Question 230 how punctuation can be included in the more complex
query types called CQL.
Question 25 What is your favo?rite
This is an instructive question, especially when the results are sorted to the right.
Asterisk: d*g finds lemmas that begin with d and end with g. Any number of letters may appear in
between. For example, during, drug, deteriorating. Try also combinations such as t*n*tion and list
the node forms.
Let's see if what the BNC can tell us anything about rare words. Even though it is a small corpus by
today's standards, no publically available English corpus has better metadata, so even though it
does not contain the most recent additions to English vocabulary, for many searches, it remains the
corpus of choice.
Question 26 What is the pattern that cahoots operates in?
Type it into Simple Query. With so few lines to look at, the pattern is clear. Then see
here13
for a Frequency listing – we start making such lists on p. 52.
Quantitative data should lead to qualitative information. Intuition will play a part in answering your
question. A look at the 16 lines of cahoots may also tell you something about its meaning and usage.
Quantity also varies greatly due to the size of the corpus. The almost 2.4 million words of English as
spoken in London and recorded in the LOC corpus (2011), will provide far more focussed data for its
tenor, field and mode than the NMC (2008), a general English corpus many thousands of times
larger captured from the internet – horses for courses.
Question 27 How ecstatic is life in London?
Another question you'd never thought to ask! The word ecstasy occurs in the LEC
15 times and 438 times in NMC. Which of the word's meanings appear in the two
corpora?
The answers will also depend on the patterns in the context and co-text, and on the extent to which
the data corresponds to the choices you are weighing up. The raw frequency of patterns with rare
words differs significantly from those of common words and obviously, the bigger the corpus, the
greater the chance they will appear. The mere existence of an item does not make it significant, as
we saw on p.31 when discussing the internet as a source of language information.
Question 28 What other wayward items might appear in the BNC?
Search the corpus for the following items. To learn more about their source, click
on Text Types under Frequency in the left panel. This is putting metadata to good
13 ske.li/bnc_cahoots_freq (UK)
Discovering English with Sketch Engine
36
use and indicates the degrees to which some things are possible and probable in
language.
accomodation, seperate, asshole, old fart, cotton candy, spag bol, would of done,alot
Speaking of these oddballs which do appear in the BNC, let us now consider some that do not.
zine, blog, road rage, regift, emoticon, sexaholism, snail mail, vertically challenged, metrosexual,
merkin, sunshower, website, fess up, shirkaholic, bling.
Are there any items here that you cannot live without?
The basic reason why words do not appear in the BNC is that they are not used in the texts included
in the corpus. As obvious as that sounds, it is important to bear in mind. Some words are technical
words and have limited use, others are rarely used, others are very slangy, new to the language or
outside core English. But the most important reason is they were not in the right place at the right
time. Remember, there are no texts in the BNC after 1994. Such is the nature of a synchronic corpus.
Try searching for some of these words in larger, more recent corpora such as EnTenTen, and corpora
with a wider geographical spread, such as LEXCMI.
Question 29 Did tweet meaning something before the social medium, Twitter?
Compare the use of tweet in the BNC and the NMC, two corpora of British English
of approximately the same size. Does the meaning from the older BNC still appear
in the more recent NMC? Is Twitter's logo related to either meaning of tweet?
Question 30 What do we with those words like thingamabob?
In the BNC, type thinga* into Simple Query field to get the concordances. Click on
Sort Node in the left panel to group them. Click on the second one which is thingamabobpreceded
by athing group. Clicking on the node opens the wider context box
as we saw on p.25. This particular example is of interest because it is someone talking
about this linguistic phenomenon and lists 30 examples.
Can you tell if these words are most typically used in speech or in writing?
What parts of speech are they?
Do they refer to people as well as things?
I went to that little shop on the corner, you know the one, bought by Miss Thingambob
when poor old Mr Whatshisname went to Australia or somewhere with his asthma.
Words and phrases
37
To study items that are not in the BNC, try the much larger and more recent corpus, ukWaC, or enTenTen.
To make a Simple Query in another corpus, you can use the search field in the top right of
Sketch Engine interface as inthis screenshot. The dropdownlist of corpora containsthe most recent
ten that you have accessed.
For recently coined expressions, see New words in English at the BBC/British Council site21
.
Question 31 How do company names compare in the two corpora?
e.g. Google, Amazon, General Motors, McDonalds, Kodak, Apple
Learning language from language
This brings to an end our introduction to the Simple Query field, which we will be returning to
throughout the book. By this stage, we have used the Simple Query field to obtain language data
that can answer questions about many aspects of language.
We have looked at the meaning and use of specific words and phrases, the grammatical and lexical
company they keep, compared the frequencies of near synonyms, inferred the meaning of phrases
from multiple contexts, explored the cultural relevance of some phrases, compared the significance
of both patterns of normal usage and of rare items. This is why we need to distinguish possible from
probablelanguage. We haveinvestigated somelinguistic minutiae such as articles, apostrophes and
the morphology of adjectives.
In the process, we met a number of general linguistic concepts that are important to language users
generally, and are valuable foundation constructs in DATA-DRIVEN LEARNING. The over-riding concept
here is frequency from which flows patterns of normal usage, marked and unmarked use of language,
norms and exploitations, collocation and anything else that comes under Firth's notion of
knowing a word by the company it keeps.
Not unrelated to frequency is the issue of corpus size as well as the criteria for its design. We found
that corpora contain a wealth of sentences that illustrate our findings and that they are as alive as
the real world in which they were created. In the process, we learnt some of the basics of using
Sketch Engine and familiarised ourselves with the main parts of the interface. And in accordance
with the title of this book, we are finding the answers to language questions through guided discovery
tasks.
Most of this has been gleaned from the BNC. Remember that most of the questions can be asked of
any corpora, including those you make yourself. Sketch Engine has tools for building your own corpora
so if you would like to make a corpus now, so that you can ask the following questions of your
own data and compare them with corpora provided by Sketch Engine, turn to p. 178 and make one.
Discovering English with Sketch Engine
38
We interrupt this section to bring you some
important information about customising
the interface. We will return you to the other
features of top level queries as soon as possi-
ble.
Taking ownership of your concordance page
39
2. Taking ownership of your concordance page
In this section, we learn how to customise the concordance pages and what language information
can be derived from different presentations of the data.
Regarding the phrase, take ownership of something, it was mostly used referred to something tangible:
after a transaction something becomes yours. However, it occurs only four times in the BNC
compared with 167 hits of take possession, 14 of which are take possession of goods, 13 of which are
tagged W_ac_polit_law_edu in David Lee's Classification14
. In the more current and much larger
LEXCMI corpus, take ownership yields 1,083 hits from British English and 14 of American English.
Question 32 What do we take ownership of nowadays?
We ask this question because its answer tells us how its usage has changed since
the BNC was assembled. We find that tangible items are outnumbered by intangible
items in this list of collocates that occur to the right.15
It becomes clear that taking ownership of intangible abstractions such as responsibility,
issue, development and learning involves a sense of being in control through
mental/cognitive processes. How far is this from obtaining a thing?
It is now time we took ownership of our concordance pages and customised them to suit our needs.
To do so we click on View Options in the left panel.
View Options: Top
Click on the View Options button. It takes you to a new screen in the main panel that allows you to
change the concordance view. In the Top Section of View Options, the features are determined by
the metadata in the corpus, not by Sketch Engine. These screenshots are from the BNC – other corpora
have different attributes and the Top Section varies accordingly.
14 ske.li/bnc_take_possession_goods (UK)
15 ske.li/lexcmi_take_possession_coll_r4 (UK)
Discovering English with Sketch Engine
40
Attributes
The Attributes column allows you to change
the default display.
Selecting the other attributes shows, for example,
the nodes' POS-tags and lemmas.
Lempos is a portmanteau word: lemma +
POS (part of speech). Click on Corpus Info in
the top part of the left panel to see a list of
lempos suffixes.
Under Display Attributes, choose between displaying them for the node only ("KWIC tokens only")
or for every word in the concordance line ("For each token").
KWIC stands for Key Word in Context. This is an unfortunate term, since the search item is elsewhere
referred to as the NODE, a perfectly reasonable term. Furthermore, key words in a text are
those that carry the bulk of its meaning.
Showing tags can be used to find out why a corpus line has unexpectedly matched a query: a common
culprit is an incorrect POS-tag. In this 'broken' screenshot, the node is marked as a past participle
in all cases (VVN), despite some being adjectives (AJ0).
Structures
The Structures column allows you to change from the default display to show the beginning and end
tags for structures such as sentences, paragraphs and documents.
One of the most useful is <g> which stands for glue. With this selected, punctuation will not be separated
from letters and numbers by a space on concordance pages. However, it is still necessary to
separate them when you form a query, as when making queries with apostrophes as we saw on p.33.
References
The References column allows you to select the type(s) of information about the source texts which
appears in blue to the left of each concordance line, as in the give it me screenshot
Taking ownership of your concordance page
41
The items in the list are those included in the metadata when the corpus was created. It is usual to
select one, but more is possible, in which case you use CTRL click. It is also possible to select none
of them, also using CTRL click.
David Lee’s classification
This is only available for the BNC but is generally found to be more useful than the original system.
See the Online Glossary for more information.
Question 33 Which of these variations of this ditransitive are most frequent? Do
they display any patterns in their register distribution?
Give it me, give me it, give it to me.
Enter these in the Simple Query field one at a time. Does anything else emerge, e.g.
punctuation before and after, modal verbs.
What other ditransitive verbs might be candidates for this variation? And different
indirect objects?
Question 34 Is moot point genre specific?
We studied some aspects of this phrase in Question 5. Now with David Lee's Classification,
the metadata adds another piece to the puzzle. See this link16
.
View Options: Middle
Page Size
The Page Size box allows you to specify the page width and length for the display. KWIC Context size
allows you to specify the width of the concordance
window in number of characters.
Although dramatically increasing the Page Size may
slow down the initial retrieval of the concordance,
16 ske.li/bnc_moot_point (UK)
Discovering English with Sketch Engine
42
having many lines on one page reveals patterns as soon as it is sorted. 1,000 is good, if you have
nothing against scrolling!
Jump to …
If the number of hits exceeds the page size you set, a Jump to droplist appears at the top of the page
after it has been sorted.
This is very handy for getting
past punctuation quickly, as well as going to a specific letter, or where you have sorted by References,
to a particular reference.
Sort Good Dictionary Examples
A set of criteria for selecting illustrative sentences is available here: Choosing Illustrative Sen-
tences17
. These criteria have been automated using algorithms that provide good dictionary examples,
a.k.a. GDEX. Selecting this in View Options and specifying the number of lines instructs the
program how many "good examples" to put at the top of the first concordance page. This is of great
value to language teachers and learners, as well as for the lexicographers for whom it was originally
programmed.
If you make the number of GDEX lines the same as Page size, your first page of concordances contains
all the “good examples”.
View Options: Bottom
Sentence copying
Selecting the checkbox for one-click sentence
copying adds icons to the right of
each line. Clicking on that icon copies the
17 http://bit.ly/maelt-ill_sents
Taking ownership of your concordance page
43
whole sentence to the computer’s clipboard. You can then paste it into a word processor document,
database, web page, etc.
Allow multiple lines selection: this adds icons for copying more than one line at a time as in the
screenshot below.
Checkbox for selecting lines This works with the concordance lines on the current page only. If you
need to select lines from more, it is necessary to make the Page size longer. Once you select a line,
a new button appears in the left panel under Filter: Selected lines. Click on this to shorten your
page to the selected lines only. This is very useful if you want to make a screenshot, print out or Save
a manually selected set of lines.
Show line numbers Select this to see each line numbered, as in the screenshot here. This is very
helpful when you need to refer to a specific line as often happens in lectures and classrooms. It is
also useful when mentioning a line in an article that contains a screenshot.
Shorten long references Select this to show only the beginning of long references, as in this screenshot.
This is especially the case with corpora drawn from the web, e.g. ukWaC. The full metadata
appears when the mouse hovers over it.
XML template for one-click copying is a feature used for specific projects only.
Save and Change Options
This button takes you back to your concordance page with your changes implemented.
Special note about copying sentences
When you copy sentences, it is often best to paste them into a simple text editor such as Notepad
first. This strips away formatting such as background colour, red node word(s), thereby cleaning it
for use in other programs. It may also be necessary to remove spaces between the node and the
words either side of it, and before punctuation. In some word processors, Paste Special pastes text
without any formatting, which makes the text editor step unnecessary.
Discovering English with Sketch Engine
44
We interrupted the Query types to bring you some
important information about customising the interface.
We will return to the rest of the top level queries
after we do one more thing, Menu Items.
What's on the menu?
45
3. What's on the menu?
The following menu items are used extensively to manipulate the results of a search. They appear
in the left panel once you do a search. You can place it at the top instead: click Menu Position at the
bottom of the panel. This may be preferred when working on a narrow monitor.
Save
Click on Save to save the current data as a file on your computer. You can specify
whether the output is text or xml, how many pages, whether a heading is included,
whether the lines are numbered, whether the KWIC are aligned in the
output as well as the maximum number of lines. Where it saves depends on your
browser settings.
KWIC/Sentence
This lets you toggle between the standard KWIC concordance view (which appears
by default) and full sentence view.
If you selected Icon for one sentence copying in View Options, the full sentence
will be copied whether in KWIC or in Sentence view.
Sort
As we have already seen, sorting turns data into information that reveal patterns
of normal usage. The examples we have looked at so far have mostly returned
relatively small concordances. More frequent words and phrases yield a great
deal of data and it may be necessary to use the Sample button (p.50), make
searches more specific using the Filters (p.95 ff), or use other tools such as Collocation
(p.62) and Word Sketches (p.154).
The five buttons under Sort in the left panel perform the following automated sorts.
Left and right
The default order of lines on a concordance page after a search is Document IDs – this cannot be
expected to reveal any patterns in the language. As we saw above, the right sorting of just when you
(Question 13) was very revealing. Right Sorting sorts the words on the right of the node alphabetically,
which reveal patterns and paradigms in what follows the node, e.g. punctuation, bound prepositions,
wh- words, to-infinitive, words that form compounds with the node. Left Sorting reveals
patterns before the node, e.g. words forming compounds, prepositions, punctuation. This is the
company that words keep.
Discovering English with Sketch Engine
46
Question 35 What things are said to be unthinkable?
Sometimes we come across a word in the course of our normal reading and we
would like to know more about it. However, in reading it is likely to appear only
once. To get a feel for how it works generally or variously in the language, we need
some more examples.
Do a Simple Query search for unthinkable. Left sort the concordance page. Are
there any adverbs describing the adjective?
What verbs and verb phrases are most prominent? Patterns of normal usage
emerge.
What is therelationship betweenthe verb BE and SEEM?Whyuse the latter instead
of be?
Another significant colligation is the. How often does this occur directly before unthinkable?
This is easy to calculate if line numbering is turned on. What does this
tell us about our target word? What other adjectives are used in this way?
Now right sort the page and look down the list. What typically follows it? There are
not many nouns, which is unusual for an adjective, is it not? Here are the lines with
unthinkable + noun18
. There is a general air of doom and gloom in these 25 lines.
This is referred to as negative semantic prosody. Since unthinkable is not usually
followed by nouns, what sort of adjective is it usually? Is this compatible with the
high frequency of the verb be preceding it?
Click on Right 352 in the GDEX bar at the top of the concordance page to go back
to the full set, right sorted.
In the right sorted set, what follows unthinkable? There must be some significance
to the punctuation, given that it accounts for almost half. What grammar words follow,
and what are their roles?
Are the prepositions bound to unthinkable (bound prepositions) or do they start
prepositional phrases (free prepositions) functioning as adverbials ?
18 ske.li/bnc_unthinkable_noun (UK)
What's on the menu?
47
Question 36 Do marriages break up or down?
Search for marriage break in Simple Query to get a concordance page with both
words inflected. Click Sort => Right in the left panel and see the patterns that
emerge.
Compare related nouns such as relationship, friendship, union, affair. Does any pattern
emerge?
Another approach is to make Simple Query searches for the two phrasal verbs,
break up and break down. When they are intransitive, left sorting will help identify
the subjects.
Question 37 Does whatsoever typically come at the beginning, middle or end of a
sentence? And is it noticeably used in negative contexts?
Make a Simple Query search. Click the button to right sort the words alphabetically.
If you have your page length set at 1,000, the patterns that emerge from sorting are
easy to recognise. If you have line numbering turned on, counting numbers of each
phenomenon is simplified.
What do we learn from the observation that more than half of the occurrences of
whatsoever are followed by punctuation?
This pattern of normal usage is too significant to be ignored. In Lexical Priming
(2005), Hoey speaks of words having typical positions in sentences.
What chunks emerge from a Left Sort19
of whatsoever?
You might compare whatsoever with at all (see also Question 47).
Question 38 Which is correct when listing things in prose, first or firstly?
Search for both words and observe them in both KWIC and Sentence views. Sort
left. Since we find both first and firstly, it is difficult to arrive at a descriptive, datadriven
decision. How should we deal with such mixed messages? In fact, it is a moot
point among stylists, purists, prescriptivists, and the like.
Question 39 In the name of language observation, what can we discover about the
word hamper from a left sort?
What part of speech Is it usually?
Does the verb form prefer passive or active? If hamper is typically used in passive
voice, a Right Sort should confirm this. Does it? By what means?
19 ske.li/bnc_whatsoever_left_colls
Discovering English with Sketch Engine
48
What things are hampered and by who(m) or what?
What adverbs typically qualify it?
Does it appear to have positive or negative CONNOTATIONS?
Question 40 What words in the context of hamper the noun indicate a different
meaning from the verb form of the word?
As a noun, hamper has a totally different meaning, as food, picnic and Christmas
suggest.
When a word operates in more than one part of speech, a process known as CONVERSION, the two
words very often have basically the same meaning, e.g. abuse, rubbish, perfect. But there are also
many cases where the word has a different meaning, and this is referred to as POLYSEMY, e.g. dog,
address, count, mushroom, goldfish, hamper. To complicate matters even further, there is some
overlap between conversion and polysemy. Learners need to pay attention to such features of English
vocabulary (Schmitt 2010), and anyone studying concordances needs to bear this in mind. We
have already seen that corpus software can be confused by these features, e.g. broken on p.40.
Question 41 Why does wont return so many concordance lines?
There aretwobasic reasons forthis. Perform asearch forthis word without an apostrophe.
Sort to the left and observe some patterns of normal usage, sort to the right
and observe some more. The patterns are mostly colligations20
.
In some cases wont is a typo, but in most cases it is a word in its own right. Can you
infer its meaning? And can you describe the relationship between the grammar patterns
and its meanings and uses?
Sorting Node, References and Shuffle
Earlier, we investigated opendoor. As we know,
Simple Query items that are lemmas are conjugated
and declined. To group all the possible
forms of a search such as open door, sort the
Node. We immediately see that of the 412 hits
in the BNC, the first 314 are in their BASE FORMS.
SortReferences rearranges the lines accordingto the metadatayouselected in ViewOptions.David
Lee’s classification shows that 13 hits of open door are spoken and over half are from fiction.
20 ske.li/bnc_wont (UK)
What's on the menu?
49
Given that spoken language constitutes only c.10% of the BNC, raw frequencies are not a fair representation.
For the studies in register and stylistics that corpora provide data for, these numbers
need to be normalised. See the Online Glossary for more on normalisation.
As mentioned above, a concordance first appears in order of document IDs. We use Shuffle to arbitrarily
order the order of lines so as to avoid the bias which may emerge from only looking at the first
portion.
Simple and Complex Sorting
The automated sorts discussed so far are single level only: alphabetical. If we wanted to sort, for
example, firstly by David Lee's Classification and then alphabetically to the left, we need to use the
Multi-level sort. In fact, this was used for the moot point example mentioned above. See the result
of the sorting here21
.
When you click the Sort button in the left panel, you are presented with both a Simple Sort form as
pictured above, and a Multilevel Sort form as pictured below.
To customise your sort, click on Sort itself. The Simple Sort has a dropdown list of Attributes derived
from the metadata of the corpus.
Question 42 Are references to the Irish political situation spelled with a capital T?
Perhaps even two?
Simple Query: the troubles. Click Sort. Simple sort attribute: Word.
Sort key: node. And leave Ignore case unchecked.
Question 43 Is open typically a verb or adjective with door in fiction?
To answer this question, our First level must be the domain. Use David Lee's Classification.
Position: Node. The second level will be tag, also Node. To have the lines
sorted in alphabetical order, make the third level word, 1R. Making it 1L will show
patterns in the determiners, which was not the question, but is not uninteresting.
The page will be thus sorted. W_fict_prose starts at about Line 50. To see the tags,
it is necessary to go to View Options.
21 ske.li/bnc_moot_point (UK)
Discovering English with Sketch Engine
50
Selecting backward shows the element in reverse alphabetical order, e.g. David
Lee's Classification listing starts with W_items (written) instead of S_ items (spo-
ken).
Question 44 In which text types are The Troubles discussed?
Using the Multi-level Sort, make the first level word – node. Second level Domain
for written corpus texts – node. The capitalised node starts halfway down the page
which can be marked using the Select Lines checkbox (in View Options).
Sample
Most corpora are samples of the language they were designed to represent as it is not usually possible
to gather all the examples of a text type. An exception to this is a corpus of a specific text type,
e.g. all Shakespeare’s plays in one corpus when it was designed to study the language of Shakespeare's
plays. We are generally satisfied with observations made from a sample, as is the case in
many scientific fields.
It is not unusual for thousands of corpus lines to be returned for a search. Analysing the first lines
whether the first 100 or the first 1,000, is not advised since the default order is the document ID
numbers in the corpus. To get a random sample, use the Sample button.
TIP It makes sense to make the sample size the same as your page length.
You can specify the size of the sample (i.e. the number of lines) or use the default of 250. For example,
if you search for play (verb) in the BNC and decide that you do not want to analyse 37,632 lines,
use the sample tool to reduce it to a manageable number.
Question 45 Does be bound to refer to anything else than something like 'being tied
with rope'?
Let’s look at be bound to. Take a sample of 250. Right sort and patterns appear.
Do you see any instances of something being literally bound, e.g. with a rope?
Is there any sense of being emotionally bound to someone?
What's on the menu?
51
Perhaps most striking in the concordance is the verb: is bound to + verb, as this frequency
list reveals22
. What can observe about its subjects23
?
well sooner or later er this mixture of languages is gonna spoil the English language isn't
it? Bound to. Yes it's bound to. No what the they'll do, they'll put all the words that are
used into a computer er from all these things won't they?
Breakfast was bound to arrive early enough for them to have time to talk afterwards.
A cake plate so temptingly situated is bound to be fought over.
Compare what happens when the object of to is a noun phrase and when to is the
infinitive marker.
These examples reveal the very different jobs that words do, and how the meanings
of words are influenced by the words around them.
Question 46 I’ve heard people say that they are open to various things. Any things at
all?
When you search for be open to in Simple Query, you potentially get all eight forms
of BE and all four forms of OPEN. Let’s try it. There are too many hits for a human
to process. First, make a Sample which is the same as the page length you set in
View Options. Then in the left panel click on SORT NODE. To answer your question
about the things that people are open to, click on SORT RIGHT.
Question 47 Does at all reinforce both positive and negative things? And where in
the sentence is it typically found?
Use the same procedures as in the previous question.
Question 48 Does put typically have the structure: someone puts something some-
where?
Being such an extremely frequent verb, a sample will help you start to answer this
question on typicality.
Filter
After performing a search, Filter allows you to remove lines that contain recurring irrelevant elements.
It allows you to specify constraints on the context of your search results to retrieve a subset
of your concordance. Click on the help icon to open the page-specific help.
22 ske.li/bnc_be_bound_to_freq (UK)
23 ske.li/bnc_subj_be_bound_to (UK)
Discovering English with Sketch Engine
52
Question 49 Do we prefer turkey breast to breast of turkey?
This could be an important question for someone translating a menu into English,
but such questions would be better answered using a corpus of English menus. See
DIY corpora from p.178. It is also possible that modern culinary movements express
themselves differently from the BNC days.
Search for turkey in Simple
Query. Then click filter in
the left panel. Type breast
into the field as in the
screenshot and leave all
other settings as they are.
Try a similar search for Mother of God vs. God's mother, as seen on a label in an art
gallery.
Question 50 How often does Ireland occur in the context of the troubles?
Search again for the troubles. Then click Filter. In the Simple Query field, type Ireland
or even Ireland | Irish, to make the subset. Of course, such filtering may answer
a more sophisticated question than How many?
Question 51 How are The Troubles represented in spoken language?
Once we see enough instances of something within a search, we can constrain the
data by selecting that aspect of it in Filter.
Click on filter. Beside the Simple Query, there is the Text Types button. Click it and
select the relevant aspect, and then at the bottom Filter Concordance.
At the top of the Filter form, there is also the possibility of excluding a feature. Select
the negative radio button. Excluding 1975 to 1993 clearly indicates that the
troubles had not yet taken on its association with Ireland. One can't help wondering
if the non-Irish associations of this phrase are not a substitution for the more general
idea of the problems someone has with something.
Frequency
Multilevel Frequency Distribution is another tool that summarises data. It generates
a variety of multi-word lists, depending on the Attributes you select.
These lists reveal the horizontal, syntagmatic company that your Node keeps.
What's on the menu?
53
BUNDLES strings of word forms only
SYNTAGMS strings of POS only
HYBRID FORMS strings of POS tags and words and/or lemmas
WLTS word form + lemma + POS tag (of the same word)
After you search for a word or phrase, click on Frequency in the left panel.
Question 52 How do sentences start with a word like Given?
Enter Given into the word form field and select Match Case. Click Make Concordance
then Frequency.
To see Given plus one word, set First Level as in the screenshot above, and Second
Level select 1R, i.e. one to the right. Click Make Frequency List.
To add the next word, select 2R in the Third Level, etc. And we find that sentences
start with Given like this.
Given the nature of … Given the fact that …
Given the importance of … Given the choice, …
Click on the P on the left to see the actual concordances.
The question might also be answered
with parts of speech instead
of words. Select the
Attribute, tag, in the second level
to find that determiners are the
most frequent item in the second
slot. Tag in the third slot unsurprisingly
reveals that nouns and adjectives follow determiners.
Selecting the appropriate attribute for each column is important for answering your question. Note
that First Level refers to the column of data that appears on the left of the table generated.
Question 53 Is Given that looking back to what has already been said, or forward to
what is going to be said?
These are two quite different functions of Given that and both are represented in
these concordances.
When a word, often a pronoun, refers to something already mentioned, this is referred to as anaphora.
The item of specific reference is referred to as the antecedent and the referring item is a proform:
bold and underlined respectively in the following examples.
The petite blonde star refused to go in his car and drove her own vehicle to his central
Discovering English with Sketch Engine
54
London offices.
When the pro-form precedes the antecedent, this is referred to as cataphora.
Though she had no car of her own just now, Charlotte had been driving, and driving
well, for more than four years.
In the case of Given that, the anaphoric that is stressed when spoken and marked with a comma in
the written form, as in Corp ex. 25 and Corp ex. 26. It is unstressed when cataphoric, as in Corp ex.
27 and Corp ex. 28.
Well, it seems that taxis are considered "public transport" for the purposes of this exercise.
Given that, as a group, taxi drivers are the most reckless, aggressive, and inconsiderate
road users, they should have been the first, not the last, to be banned. (NMC)
… so the increase in water storage will need to be somewhere between 10-15 percent.
Given that, the debate looks set to intensify, not evaporate. (NMC)
Given that the drive is very long, it is essential to spend a few days on this amazing journey.
(NMC)
Given that the number of different platforms that IE is not even available for is increasing
as time passes ... (NMC)
Question 54 What patterns does unthinkable and other such adjectives work in?
Make a Simple Query search for un*able. We will now use the Frequency tool to investigate
its colligation patterns.
It is not necessary to look at the > 22k concordance lines. Click Frequency in the left
panel and make a First level list only: node – tags.
This list shows that the to-infinitive marker, to,is the most frequent POS after these
adjectives. Click on the P beside TO to see the concordance page and then click on
Node forms to see which words are followed by to.
The second and third most frequent POS after these adjectives is NN and SENT.
One of these suggests predicative adjectives and the other attributive. Which is
which and why?
The fourth tag on the list, IN – preposition, does
not include to as the infinitive marker. Click on
the P beside IN. Make a two-level frequency list
of the adjectives followed by prepositions, as
shown in the screenshot. Set the Frequency
limit to 2 or 3. Then sort by clicking on the columns
headings – the left column to put the adjectives
in alphabetical order, the right column heading to put the prepositions in
order.
What's on the menu?
55
We have made some observations about a set of adjectives formed with the affixes,
un- and –able, concerning some of their colligation patterns. Perhaps the same patterns
apply to all adjectives, other sets of adjectives. There's only one way to find
out. Actually, that last sentences is a fixed phrase in English (with over 1,000 hits in
enTenTen) – there is always more than one way to find something out.
Question 55 How does the collocation problem lie work?
We met this collocation in Question 4. Using frequency, we can now show what patterns
follow it. It is quite instructive.24
Question 56 How does arresting officer come to have more than one meaning?
It is worth observing the POS in compounds, as we saw with open door. The variety
of combinations often needs to be borne in mind when interpreting concordances.
We can generate a word family, lemma, tag (WLT) table by setting all of the levels
at Node but with different Attributes.
Here is the result of arresting as found through a Simple Query. It is interesting to
see it almost as often an adjective (JJ) as an -ing verb (VVB)25
.
Click on the P to see the contexts. Here are six of the 104 adjective (JJ) sentences.
The polysemy of this adjective is clear from these examples. These were Selected
using the check boxes, and Saved as a text file, and then imported into this
Word document, complete with the angle brackets and spaces as shown.
This area is the Coulin Forest, … its mountains forming a compact
group but individually having distinction of character and outline,
some being of <arresting > appearance.
Mark usually achieved this by thinking out an <arresting > beginning, nearly always of
the same type, asking his congregation to imagine themselves standing gazing at the Pyramids
or the Acropolis …
Ways of <arresting > decay also included covering eyes and mouth with carefully shaped
24 ske.li/bnc_problem_lie_bundles (FI)
25 ske.li/bnc_arresting_wlt
Discovering English with Sketch Engine
56
pieces of jade and inserting jade plugs into other orifices of the body.
Mrs Spurling added: But this <arresting > image takes no account of the fact that
Freya's was essentially an imaginative achievement. …
She swung around and tapped the <arresting > officer on the breastplate.
Theword iscloselyrelated tothe French,arrêter, meaningtostop. As apositive adjective, it suggests
that something was impressive enough to make the viewer stop in their tracks – at least metaphorically.
What else does it suggest in these examples?
To see what immediately follows the Node, choose 1R under second level. 1R generates a list of
items that occur one to the right of the node, e.g. words, lemmas, tags. 5L, generates a list of items
that occur five to the left of the node. Once you have determined the parameters for your columns,
click Make Frequency List.
Question 57 Would you consider well within a collocation?
If well within is not in your active vocabulary, the little piece of research we undertake
here should raise your awareness of it.
A Simple Query search returns a fairly small number of hits. Sorting to the left and
again to the right gives some idea of how this pair of words works. Does it match
your criteria for a collocation?
Using Frequency, however, reveals bundles that indicate that it is a semi-fixed
phrase: the elements are finite but they mix and match in a variety of ways. Perhaps
you can diagram its template: what types of elements precede and follow it? See
Template in the Online Glossary.
Zipf's law
As frequency graphs often show, relatively few items make up the bulk
of the whole. This statistical phenomenon is known as Zipf's Law. In essence
it states that the frequency of any word is inversely proportional to
its rank in a frequency table (Johns). This thumbnail screenshot demonstrates
this tendency, as do many other such graphs in this book, regardless
of the feature of language being investigated. See the Online
Glossary for more information on this. This particular screenshot is the
top of a list of all lemmas in the TED corpus26
.
26 ske.li/ted_top_lemmas_freq (UK)
What's on the menu?
57
Question 58 Is the phrase a long way complete? Does it occur at the end of an information
unit?
Search for the phrase, a long way and click on Frequency. Select the columns appropriate
to your investigation. This screenshot shows us that the phrase, a long
way, is most typically followed by (a) one of three prepositions, (b) a full stop indicating
that it is common at the end of an information unit – sentence or clause, and
(c) it is also followed by away, which is not insignificant.
We further find that a long way away is mostly followed by a full stop, comma or and, confirming its
final-position status27
. Click on the P beside a long way to go. And then Frequency again. This time,
First level – 2L, Second level – 1L, Third level – node. The word still is particularly salient before this
phrase.
Question 59 Is it true that practice is the noun and practise the verb?
Frequency is an invaluable tool
for observing Firth's company a
wordkeeps. Search the BNC for
practi?e and then ask Frequency
to create a list of Node forms using the Attribute, lempos. It should look like this, which
means that it is true. But what about mysterious 315? More tagging errors or is there another expla-
nation?
Across the pond, the answer to the practice vs. practise question is different.
Use the 'am' English part of LEXCMI to discover what happens in
the US.
Question 60 According to everyone, according to me is
wrong. Is it? According to whom?
27 ske.li/bnc_a_long_way_freq
Discovering English with Sketch Engine
58
Who decides what is right and wrong? Where do rules come from? Cambridge Dictionaries
Online has a usage note about according to me here22
, for example. Where
do they get their authorityfrom? What is the difference betweentheir authority and
yours?
Perform a Simple Query for according to, then in Frequency, select R1 to find out
what follows it. Does this answer the question? Or another question?
See also p. 113 for more on according to.
Question 61 Is thereby always following by –ing forms?
After performing a Simple Query on thereby, click Frequency and make the first
level the node and the second the tag of all the items that immediately follow it, as
shown in the screenshot.
The result makes it quite
clear that verbs are the
most frequent elements after
thereby, and the most frequent of these is the –ing form. To learn more about
the patterns of normal usage of thereby, use the Frequency listings putting thereby
in different columns (levels) and observing what goes before and after it, thereby
listing the bundles in which it occurs.
Node tags
Clicking on the Node Tags button in the left panel shows a table of the POS tags for the search items.
Question 62 I thought lend was the verb and loan was the noun.
But I hear loan as a verb more and more often these days. After searching for loan
in Simple Query, click Node Tags. Try this in both the BNC (1994) and ukWaC (2008)
to see if a change of use is in evidence.
Question 63 What is the difference between be interested and be interesting?
In Simple Query, type be interest.*. Click Node tags to see the combinations of POS
tags that result. Click on the P beside the most frequent ones to see their concord-
ances.
Question 64 Is friend a verb?
In the BNC, try a Simple Query on friend then click Node Tags or Show Tags in View
Options. Open a new tab in your browser and repeat these steps using more contemporary
corpora. The following examples all come from enTenTen.
Come for food, drinks and to meet all the people you’ve tweeted, friended or (gasp!)
You
couldn't
be
wronger
What's on the menu?
59
emailed who are in town for the conference.
… you want to be actively friending and directly connecting with 3-5 people a week that
have influence in your industry.
The following examples are not verbs, but are derived from the new usage of to friend. Corp ex. 37
goes one step further using the negative prefix as well.
The Host Community on LiveJournal was hosting a friending meme.
Perhaps this will help us develop a theory of the entire cycle of friending and unfriending.
The next example seems to say it all, really!
Researchers flew under Facebook’s radar, automated friending random strangers and
THEN friended friends of those that friended the socialbots.
Question 65 Is will a lexical verb in addition to being a future marker?
Search for willinSimple Query.Then click Node tags: MDindicates its modalstatus,
while tags starting with V indicatelexical verbs. Having found that it is a lexical verb,
how does it work? Is there one dominant V tag? Why? What are its typical subjects,
objects, complements and its typical grammar patterns?
Is this worth adding to your active/productive vocabulary?
Question 66 Are the typical subjects of couldn't bear to nouns or pronouns? And are
there any typical verbs following to?
Remember to separate: could n't. Here is one of many examples of this chunk in the
BNC.
Jessamy prized her privacy far too much, she couldn't bear to have a pack of journalists
prying into every corner of her life.
Question 67 Who, apart from journalists, pry?
Search for this word in the BNC and use a simple webpage search (CTRL F) to find
journalist on the page. A left sort will indicate other subjects of this verb, and much
more, as certain phrases appear.
Question 68 Is pry only followed by into? Can it be followed by an object instead?
Could it be a phrasal verb? And/or a transitive verb? As it is such an infrequent verb
in the BNC, a simple right sort will answer these questions.
In the process of answering these 'prying' questions, the connotations of the word
appear from its context. David Lee's Classification (View Options) reveals the one
text type in which this word is most significant.
Discovering English with Sketch Engine
60
Question 69 Pack of? When pack is not a short form of packet, what are the most
common things that come in packs?
Search for pack of in Simple Query and use Frequency to investigate the 1R column.
Do you think the author of the 'Jessamy sentence' associates journalists with lies,
wolves, etc.?
Question 70 Why might the author have chosen prying into instead of look into, investigate,
put under the microscope, etc?
This is a typical choice we make when we speak and write. Using Simple Query and
Frequency as described above demonstrates the negative connotations of to pry28
.
The cumulative effect of couldn't bear, pack and pry is an example of so-called LEXICAL SUPPORT.
When we are speaking and writing, one of the criteria we apply to selecting a word from among its
synonyms is its connotations: words with similar connotations lend support to each other to create
the desired rhetorical effect. See, for example, Corp ex. 25.
The difference between connotation and lexical support is that the former is a property of a word:
the consistent understanding of a word's connotations across a language community can be shown
through word association tasks29
. Speakers and writers achieve their rhetorical effect by co-selecting
words with similar connotations, and this is lexical support.
Serendipity
Finding something valuable unintentionally. It is not unusual to find something of great interest in
a dictionary when you are looking for something else. In Question 66 we were looking at the objects
of couldn't bear and within four or five questions have ended up discovering some rather unrelated
but valuable features of English and language in general. These serendipitous findings are as common
in corpus work as they are in dictionary searches, and other areas of life. The index of this book
lists several other references to serendipity.
Node forms
The next questions deal with prefixes and suffixes
Question 71 Which words employ these meaning-laden suffixes?
–cide, -phobia, -phyte, -dactyl? –mania?, -proof, -gate, -esque
28 ske.li/lexcmi_pryinto_freq4r
29 Paul Baker on the new use of gay
What's on the menu?
61
After a search involving operators (p.34), click on Node Forms to produce the list of
word forms.
It is worth observing:
a. what proportion of the word forms occur once only. See nonce words (p.85).
b. if they look like someone is playing with, or exploiting, language.
Question 72 Are words with the femine suffix, -ess, fading from use?
The word actress seems to be fading from use as most male and female thespians
refer to themselves as actors. What other nouns end with the feminine marker, ess?
Can we conclude that using it is a marked choice these days? A comparison of
old and new corpora may be instructive.
Is the use of the –ess ending a marked choice that people still make?
Question 73 Which words that end in –sis form their plural with –ses?
Once you have built a list of these, can you observe the text types that they are most
typical of?
Doc IDs
Staying with the 58K hits of –sis (BNC), click on Doc IDs. A list of codes for each text in the corpus
that contains words ending in –sis is generated. Given the Greek origin of words ending in –sis and
given the fact that Graeco-Latin words figure significantly in academic prose, it is not surprising to
find that the texts with the most uses of such words are academic in nature.
Click on P beside one of them and then click on the blue reference at the left of the concordance.
The yellow metadata detail frame pops up.
Text Types
Click on the Text Types button to see a summary of the
metadata for a particular search. For example, the academic
nature of the –sis words is confirmed: small
proportions of spoken language and written texts are
tagged as Arts domain, Imaginative domain. At the
bottom of Text Types, you can specify David Lee's
Discovering English with Sketch Engine
62
Classification categories which more specifically reveal the domains and sub-domains. To select
more than one of David Lee's Classifications, use the vertical bar as in this screenshot.
See p.112 for more information about Text Types.
Question 74 Blench seems like a perfectly normal English word, but one does not
come across it very often.
In what sort of texts does it appear?
Question 75 Harmony is mainly a technical term in music. How do we use it else-
where?
Find harmony via Simple Query. Click Text Types and see that according to David
Lee's Classification it is mostly in the humanities. Clicking on P (positive) shows that
it is mostly used in music contexts. To answer your question, click on N (negative)
beside W_humanities_arts. This will show the examples of harmony that are not in
this domain. There are still references to music, but it also answers the 'elsewhere'
question.
Collocation
The next button in the Left Panel is collocation. This topic is dealt with at considerable length starting
on p.143.
Visualize
The visualisation graph divides a whole corpus into 100 chunks of equal size. These chunks are not
related to specific documents, domains, genres etc. Rather, it is necessary to think of the corpus not
as a collection of documents
but as c.100 million words that
have been drawn from the documents
of the corpus.
The word eagle occurs 1,000
times (according to Word List30
)
and the graph of its distribution,
30 ske.li/bnc_lemma_1000
What's on the menu?
63
as seen in this screenshot, indicates that it occurs significantly at about 42% and very significantly
at 100%. Its other occurrences across the corpus are trivial.
By contrast, the second graph depicts the word shared, which also occurs 1,000 times – it is evenly
distributed across the whole corpus.
Visualise reveals that some
parts of a corpus are quite specific.
It is usually caused by the
fact that in these parts there
are some specific documents in
which specific words are used.
The Visualize tool appears several
more times in the book,
demonstrating further applications of this data representation.
Question 76 Compare the distributions of blench and other word forms that also occur
exactly 21 times in the BNC.
chromaticism cannibalistic lowbrow symbolist tone-units self-love ill-educated
cash-strapped rail-coaches shoebox tweet subliminally polysyllabic edification
Discovering English with Sketch Engine
64
Top Level Queries continued
65
4. Top Level Queries continued
We interrupted First Level Queries (from p. 21) to provide some important information about the
Left Panel buttons. Via the Simple Query searches we learnt about customising our View Options
(p.39) and then we proceeded with the Menu Items (p.45).
TIP It is not necessary to click any of the radio buttons beside Query type – it’s enough to click in
your field of choice.
We now return you to the rest of the top level queries, as shown in this screenshot. For this, we need
to have a grasp of the lower branches of the Hierarchy of Language which come under the heading
of morphology.
Morphology
The taxonomy of morphology accounts for the processes that words undergo grammatically and
lexically as can be see in the left and right sections of the morphology taxonomy respectively. Its
metalanguage is used throughout this book as many types of searches and post-processing involve
these concepts.
Conjugation is the name of the process
that verbs undergo when they change
their forms to indicate person, number,
tense, e.g. compute, computes,
computed, computing.
Declension is the name of the process
that nouns, adjectives, adverbs and determiners
undergo. Nouns change
their forms for number (computer –
Discovering English with Sketch Engine
66
computers; corpus – corpora), adjectives and adverbs for degree (comparative, superlative) and determiners
such as this those his their for various combinations of number, person.
Derivation is the name of the process that involves adding affixes (prefixes and suffixes) to words. In
comparisonwith theabove twoinflectionalprocesses, derivationcreatesdifferent meaningsand/or
changes the part of speech which gives it a different syntactic role.
Compounding sees the combination of two or more words to make a new meaning. Sometimes they
are written as one word, they are hyphenated or they remain as two words. Nevertheless, mobile
phone, ballet dancer and boarder guard, for example, each expresses a single concept.
There is another word formation process which involves the derivation of one lexeme from another
(e.g. the verb father from the noun father) without any overt morphological change. This phenomenon
is labelled variously: conversion, zero-derivation and zero-affixation. (See Carstairs-McCarthy
2002).
A set of words with the same stem or root that are the product of both inflectional and word formation
processes is known as a Word Family.
Part of Speech Tags
A part of speech tag is a code that indicates a word's part of speech in the context in which it appears.
The word like, for example, is tagged as a preposition (IN), a lexical verb (VV), an adjective (JJ) and a
plural noun (NN) in the BNC, which the following four extracts show respectively.
Europeans have turned whales into a sacred animal, like the Hindu cow.
Where appropriate, we like to take an active role.
Methinks it is like a weasel.
You give me anyone's phone line and mail for a month and I'll tell you exactly who he is,
how old he is, his likes and dislikes, his character, his worries, what he had for breakfast
and supper, what time he wakes up.
The metadata in this corpus tells us that this last sentence is from Esquire, 1992, long
before Edward 31
Snowden's NSA revelations. It also seems to predate the preference
for non-sexist language, as anyone here is assumed to be he. Nowadays we are more
likely to use they.
Notice that the four like tags above, indicate more that POS. It is just the first
element of the tag that identifies the POS. The following elements provide 'sub information'. For
example, VV indicates a lexical verb, VM a modal verb and VVZ is the third person singular (Z) of a
lexical verb.
31 https://openclipart.org/
Top Level Queries continued
67
Given that like occurs 144,481 times in this corpus, it is helpful to have some criteria to filter our
searches. It is not usually necessary to show the tags, but Sketch Engine can be set to display the
part of speech tags of every word if required. For example, when 'show all tags' is on (see p.40), the
format is:
You might like to imagine the scenario in which this sentence occurred.
Part of speech (POS) tags are used by humans when we create searches for words functioning in a
particular part of speech, and for identifying patterns in the grammatical company that words keep.
As we will see time and again, they are used by concordancing software to group language phenomena
in many ways.
Tagging is almost always done computationally using man-made algorithms: it is therefore unreasonable
to expect 100% accuracy. Language is a very complicated business and not even linguists
always agree on the part of speech of words in context. There are many tagging errors. For example,
in the BNC, broken is never tagged as an adjective despite 150 occurrences of broken glass, 72 of
broken heart and 26 of broken promises. See a 'broken. screenshot on p.40. It is not considered a
lemma which means that it does not have a word sketch. Another example appeared when studying
cause as a verb followed by a to infinitive. Most of the causes are in the phrase, have cause to, where
it is a noun32
.
Later we shall have cause to doubt its accuracy and validity.
Well, I wish he did have cause to be jealous of me.
The 18 examples of likest are tagged as superlative adjectives in the LEXCMI corpus. They are, in
fact, all second person singular forms from an earlier period before English verbs shed most of their
inflections.
Out of my doors, knave, if thou likest it not.
This could be a useful sentence to learn as a chunk – it's far more elegant than today's, If you don't
like it, piss off, prat. Corp ex. 46 is from Thomas Lord Cromwell, a play bearing Shakespeare's name
but generally considered apocryphal.
The BNC was originally tagged with a tagset called CLAWS and was retagged by the University of
Lancaster using an extended CLAWS tagset. Information about the so-called BNC2 is available on
their website33
, which describes related activities such as other tagsets, tokenisation, ambiguity.
Everything except when BNC2 was released!
32 http://ske.li/bnc_cause_to_vb.
33 http://ucrel.lancs.ac.uk/bnc2/bnc2postag_manual.htm
Discovering English with Sketch Engine
68
However, most corpora use another tagset called the Penn Treebank Tagset, using software called
the Tree Tagger (TT). Sketch Engine team recently retagged the BNC with PTB, which is the only
version used in this book. A comparison of these two tagsets is available on the Versatile site.
Another layer of metadata that Sketch Engine adds when compiling a corpus is the lemma of each
word. The most familiar uses of lemmas are headwords in dictionaries. Lemma can be seen in
(Error! Reference source not found.) which we will take a quick look now at in preparation for the
notes on lemmatizing.
Lemma
(Stubbs, 2002) also stresses that “word is ambiguous and
we need to distinguish between word-forms and lemmas,
because we need to differentiate between units of texts and
units of the vocabulary of a language”. (p.25)
The LEMMA GO consists of the five WORD FORMS, go, goes,
going, went, gone. This is another stratum of data that is
available in Sketch Engine corpora, and is therefore an
important choice when creating a query: are you searching
for all forms of GO, or just the WORD FORM go?
Nouns also have lemmas and word forms, though declension
is not a particularly notable feature of English
– it mostly distinguishes singular and plural forms, and possessives. Nouns in highly inflected languages
have many word forms that express number and case. Czech, for example, has seven cases,
two numbers singular and plural plus some historical relics, such as pluralia tantum and dual. This
gives Czech nouns a potential for fourteen word forms.
English also has some historical relics in words deriving from Greek and Latin. For example, behind
the word form criteria (3,798), we find its base form (lemma) CRITERION; ditto bacteria, media, millennia,
hypotheses, analyses, algae.
Perhaps more interesting is the noun GO and its plural goes, which both appear in our first corpus
example.
Choddo had a go at a few and scored wins after just two goes.
When you search for a lemma, you get all of its word forms. Error! Reference source not found.
shows the word forms of GO, and their frequencies in the BNC resulting from a lemma search not
limited to the verb.
Such an observation might pique the linguist's curiosity as to whether the base form of verbs is always
significantly more frequent than the other word forms. On finding that it is, the data needs to
be interpreted to account for why this might be the case.
Top Level Queries continued
69
Thus when a corpus is compiled, one of the main machine processing tasks is tagging and lemmatizing,
which involves putting every word in every text on a separate line accompanied by its lemma
and POS tag . In this book, this is thus referred to as WLT.
The lemma field is case-sensitive. For example, grant (n) (BNC:
6,121 hits) vs. Grant(BNC 2,412 hits); centre (BNC 22,250) vs Centre
(BNC 6,930). But capitalisation does not entirely guarantee a
proper noun due to sentence capitalisation. Proper nouns can be
specified using tags in CQL, as we see on p.124.
Question 77 Is holiday used as a verb more today than
in the BNC days?
In the lemma field, type holiday and select verb from the POS droplist. Try this in
both BNC and ukWaC. Because of the very different sizes of these two corpora,
compare your findings using per million. What do you find?
Furthermore, Node forms reveals a preference for one form.
Visualize reveals its quite equal distribution across the corpus. Do any of these observations
shed further light on Question 7?
Question 78 Compare thatcher and Thatcher in the BNC.
Given the period in English history that the BNC covers, it is not surprising that the
Iron Lady is well-represented.
Inferring meaning of unknown words in a text is much discussed in language teaching literature.
Language learners are often tasked with inferring the meaning of a word which occurs once in a long
text. Compare this with what we have been inferring from multiple examples in short extracts from
thousands of native speakers.
Question 79 If you are not familiar with the common noun, thatcher, can you infer
its meaning from these contexts?
Can you deduce the verb from the noun thatcher? Search for thatch* and right sort
and you'll be well on the way to answering the question.
We prefer Simple Query to
Lemma and Word Form when
we don’t have much knowledge
about a word. It is often a good
starting point, like a general internet
search.
Discovering English with Sketch Engine
70
Question 80 What can we learn about the word slouch?
If you do not know this word, resist the temptation to
look it up before doing the following. It could be an interesting
voyage of discovery inferring its meanings and
uses.
Do a lemma search without specifying part of speech.
Then …
1. Click on Node Forms and see if all the word forms
are represented.
2. Click on Node Tags to make some other useful observations.
3. Click on Doc IDs and check that the word is not used disproportionately in one
document, which does happen, especially with uncommon words.
4. See what Visualise offers in terms of spread across the corpus.
5. Click on Text Types to see in what genres slouch is most common?
Question 81 What other patterns of normal usage can we infer about slouch from a
left sort?
• Is there anything of interest about the subject when slouch is a verb?
• Only people? Both males and females slouch?
• Does slouch ever start a sentence?
• Is it used in both positive and negative contexts?
If being no slouch is a positive thing, what does this tell us about slouch itself?
Do any patterns of normal usage emerge about slouch from a right sort?
• Is it followed by any particular prepositions?
• Are the prepositional phrases places indicating where people slouch?
• Are any of the prepositions part of a phrasal verb with slouch?
• What is a slouch hat?
• Is the verb a troponym of (a) standing or (b) moving?
Question 82 Do we elaborate something, or elaborate on/about/over something?
Are there other COLLIGATION patterns to be aware of, e.g. preference for passive?
Remember that if you
have your page length
set to at least the number
of hits (p.41), you can
scroll down and see
quite a few patterns at
once.
Top Level Queries continued
71
Search for the lemma and specify verb. Sort to the right as that is where the patterns
in our data need to be observed34
. Beware of the tagging errors here – this
word is also an adjective, in which case its pronunciation is different35
.
To answer the question about elaborating something, it is necessary to look for S V
O patterns: who elaborates what? It is worth noting all of the objects and to see if
they form semantic sets. In the section on Clustering (p.160) we look into this idea
of grouping words into semantic sets that occur in a particular syntactic slot.
Left sorts do a similar job. Verbs are preceded by other elements of the verb phrase,
such as modals and auxiliaries. This grouping of the auxiliaries reveals the word's
use in passive voice, perfect aspect, to-infinitive form, thanks to the PERIPHRASTIC
nature of English.
To elaborate a little on the definition I gave earlier, an 'edition' is any number of books,
small or large printed from one setting of type.
Her answer was crisp, and she did not elaborate any further.
This point will be elaborated further in the discussion of 'crime' in the next section.
Although he discusses and exemplifies the other maxims, Grice does not elaborate on the
simple instruction 'Be relevant'.
William Stukeley further elaborated the legend in the eighteenth century when he proposed
that …
These corpus examples also represent the polysemy of this verb. Examining its collocations and colligations
even in this tiny sample, the associated patterns with each meaning can be gleaned. These
lexicogrammatical patterns are at the core of WORD TEMPLATES as we explore later.
The high frequency of and suggests a pattern of normal usage when another verb precedes it frequently,
e.g. developed and elaborated occurs five times.
Further discussion of elaborate can be found at the MAELT website36
.
Question 83 Are both centre and center used as verbs?
Does the BNC represent both spellings? Use LEXCMI for a better sample of US and
UK varieties? When did American reform its spelling?
Search for the lemma and then click on Node Tags in the Left Panel to see what its
parts of speech are.
What are the patterns of normal usage of the verb, centre? Is it used in both transitive
and intransitive structures? What are its typical subjects and objects? Are there
any salient prepositions?
34 ske.li/bnc_elaborate_vb_colls_r
35 http://bit.ly/howjsay_elaborate
36 http://bit.ly/maelt_elaborate
Discovering English with Sketch Engine
72
Verbs of Motion
Verbs of motion, such as the troponyms of go, are intransitive, i.e. Subject + Verb, so when investigating
the valencies of these verbs, we are primarily interested in the subjects. This information is
part of the semantics of the verbs and it provides a structure for using them. For example, are the
verbs performed by people only?
Here are some more words for ways of walking:
tramp, ramble, rove, tiptoe, lope, wander, swan,
flounce, mince, pace, step, tread, trudge, meander.
Such a list raises such questions as:
• What are the differences between them?
• What use are they? Who uses them and when?
• Why are there so many ways of walking?
• Are there as many ways of doing other things?
We look at approaches to these questions under componential analysis (p.76), word templates
(p.161) and sketch differences (p.168).
Question 84 Do only people lope, meander, tread and mince?
If such verbs have other subjects, do they mean something quite different, or are
there common elements of meaning?
Also of interest to verbs of motion are the adverbs that describe how they are performed and more
significantly in valency studies, the spatial adverbials that express where the verb takes place, the
range and the destination.
Many verbs of motion are also used transitively. This means that the pattern is Subject Verb Object
(S V O). To find some examples, search for some of these verbs + the, e.g. walk the. You can expect
to find S V O, which by definition, is not a verb of motion. They will have a different meaning and/or
usage.
These verbs often combine with adverbs and prepositions to make phrasal verbs, which are discussed
in many places up throughout this book. See the Index. They express a wide range of transi-
tivity.
Theseverbs alsooccur in idiomatic phrases,e.g.walktall,walk the plank,walk the beat.Match these
three phrases with these three glosses: (a) a form of execution, (b) patrolling, (c) pride. Now find
idiomatic usages yourself.
Many verbs of motion also occur in delexical structures, such as take a stroll, in which the verb is
"nouned". And many of the noun forms of these verbs are used in various compounds and expressions
such as a walkover, strollathon, runaway, bush walk, protest march.
Top Level Queries continued
73
Question 85 Are there any patterns of normal usage in the adverbs that describe different
semantic groups of verbs of motion?
Compare the adverbs used with verbs of walking with those of travelling. Or those
of Body-Internal Motion (Levin 1993: 49: 261)23
fidget, gyrate, rock, squirm, twitch, wriggle.
Question 86 Can you think of some other series of troponyms that start with go, not
in a walking sense?
Series of troponyms and other semantic relationships are given at Wordnet24
.
You could make a hierarchical diagram or mindmap with go as the headword.
By now we have arrived at some answers to 0. And hopefully much more. Let us
now move on.
Phrase
The definition of a phrase for this search field is 'a group of adjacent (a.k.a. contiguous) word forms
in a fixed order'. The definition of phrase in general parlance is not as clear. According to Oxford
Dictionaries37
, a phrase is:
1. A small group of words standing together as a conceptual unit, typically forming
a component of a clause.
1.1 An idiomatic or short pithy expression.
The following definition comes from the Glossary of Linguistic Terms25
:
A phrase is a syntactic structure that consists of more than one word but lacks the
subject-predicate organization of a clause.
So we can see that the definitions of phrase are more general than its definition for the Sketch Engine
search field.
A search for go fish in the phrase field returns these four lines only in the New Model Corpus (NMC),
37 http://bit.ly/oxdict_phrase
Discovering English with Sketch Engine
74
I had to laugh at your comment, go fish. Yes, we were a happy bunch in the 70’s weren't
we?
We know that searching in the Simple Query field returns the inflected forms of
both items: go fish returns 20. However, when go conjugates and fish declines, we
no longer have the fixed phrase, go fish.
By the way, is the apostrophe correct with 70’s? (See p. 33). It is important to remember
that incorrect, non-standard, deviant, erroneous forms, call them what
you will, occur in the language that we are exposed to, our INPUT.
The phrase field is also case-sensitive: Go fish returns seven lines, and Go Fish returns four.
We havejust seen thatsearching for gofishin the SimpleQueryfield produces differentresultsfrom
the phrase field. But in both cases the two lemmas are adjacent. If we want to search for words that
might appear between them, we use the Context search which is described from page 95 onwards.
Question 87 I wonder if I had to laugh is a fixed phrase in English.
0 opens with the phrase, I had to laugh. This could arouse one's suspicions, pique
one's curiosity.
The BNC seems to think so, but only in a few text types. It usually stands alone.
What can you find out about this phrase?
Question 88 I wonder if or I wonder whether?
Both of these are bountiful in the BNC but not equally so. Can you identify any systematic
usage in one over the other?
Question 89 Does I wonder open an indirect question that requires subject-verb in-
version?
For example, What time is it? inverts after an introductory clause: I
don't know what time it is. Look in the BNC at I wonder and see if
this inversion is standard. A sample of 1,000 right sorted lines is
quite revealing.
Question 90 Do the subject and verb typically change places in indirect
questions?
To answer this question convincingly, study the different wh- words
and the word order of their clauses.
Search in the Phrase field with: Do you know wh.*.
Here is just one example exhibiting this inverted word order.
Wh- words:
who
whose
what
when
where
why
Top Level Queries continued
75
Do you know whose button this is?
After performing the search, set First level as node. Change the attribute in the Second
and Third levels to Tag38
. Pronouns certainly dominate.
Click on the P beside any of the lines to see sentence examples.
Question 91 Is the phrase in mixed company a euphemism, code, a cultural refer-
ence?
Mixed company was referred to earlier in the context of fig-leafed language (p.9):
Victorians and vicars birdwatching in mixed company blushed when they
spotted one.
Search for it in the BNC using the phrase field. The small number of hits is indicative
of its limited use in English. Reading either side of the node may lead to recognising
its reference to social acceptability.
Question 92 To return briefly to Margaret Thatcher, does her nickname appear in
the BNC?
Is it used as frequently? Do iron lady and Iron Lady return the same results?
Before you look, try to predict the answers. Check it in the Phrase field, then using
Sentence instead of KWIC view, compare what is said about the lady with what is
said about the nickname itself.
Question 93 Is the 'way how to do something', a structure in English?
Before considering the structure, consider the collocation: how frequent are the
two words individually? Search for them and write them under Frequency here.
Word 1 Frequency Word 2 Frequency Co-occurrence
way how
Before searching for the phrase, way how, think about how frequently you would
expect them to co-occur.
Type it into the Phrase Field and see how often this pair co-occurs. Then see how
often it is followed by a to infinitive. Is this a pattern of normal usage?
38 ske.li/bnc_doyouknow_tag_tag (UK)
Discovering English with Sketch Engine
76
It is worth noting that other 'way-like' words
do not occur in front of how either. Try some
of the following: strategy, system, approach,
plan, method, technique + how. The nouns
that do precede how can be seen here39
. This
search was filtered as shown in this screen-
shot.
Question 94 more than likely is perfectly standard English. In what situations?
In what sort of texts is it used?
Is it used just as a reaction, or in the course of text?
What does it express the speaker's attitude towards?
Question 95 Can we say more than unlikely?
We can say anything. But do we? Significantly? In specific contexts? If not, what are
some alternative wordings?
Question 96 Are there any patterns of normal usage of more than?
Search in the Phrase Field and form bundles using Frequency 1R. And 2R. Select Ignore
case.
Fixed phrases, chunks and bundles
The explosion of interest among linguists in multi-word units in the last twenty years has seen an
explosion of terminology. The innovating linguist has sometimes used a known term in their own
specific way or created a new one. The following componential analysis of three of these terms is
unlikely to meet with unanimous acceptance among linguists: it can be seen as an attempt to distinguish
them as a basis for further work, and to clarify how they are used in this book.
39 ske.li/bnc_1_n_how (UK)
Top Level Queries continued
77
MULTI-WORD
UNITS
Phrase Chunk Bundle
semantically whole
yes yes not necessarily
flexibility of words
permits variation fixed word forms only
length
depends on mean-
ing
depends on mean-
ing
fixed (n-gram)
storage
holistic holistic not stored
phonology
usually one nucleus usually one nucleus
not pronounced as
units
Native speakers of any language have vast stores of multi-word units (MWU): using them contributes
significantly to fluency and idiomaticity. They are not tagged as MWUs in corpora which means
that we cannot derive them from corpora automatically. Rather, searches reveal them as patterns
that hover around specific words and combinations.
Question 97 What can you discover about the following chunks?
What we need to know about chunks is how they fit into streams of language. When
they appear in KWIC format, and sorted left and/or right the company they keep
emerges quite clearly. Furthermore, when observed in the BNC, the metadata will
indicate the registers and domains in which they occur.
Some of these will be more fruitfully sought in larger corpora such as ukWaC and
enTenTen.
every so often if truth be told only a matter of time
age before beauty never walk alone alive and well
by and large all very well be it on your own head
clean and tidy gloom and doom from cradle to grave
crux of the matter food for thought given half a chance
Learners of English would do well to prioritize such chunks in their study plans. This involves observing
them in their contexts and getting a feeling for how they might meet a need, especially when
there is no direct equivalent in the first language.
Discovering English with Sketch Engine
78
Question 98 Does the phrase, once in a lifetime, function as an adjective?
It occurs 23 times in the BNC, followed by nouns such as affair, event, find, holiday,
opportunity. How is it tagged?
Sentence stems
At the heart of an English sentence is a clause which expresses its message. In addition to messagebearing
clauses, sentences also contain single and multi-word adverbials, e.g. although, be that as
it may, that express relationships between sentences and across text.
In some linguistic work, such organisational language is referred to as sentence stems. They make
important contributions to text on the levels of discourse, stylistics and pragmatics. These are often
chunks expressing temporal and causal relationships in the discourse.
Apart from single word items, they are mostly prefabricated chunks, which ensures that they are
well-attested in corpora. With the Phrase Field being case sensitive, these sentence-initial chunks
can easily be found. For example,
Question 99 How are the above sentence stems followed up?
As we asked in Question 97, what are their collocation and colligation patterns?
It is valuable to have such chunks at your fingertips, but knowing how they are typically
followed up is also part of knowing them. Look them up in the Phrase Field
and use Frequency to see what follows them at various levels. It is sometimes surprising
to find that even the most general of these are followed by quite a limited
set of items.
Make sure you are showing the David Lee's Classification as well. There are observations
to be made.
Remember that Given that
It may be the case that It should be noted that
There is no doubt that It is clear (and other adjectives)
that
It appears/seems that It has been suggested that
Having said that This does not mean that
Despite the fact that It was agreed that
Top Level Queries continued
79
Word Form
Unlike Simple Query and Lemma searches which search for lemmas as well as word forms, words
sought via the Word Form field are returned only as entered. This is particularly useful when investigating
the use of a word form that is also part of a lemma, for example, the noun going is a word
worth studying separately from its lemma go.
Question 100 Are both though and although used to start sentences? Equally?
Perhaps you already know the answer to this through intuition, or through being
taught it. Match Case is useful for identifying words at the start a sentence.
You don't have to process tens of thousands of lines. Rather, click on Node Forms
in the left panel.
Question 101 Is it true that and and but cannot start sentences?
These two questions deserve deeper answers than yes or no – check the Text Types
as well. Also, compare the hits per million of those that start sentences with those
that don't.
Question 102 Is the plural mouse the same for the animal and the computer periph-
eral?
If mouses occurs in corpora at all, check the text types as well as context and cotext.
The frequency of your findings with its co-text should be decisive.
The CHILDES corpora are used for the study of first language acquisition. Even
though young children are unlikely to ever hear the word mouses, at least not from
their caregivers, once they start to form plurals according to the paradigm that they
figure out, they do form the plural as if it were regular. This is after a period of having
used the irregular form correctly, based on imitation.
See the English CHILDES corpus40
for the instances of mouses sorted according to
the age of the children. They also form it correctly: search mice in the Word Form
field. Compare the ages of speakers of mouses and mice.
Speaking of childs, do children form this plural according to the rules also?
Children go through the same language acquisition stages with irregular verbs: they
use the correct forms first, then goed, comed,taked, then they realise that there are
40 ske.li/childes_mouses_age
Discovering English with Sketch Engine
80
both regular and irregular forms. For more on this a well-documented phenomenon
in FLA, see Jean Berko and her wug tests, which has many citations on the in-
ternet26
.
Question 103 Have you ever wondered if some surnames are also common nouns?
For example: Thatcher, Butcher, Baker, Teacher, Hooper, Cooper, Blower, Archer,
Brewer, Parker. This must pique your curiosity about the origin of surnames.
Some readers may be familiar with Bill Posters. Open this link41
and identify some
of the patterns of normal usage around the node. You might wonder if Bill is a person,
etc.
If the collocation pique curiosity piques your curiosity, see this page via the per-
malink42
.
Countability
The issue of countability of nouns looms large43
in learning English as a foreign or second language,
although it seems that few native speakers are aware of the concept. In the first of the next two
corpus sentences, scholarship refers to academic pursuit and has no plural form, whereas in the
second it is a financial award and there can be more than one of them.
Fine art, architecture, scholarship, literature, music and piety jostled for attention alongside
hunting, feasting, jousting, politics, diplomacy and war. (ukWaC)
Lawrence won a scholarship to Nottingham High school, leaving at the age of 16 to work
as a clerk in a local factory. (ukWaC)
Another example is the word experience. In the first of the next two sentences, it refers to
knowledge of the world and has no plural form, whereas in the second it is an event of which there
can be more than one. Note that in other languages, these two notions have quite distinct forms,
e.g. Czech has zážítek and zkušenost, while German has Erfarhung and Erlebnis.
I am a political refugee with personal experience of being the target of hate speech …
(ukWaC)
British gay men reported having at least one experience of unprotected sex last year, ...
(ukWaC)
Countability is an important factor in the use of determiners including the definite and indefinite
articles. We shall not investigate this here.
Mass nouns such as water, toothpaste, money, homework and coffee only have plural forms when
they refer to discrete entities, however, not all of them do. Can you find some illustrative sentences
in corpora comparing these uses?
41 ske.li/ett_bill_posters (UK)
42 ske.li/bnc_pique_curiosity
43 ske.li/bnc_loom_large
Top Level Queries continued
81
We therefore use quantifying nouns that do take the plural if necessary, e.g. tube of toothpaste, bar
of soap, cup of coffee, piece of homework.
English also has expressions of quantity such as array, bunch, group, heap, multitude and range,
which are typically followed by of.
All of these nouns + of can easily be sought in the Simple Query field and a frequency listing will
show the items that follow them. It will show any items that meet the criteria, but human intuition
will distinguish bars of chocolate and soap from bars of music and bars of a cage.
Question 104 Are the 'quantity nouns' that follow of in the singular or plural?
Here are two examples. Note that in the first, multitude are (plural) and in the second
a multitude binds (singular).
A multitude of medical conditions are due to being overweight, which in many people
may simply be the product of self-indulgence rather than food addiction.
…, yet a multitude of non-verbal signals binds them together in a group.
Use the Word Form field to investigate some more examples of these words as they
are used in the singular. They can also be used in the plural, but then agreement is
not contentious.
Some nouns are used mainly in the plural. To search for the plural form only, use Word Form.
Question 105 Are singular nouns always used with a determiner? Do they have singular
verb agreement?
Singular nouns are nouns which have no plural in at least one of their senses. For
example, ability, brainchild, crux, dearth, epitome, fray, grounding, heartbeat, impasse,
and etc. through the alphabet. See Hunston and Francis27
.
1. Choose some of the nouns listed above and search for them in the Simple Query
field.
2. Then look at the Node Forms – we are expecting they will reveal a high proportion
of the singular form. Make a note of the frequencies.
Discovering English with Sketch Engine
82
3. Click on the P beside the singular form. To isolate
those preceded by determiners, click on Filter.
Here we need to enter the CQL query, as in
the screenshot here, [tag = "DT"] under Query
Types, and set the range -1, 0, as in this screenshot.
In the Contextsection (p.95), wewill see an
alternative way of filtering such a query.
4. Compare this frequency with the one you noted.
To see which determiners are used, use Frequency
again – 1L + node. Is there one that is salient,
in that is, it stands out from the rest?
5. To see the bundle in which this appears, use several
levels in Frequency, e.g. 1L Node 1R. Are there any salient bundles? It's Zipfian!
This was a long journey, but the destination was not the only aim. There are no shortcuts for the
pioneer – only once the trail has been blazed can the search for answers be shortened.
Character
While this field is of particular relevance to languages such as Chinese and Japanese, it serves English
in some useful ways too.
Brackets and parentheses
Bracketed or parenthetic information is often an admission of irrelevance or downgrading: by putting
a comment in brackets, the author is saying, well it's not germane to my point, but it occurred
to me while I was writing and I thought you might be interested ….
In contemporary writing, therefore, there is very little information
in brackets (even footnotes rarely contain information). This makes
reading a very smooth process indeed. It places a demand on authors
to work everything germane into the regular syntax of wellcrafted
sentences, or to leave some material out. In the following
sentence from the BNC, the author uses parentheses twice and for
different purposes. Curiously, she gives some examples of parenthetic
information in parentheses!
His stimulus material was a film and he particularly noted the expression of parenthetic
information (asides, or personal expression of feeling) in the recall of such events (something
which does not appear in recall of verbally presented stories).
The second use is more of an aside, something that she mentions in the first paren-
theses!
An aside is a dramatic device
in which a character speaks
to the audience. By convention
the audience is to realize
that the character's speech is
unheard by the other characters
on stage. (Wikipedia)
Top Level Queries continued
83
Question 106 How much information typically appears in brackets?
Type a left bracket into the Character Field. Make a sizeable Sample and Sort to the
right.
What appears in brackets? How many words? Are there any clauses, i.e. containing
a finite verb? Give some thought to the reasons for putting such things in brackets.
And observe the text types of your findings.
Consider carefully the use of brackets in your own writing.
@ a.k.a. at
If you are looking for email addresses, enter @ and click Node Forms. These are somewhat more
frequent in more recent corpora than the BNC.
Question 107 How early did people have email addresses?
Starting with the BNC, enter @ into the Character Field. Use the Select Lines check
boxes to separate the email address from the other uses of @. And in View Options,
choose Publication Date.
Question 108 How is the at symbol used when not in email addresses?
Search for it in the character field. Sort node.
Question 109 Are capital letters used after colons and semicolons?
Type these punctuation marks into the character field, perform searches and then
use Frequency to make a list of at least 1R.
Question 110 Is it true that above-mentioned (with or without a hyphen) and aforementioned
are used in written and spoken language respectively?
Question 111 How do answers begin?
Type a question mark into the character field. This contrasts with Question 24: Do
you love me (p. 34), where we could not use a question mark in the Simple Query
field. There are more ways of killing a cat, as the old saying goes.
… there are more ways of killing a cat than skinning it …
Truly, as the old saying goes - 'there are more ways of killing a cat than choking it with
cream!'
Discovering English with Sketch Engine
84
Operators used in fields other than Simple Query
The following operators work in these fields: Lemma, Phrase, Word form. Some of them behave differently
in CQL, as we will investigate later (p.124ff).
full stop before, during or after a word, a full stop finds that word with any letter in
place of the full stop.
In the Word Form field, b.g gives big bag beg bog bug. Click on Node Forms
to obtain the list. This is useful for pronunciation teaching in generating lists
of words that can be used in minimal pairs activities.
In the Lemma field, .rick gives all word forms of the lemmas that have these
four letters starting in second place, e.g. brick, tricks, pricked, tricking. This
typically offers words that rhyme.28
Note that in Simple Query, a question
mark provides this list.
asterisk after a full stop finds any number of letters, including none, in place of the
full stop. In the Lemma field, b.*g gives 153,551 hits44
, including broadcasting,
brainstorming, bing, beg. In the Word Form field, it returns 264,766 hits.
.*ship finds words ending with ship
oxy.* finds words beginning with oxy
.*advantag.* finds words containing advantag of which there are 27 different
ones. Use Node Forms to create this list.
backslash To search for a full stop, use the backslash and the full stop. In computing,
this is referred to as escaping.
Dr\. finds Dr. only. Without the backslash, you also get Dr., along with hundreds
of other three-letter combinations starting with dr.
And now for a quick comparison:
Simple Query ?ock ⇒ 22,695 does not work with a full stop
Lemma .ock ⇒ 22,685 does not work with a question mark
CQL ".ock" ⇒ 22,685 lemma attribute does not work with a question mark
44 http://ske.li/bnc_b.*g
Top Level Queries continued
85
Affixes
The full stop operator is like a blank tile in Scrabble – it can stand in for any letter. Used alone it is
particularly useful for alternative spellings. In combination with the asterisk however, we discover
what prefixes and suffixes attach to stems. And as we have already seen many times, some things
will be very typical and others less so. Given our creative use of language, many novel words are
created in this way, as we see among the less frequent ones. Here, for example, are six of the dozens
of recent coinages, a.k.a. neologisms, starting with faux that occur once only in enTenTen.
faux-modest, faux-sarcasm, faux-sophistication,
faux-patriotic, faux-lesbian, faux-deferential
A word that only occurs once in a corpus is referred to as a hapax legomenon.
Question 112 Which words beginning with faux occur in the BNC?
Search *faux in Simple Query or .*faux in other fields. What does the BNC data indicate
about the recent history of this prefix?
Question 113 Did any of these words catch on and enter the language?
The biggest corpus in Sketch Engine is enTenTen. Search for them there. And there
is always the world wide web.
By the way, do words catch on? What do you think of this collocation?
Question 114 Look for them in corpora more recent than the BNC, e.g. enTenTen,
ukWaC.
Nonce words are created for a specific occasion and cannot be expected to be met
again – they often go unrecorded. In corpus terms, many nonce words are also hapax
legomena. According to Zipf's Law, between 40% and 60% of words in a corpus
can be expected to occur only once. We will use the Word List tool (p.173) to search
for words that occur once only.
New words in a language may start their lives as nonce words. These can only be
open class words.
Question 115 What are some examples of the suffix –ity indicating the state or quality
of being the adjective (LGSWE, pp 232-3)?
Try .*ity. This gives all the words that end in -ity. Those which are the abstractions
ofadjectives are typicallysingular nouns (seealsoQuestion 106). Nevertheless, performing
this search in the Lemma field returns plural forms as well, e.g. opportunities,
eccentricities. Are these also abstract forms of adjectives?
Discovering English with Sketch Engine
86
Question 116 Is dis the only prefix that our titular word, discovering, takes?
Question 117 Search for .*cover in the Word Form field. What are the differences between
the words that you find? Do their contexts and cotexts provide
any clues?
Question 118 How do other prefixes fare?
Try mega.* in the Word Form field in the BNC. Click on Node Forms – the list has
Megan at the top while the following occur only once:
megafirm, megafabulously, megamoney, megamum, megamouth
You might also try hyper.*, over.*
Question 119 Given that Megan is a fairly uncommon name, how did it get such a
high score?
How many Megans do you know or know of?
Wikipedia's list of Notable Megans29
might jog
your memory.
Clicking on Text Types and David Lee's Classification
shows that more than half come from fiction.
Click on the P.
To find out what books they come from, select
bnc.doc title in View Options' References as
shown in the screenshot.
Question 120 Can we identify any language change through suffixes?
The masculine form of words used to refer not only to men, but to mankind, and
was therefore considered gender neutral. There has been a shift away from this and
person has become the suffix replacing man. For example, chairperson is now preferred
over chairman, regardless of the gender of the person occupying that position.
Search various corpora to find both of these endings and compare them.
Similar investigations could be conducted with the new, politically correct suffix
challenged.
Top Level Queries continued
87
Discovering English. Does English need discovering? Many linguists, of the armchair and empirical
varieties, have been writing for over a century about English as she is spoke30
and writ. Now that
English has achieved lingua franca status, describing her every curve and nuance has virtually become
an industry. This serves as a great impetus to linguistic departments in universities – their
findings are filtered down to a vast community of language professionals by a relatively small band
of journal and book publishers. Since the work of teachers is above all to teach, and of translators is
to translate, language professionals are very glad of the work that linguists do. As Gleiser recently
wrote, in defence of scientists:
Behind the scientist’s complicated formulas, the tables of data obtained from experiments,
and the technical jargon, you will find a person eagerly trying to transcend the immediate
boundaries of life, driven by an unstoppable desire to reach some deeper truth. (2005)
Nevertheless, of great interest to the current book is treating "the learner as a research worker
whose learning needs to be driven by access to linguistic data and whose role is to identify, classify
and generalize that data" as Tim John wrote almost a quarter of a century ago. For this he coined the
widely-used term, data-driven learning. Thus, we are less interested in the findings of linguists for
the consumption of teachers, learners and translators than in the process of discovery which allows
learners to create knowledge for themselves.
Piaget (1896–1980) said: [pic]
Each time one prematurely teaches a child something he could have discovered for himself,
the child is kept from inventing it and consequently from understanding it completely.
As elegant as this may be, the support of a teacher can enable leaners to achieve things that they
never would alone. This is part of what Vygotsky (1894–1936) meant by his Zone of Proximal Development.
And guided discovery is a logical consequence of this co-operation between the teacher as
the guide on the side (King 1993), and leaners. Teachers facilitate learning rather than transmit information.
Johns again: the task of the learner is to "discover" the foreign language, and that the task
of the language teacher is to provide a context in which the learner can develop strategies for discovery
- strategies through which he or she can "learn how to learn".
To return to our question: Does English need discovering? In our work so far, Discovering English
with Sketch Engine has been enabling you to obtain language data that you can interpret to answer
set questions with the help of instructions for the Sketch Engine's tools as they are introduced step
bystep.The process isintendedto developyour awarenessoflanguage asa linguisticobject, provide
youwith information about English that may benew to you,and above all to equip youwith theskills
to find answers to your own questions. It is hoped that knowledge is being created within you every
step of the way. This is what needs discovering.
On with the show!
Discovering English with Sketch Engine
88
Question 121 Are there any affixes making verb forms out of friend ?
In the BNC, put .*friend.* in the lemma field and choose verb from the POS droplist.
Then click on Node Forms.
Repeat this search in the enTenTen corpus. How do the node form lists compare?
Do you find that the standard form still has a strong place in contemporary English?
Question 122 It emerges that be- is often used as a prefix. Does it have a meaning, or
perhaps a function?
Is this the case with both nouns and verbs? Perform a search and use Node Tags.
Now search for be.* in the lemma field and select verb. Make a frequency list with
lemma as the attribute.
The words of interest are those which are obviously be + free morpheme, e.g. befriend,
bewitch, bejewel.
The data is not the answer – it is a starting point, as we have seen in the more or less
100 questions so far. The question asks about the meaning of the prefix.
Question 123 Why do people say things like not uninteresting? Is it not enough to say
is interesting?
We find that not uninteresting occurs once only in the BNC and 162 times in en-
TenTen45
. Using .*, however, we find that not un… is quite well-represented, even in
the BNC46
.Is this negative particle + negative prefix a patternof normalusage?Does
it belong to any particular text types? If not, can any other explanation be found in
the data?
We will now observe how this operator can be used between words.
Question 124 What’s the correct form: no holds barred or no holes barred?
Search for no .* barred in Simple Query.
Furthermore, can you infer the meaning and/or function of the phrase from its cotext
and contexts?
Question 125 I’ve heard something that sounds ironic: in the looks department.
Apart from actual departments, what departments are referred to?
Try a search for in the .* department, then click on Node Forms.
45 ske.li/ett_not_uninteresting (UK)
46 ske.li/bnc_not_un_freq (UK)
Top Level Queries continued
89
As we have already noted a number of times, patterns of normal usage emerge at
the top of the lists and graphs, while the exploitations of the language are inevitably
lower down. Hanks' Patterns of Norms and Exploitations (2013) is a useful dichotomy.
Some examples in the BNC: the brain department, the dick department, the
butchering department are at the bottom in terms of frequency. There is even one
sighting in the NMC of in the irony department! Is that eponymous or just plain
ironic?
An interesting facet of languagethatcorpus studies haveunearthed, is the influenceword meanings
have on each other when they habitually co-occur.
Firth noted in his work on phonology, there are many ways in which a sound can be influenced by its
environment – he referred to this as prosody31
. By analogy, some of his followers applied his term to
meaning (semantics) – thus SEMANTIC PROSODY. For example, the objects of the verb to harbour (BNC
434: 3.8 per million) include grudge, resentment, bacteria, terrorist, hatred, prejudice. Adjectives in
the vicinity include ineffective, deep-seated, irrational, hostile, illegal. This negativity around the
verb is its semantic prosody. A list of collocates47
of harbour is quite revealing, as we will study in
greater detail from p.143 onwards.
Question 126 What semantic prosody emerges when the phrase or just plain is used?
Search for this in the Phrase field, and make a Frequency list of the words 1R. They
are either nouns or adjectives, which means that the POS of plain must vary here.
To find out how it has been tagged, click on Node tags or use lempos as an attribute.
And what do you notice about the nouns and adjectives in terms of their prosody or
connotations?
Question 127 Is it true that more English words have k as the third letter than as the
first?
This may seem like a bizarre question, but it concerns a finding by psychologists
Kahneman and Tversky (1973) about the way humans process information32
. Before
we investigate it, what do you think the answer is?
To find words with k as their third letter, type ..k.* into the lemma field.
To find words that start with k, type k.* into the lemma filed.
With the results of both, use Frequency: choose attribute lemma, Ignore case and
set the Frequency limit to 1. The first thing we notice is the Zipfian tendency for a
small number of words to have a great number of occurrences.
47 ske.li/bnc_harbour_v_colls (UK)
Discovering English with Sketch Engine
90
To find out how many such words there are, it is necessary to keep clicking Next till
you get to the end, and multiply the number of pages by the number of lines per
page. There are 79 pages of lemmas with k as their third letter48
.
Unfortunately, there is no 'Go Last' button that would jump to the end of the list in
a single click. However, along the way, you will observe some interesting words and
non-words33
!
To exclude hapax legomena, make the Frequency limit 2. In the case of words starting
with k, this reduces the number of pages from 43 to 2049
, in accordance with
Zipf's Law.
The answer to the actual question is yes, contrary to the intuition of most of their
subjects. Remember that computers process information differently from people.
Vertical bar
The vertical bar can be placed between items so that the concordancer searches for them at the
same time. In regular expressions it means OR.
Question 128 Can amid and amidst be used interchangeably?
Use the vertical bar between them: amid|amidst. Make a Sample, investigate their
text types, etc. Can you find any systematic difference between them? Whatever
answer you arrive at will influence how you choose to use one or the other.
Whilst investigating this pair, observe their semantic prosody. It is of surprising interest.
Once again, a list of collocates50
is revealing.
Question 129 Are the same prepositions used with synonyms?
struggle|battle|fight searches for all of these items. This particular search reveals
interesting patterns in preposition use. Since all three are both nouns and verbs, it
is worth choosing PoS when making the query. Various frequency listings are useful
here.
Question 130 Can the vertical bar be used to look up British and American spellings
at the same time?
Try color|colour and centre|center, for example. Try such searches in both the BNC
and LEXCMI and see which corpus is the more British!
48 ske.li/bnc_k3_lastpage (UK)
49 See where the hapax legomena begin: ske.li/bnc_k_words_p20 (UK)
50 ske.li/bnc_amid_colls (UK)
Top Level Queries continued
91
Corpus Query Language – a brief introduction
Until now, we have been searching corpora
by typing into the Simple Query, lemma,
phrase and character fields. In the background,
Sketch Engine converts these into
Corpus Query Language (CQL) queries,
which you can see by clicking on ConcDesc at the bottom of the left panel. This screenshot shows
the CQL that was generated for Question 131.
CQL is not only an alternative way of forming searches, but it has many features that enable very
specific searches. In many cases there is no alternative to a CQL query. Here are some questions
that simple CQL queries can provide data for.
Before we start, check that you are using BNC (TreeTagger). The BNC without TreeTagger uses
CLAWS with which the following steps are incompatible.
Question 131 How many sentences are there in the corpus?
The Penn TreeBank tag, "SENT", which marks full stops, question marks and exclamation
marks, invites a wide range of questions, as we are about to see.
Knowing the number of sentences in a corpus, especially a specialised corpus or a
subcorpus, allows calculations of average sentence length, which is an important
stylistic consideration.
Type the query, as shown in the screenshot: the tag in capitals between non-curly
quotation marks; choose tag as the Default Attribute. Then use some of the tools to
check the reliability of the tagging. For example, are decimal points counted as full
stops?
Question 132 What are some common ways of starting sentences?
This might not be a question that keeps you awake at night, but the observations
that follow from it can lead to great things. It is not uninteresting to compare the
use of stance adverbials, disjuncts, conjuncts,
discourse markers, backchannels and sentence
stems across different corpora, different
text types and different contexts. The data
can also be used for questions concerning
given and new information, theme and
rheme34
.
Discovering English with Sketch Engine
92
This asks the software to generate concordances of sentence punctuation, thus full stops, question
marks and exclamation marks becomes the node and new sentences follow.
It is then simply a matter of using the Frequency tool to make lists of words or BUNDLES up to four
TOKENS long. Some of the tokens in these four positions are not words – sometimes they are punctuation
and we often see the negative particle separated from its stem, as in do n't, which makes it
two tokens. Click the P to see the sentences. This screenshot shows a Frequency list of some spoken
sentence beginnings.
Quite different and informative data emerges when we process the data differently. For example,
you can:
1. make a sample.
2. include the Node in the Frequency list.
3. select different attributes, e.g. a Frequency list of tags generates SYNTAGMS, as we
saw under Frequency on p.52.
4. experiment with the Frequency limit at the top of the Multilevel frequency distri-
bution.
5. The default sort is according to frequency. Click the column headings and sort by
the chosen attribute.
Question 133 Does English permit numbers at the beginning of sentences?
In other words, are sentence initial numbers written as words as the style manuals
tell us? After the sentence punctuation tag, we add the tag for cardinal numbers,
then make a node forms list. Enter this into the CQL field and select TAG as the default
attribute.
"SENT" "CD"
To use the data to answer the question, It is necessary to click on some of the "P"s
in the list to see why some of the numbers are not written as words.
Question 134 Is it true that past participles are only used in compound forms, that is,
with auxiliaries?
This question surprises a lot of people and a lot of people benefit from knowing the
answer. To search for past participles, we need "VVN" in the CQL field. Set the Default
Attribute to tag.
We then look to see if there are any instances without be (passive voice and continuous aspect) or
have (perfect aspect and causative). The Sample button can be put to good use here as there are
over 2.5 million past participles in the BNC.
Alternatively, use the Frequency tool to generate columns to the left of the past participles.
Top Level Queries continued
93
Do the concordances contain the lemma have for perfect and be for passive? Does the past participle
perform any other function(s)? There are so many examples of bad tagging that it is necessary
to use the data to answer the question manually. The Select Lines check boxes are useful in such
situations (see p.42 – View Options: Bottom).
Periphrasis
Periphrasis is when grammatical functions are expressed by auxiliary words in combination with
content-bearing lexical words, as is the case in analytic languages. Chinese is a prime example: it
has no verb forms other than the base form and nouns do not have plural or possessive forms. In
contrast, synthetic languages express grammatical function such as tense and aspect with inflections.
See the Italian example below. However, as David Crystal notes, most languages have features
of analytic and synthetic languages (2006:370).
Forexample, in English, the comparative and superlativeforms of adjectives and adverbs are mostly
inflected with –er, -est, but the use of more and most are examples of periphrasis. Delexical verb
structuresalsoexemplify theperiphrastic nature of English becausethe verbs aresemanticallylight
or even empty. We investigate these structures on p. 107.
From a linguistic standpoint, English has two tenses only, past and present, as we have already
noted. From a pedagogical standpoint, however, tense refers to all of the
expressions of TIME: continuous and perfect, which are technically referred
to as aspects, and the passive, which is a voice, contrasting with the
unmarked form, active voice and which does not involve an auxiliary. Unmarked
aspect is referred to as simple.
This chart depicts the use of the two verbs that are used as auxiliaries to form aspect and voice. It
also shows the periphrastic role of do in creating other language constructions.
“Time is a game played
beautifully by children.”
Heraclitus
Discovering English with Sketch Engine
94
In Italian, the passato prossimo is formed with the verb avere (italicized in the table below). This
periphrastic structure is quite similar in both meaning and form to English present perfect. Italian's
imperfetto however, is formed with endings (underlined) and expresses the English used to / would
always, a periphrastic realization.
Future time in English is expressed in ways that reflect or respect our uncertainty about things
which have not yet happened. Because there is no inflection to indicate the future, it is mostly expressed
periphrastically with will, shall, going to and be to. Italian has a set of endings (in bold) to
mark the future.
Pronouns Presente Passato prossimo Imperfetto Futuro semplice
io
tu
egli
noi
voi
essi
scrivo
scrivi
scrive
scriviamo
scrivete
scrivono
ho scritto
hai scritto
ha scritto
abbiamo scritto
avete scritto
hanno scritto
scrivevo
scrivevi
scriveva
scrivevamo
scrivevate
scrivevano
scriverò
scriverai
scriverà
scriveremo
scriverete
scriveranno
Auxiliaryverbs
BE
-ing Continuous
-ed Passive
HAVE
-ed Perfect
obj + -ed Causative
DO
obj + base Questions
neg + base Negative
base Emphasis
- Pro-forms