Mapping Meaning onto Use
Patrick Hanks
Institute of Formal and Applied Linguistics,
Charles University in Prague
Masaryk University, December 2009
Talk Outline
1. How to map word meaning onto word use?
Contrast American and European linguistic theory
2. Dictionaries
before and after corpora
3. Collocations and lexical preferences
4. Practical corpus analysis
Look first for the pattern, not the meaning
5. Creativity in language: how writers and speakers
‘exploit’ normal patterns of word use
Lexical theory in American
• Bloomfield; Chomsky and his followers –
– say a lot about syntax, but little about words and meanings,
– even though, according to Chomsky’s projection principle,
the syntax of a well-formed sentence is “projected” from the
lexical items in the sentence, plus “selectional restrictions”.
– “Meaning is dangerous ground.” – N. Chomsky
• Pustejovsky: The Generative Lexicon.
• Fillmore; Goldberg; Jackendoff: Construction Grammar
– Meanings are carried by constructions
– A construction can be a word (e.g. “sleep”) or a phrase (e.g.
“she slept her way to the top”)
Foundations of lexical research
(in European linguistics)
• Saussure and C20 structuralists
– Semantic field theory (Trier, Porzig, Gipper)
– Structure of meaning in language (Coseriu, Pottier)
– Lyons, Ullmann
• Firth, Halliday, Sinclair
– “You shall know a word by the company it keeps”
– To study the lexicon is to study collocations
– By empirical analysis of corpus data
Making lexical analysis possible
• Wittgenstein (1953):
– “Do not ask for the meaning, ask for the use.”
– “What is a game? Do not assume a common property,
but look and see.”
• Grice (1957, 1975):
– Conversational co-operation presupposes a set of
linguistic conventions that users of a language know
and expect others to know
• Rosch (1975, etc.): Prototype theory
• Lakoff and Johnson (1981):
– The fundamentally metaphorical nature of many abstract concepts
– Regular alternation between concrete and abstract ((e.g. grasp a
handle, grasp an idea)
Dictionaries before corpora
• Based on collections of citations
– Literary language, not everyday language
– “Historical principles” – oldest meaning first
• Or based on introspection
– Introspection distorts data. Example, “total” as a verb
• EXERCISE: Invent a sentence using total as a verb.
– What we think we say and what we actually say are
– Cognitive salience (ease of recall) vs. social salience
(frequency of use)
James Murray (1878) predicts
the need for corpus data
• “The editor and his assistants have to spend precious hours
searching for examples of common everyday words. Thus,
in the slips, we have 50 citations for abusion, but for
abuse, not five.” – James Murray, Presidential address to
the Philological Society, 1878
• Murray was the first editor of the Oxford English
Analysis of Meaning in Language
• Analysis based on predicate logic is doomed to failure:
– Words are NOT building blocks in a ‘Lego set’
– A word does NOT denote ‘all and only’ members of a set
– Word meaning is NOT determined by necessary and sufficient
conditions for set membership
• Instead, a prototype-based approach to the lexicon is
mapping prototypical interpretations onto prototypical phraseology
Prototypical phraseology includes collocational preferences
E.g. what do you hazard? – Typically, you hazard a guess.
Not a necessary condition, but a collocational preference
Patterns in Corpora
• When you first open a concordance, very often some
patterns of use leap out at you.
– Collocations make patterns: one word goes with another
– To see how words make meanings, we need to analyse collocations
• The more you look, the more patterns you see.
• When you try to formalize the patterns, you start to see
more and more exceptions.
• The boundaries are fuzzy and there are many outlying
John Sinclair (1933-2007):
Why is Sinclair important? (1)
Sinclair was editor-in-chief of the Cobuild dictionaries for
foreign learners. (I was the managing editor)
• “Many, if not most meanings, require the presence of more
than one word for their normal realization.”
• “Patterns of co-selection among words, which are much
stronger than any description has yet allowed for, have a direct
connection with meaning.” (Sinclair 1998, The Lexical Item,
page 4)
Why is Sinclair important? (2)
The idiom principle (also known as the phraseological
tendency) vs. the open-choice principle:
“The principle of idiom is that a language user has available to him
or her a large number of semi-preconstructed phrases that constitute
single choices, even though they might appear to be analysable into
segments.” (Sinclair 1991. Corpus, Concordance, Collocation, p.
“Tending towards open choice is what we can dub the
terminological tendency, which is the tendency for a word to have a
fixed meaning in reference to the world. ... tending towards
idiomaticity is the phraseological tendency, where words tend to go
together and make meanings by their combinations.” (Sinclair 2004.
Trust the Text, p. 29)
Some problems with current
• Definitions are not mutually exclusive.
• There is no general agreement on what counts as a
word sense.
• No clear criteria are given in dictionaries for
distinguishing one sense from another.
• There is very little syntagmatic or collocational
information in English dictionaries
• Some Italian, German, Greek, Italian dictionaries
try to give it, but without sufficient evidence
Dictionary definitions are often
not mutually exclusive
• People have tended to assume that dictionary sense
distinctions are mutually exclusive. Quite often, they are
not. E.g.:
pour, v.t.
1. cause a liquid to flow.
2. serve (tea, coffee, etc.) by putting it into a cup.
These defs. look different, but actually 2 is just a subset of 1.
Tea, coffee etc., are liquids, and “putting” is done by “causing to
flow”. Only the social context of 2 is different (more restricted)
Word sense disambiguation
Lesk (1986): ‘How to tell a pine cone from an ice
cream cone’, using OALD definitions:
pine 1. kind of evergreen tree with needle-shaped leaves. 2. waste
away through sorrow or illness.
cone 1. a solid object with a round flat base and sides that slope up
to a point… 2. something of this shape whether solid or hollow.
3. a piece of thin crisp biscuit shaped like a cone, which you can
put ice cream in to eat it. 4. the fruit of certain evergreen trees.
• Lesk here is trying to “disambiguate” by contextual criteria.
• Existing dictionary definitions are his starting point
• He did not look for patterns of word use in corpora
• We want to research new definitions based on actual usage
• Associate the meaning with the PATTERN not the word
Lumping and splitting
Most dictionaries are splitters. E.g. why did
OALD 1963 make these two senses (cone)?
• 1. a solid object with a round flat base and sides that slope
up to a point… 2. something of this shape whether solid or
Why not make it:
• a solid or hollow object with a round flat base and sides
that slope up to a point
This problem is endlessly multiplied in entry after
entry in most English dictionaries.
Dictionaries, collocations,
phraseology: a new approach
• Start with verbs
– The verb is the pivot of the clause
• Distinguish normal uses from creative uses and
– Dictionaries should record norms, not
exploitations of norms
• Use the corpus to find the phraseological and
collocational patterns associated with each verb
Meaning is context
• What does fire mean? file? treat? land? …
• Words in isolation don’t have meaning!
• They have one or more meaning potential
– Meaning potential is multiple and vague.
• Put a word in context, and its vagueness is reduced
or even (for practical purposes) eliminated.
– Theoretically, there is always the potential for extended
Context and Collocations
• “You shall know a word by the company it keeps” – J. R.
• Corpus analysis can show what company our words keep.
• Frequency alone is not enough: “of the” is a frequent
collocation – but not interesting!
• “(the) storm abated” is less frequent, but more interesting.
Contrasted with “(the) threat abated”, it can give a
different meaning to the verb abate.
• We need a way of measuring the statistical significance of
collocations. Sketch Engine provides one.
Mutual information
• A way of computing the statistical significance of two
words in collocation.
• Compares the actual co-occurrence of two words in a
corpus with chance.
• Church and Hanks (1990): ‘Word Association Norms,
Mutual Information, and Lexicography’ in Computational
Linguistics 16:1.
• Kilgarriff, Rychlý, et al. (2004): “The Sketch Engine”,
Proceedings of Euralex 2004. Lorient, France.
Deciding relevant context
Peter treated Mary.
Peter treated Mary respectfully.
Peter treated Mary with respect.
Peter treated Mary with antibiotics.
Peter treated Mary to lunch.
Peter treated Mary to his views on George W. Bush
Peter treated the woodwork with creosote.
Normal implicatures: taking
prototypes and domain seriously
• If someone files a lawsuit, they activate a procedure
asking a court for justice.
• When a pilot files a flight plan, he or she informs ground
control of the intended route and obtains permission to
begin flying. …
(12 more such definitions of file, verb.)
Now we have to find whether any other words that have
similar meaning to “lawsuit” or “flight plan” in this context
• When a group of people file into a room or other place,
they walk in one behind the other.
Norms and Exploitations
• Anyone acquiring a language must learn not one, but two,
forms of rule-governed behavior:
1. The ability to use words normally.
2. The ability to exploit the norms mentioned in 1 in a creative way,
typically in order to say new things or to say old things in a new,
interesting way.
• The norms of natural languages have not yet been
satisfactorily described
due in part to lack of data until about 10 years ago and in part to
linguists treating exploitations as if they were norms).
• Exploitation rules – second-order rules – cannot be fully
described until the norms are known.
How words are normally used.
Patterns of usage.
Descriptive (not prescriptive).
Norms are acquired by priming (M. Hoey) and by
reinforcement (repeated exposure).
• Norms can be discovered by systematic, empirical
Corpus Pattern Analysis (CPA).
• People don’t just say the same thing, using the same words
• They also exploit norms in order to say new things, or in
order to say old things in new and interesting ways.
• Exploitations include metaphor, ellipsis, word creation,
and other figures of speech.
• Exploitations are the rules that govern linguistic creativity.
The CPA method
• CPA: Corpus Pattern Analysis (based on TNE: the Theory
of Norms and Exploitations).
1. Find the significant collocates and tag those lines first.
2. Then create a sample concordance (KWIC index):
from a ‘balanced’ corpus (i.e. general language): BNC50
250 examples of actual uses of the word to start with
3. Classify every line in the sample, by context.
4. Analyse a larger sample if necessary.
5. Use introspection to interpret data, but not to create data.
In CPA, every line in the
sample must be classified
An important principle of statistical analysis.
The classes are:
• Norms
• Exploitations
• Alternations
• Names (Midnight Storm: name of a horse, not a storm)
• Mentions (to mention a word or phrase is not to use it)
• Errors (e.g. learned mistyped as leaned)
• Unassignables
Methodological precepts
• Don’t look for necessary conditions for the meaning of a
word. (There aren’t any.)
– “This elephant is a mouse” is an unlikely sentence of
English – but not meaningless.
• There are innumerable possible but unlikely sentences.
• Don’t try to account for all possibilities. Very few of them
are interesting.
• Corpus linguistics and prototype theory provide a new
focus – on actual and probable sentences and meanings.
A typical pattern dictionary
• irritate
PATTERN 1 (90%): [[Anything]] irritate [[Human]]
IMPLICATURE: [[Anything]] causes [[Human]] to feel mildly annoyed.
PATTERN 2 (8%): [[Stuff]] irritate [[Body Part]]
IMPLICATURE: [[Stuff]] causes [[Body Part]] to become inflamed and
somewhat painful.
• Notes:
Semantic values are assigned to arguments.
Both patterns are transitive (V n), but they have different meanings.
They are distinguished by the semantic types of the nouns.
Getting the right level of semantic generalization for each n is hard.
Another CPA verb norm
BNC frequency: 185 in 100m.
1. [[Event = Storm]] abate [NO OBJ](11%)
2. [[Event = Flood]] abate [NO OBJ] (4%)
3. [[Event = Fever]] abate [NO OBJ] (2%)
4. [[Event = Problem]] abate [NO OBJ] (44%)
5. [[Emotion = Negative]] abate [NO OBJ] (20%)
6. [[Person | Action]] abate [[State = Nuisance]] (19%)
(Domain: Law)
But if you wanted to, you could lump 1-5 together into a single sense.
Unsorted sample from a
incessant noise and bustle had
after dawn the storm suddenly
Thankfully, the storm had
storm outside was beginning to
Fortunately, much of the fuss has
, after the shock had begun to
been arrested and street violence
he declared the recession to be
‘soft landing’ in which inflation
the threshold. The fearful noise
ability. However, when the threat
bag to the ocean. The storm was
ferocity of sectarian politics
storm. By dawn the weather had
abated. It seemed everyone was up
abated. Ruth was there waiting when
abated, at least for the moment, and
abate, but the sky was still ominous
abated, but not before hundreds of
abate, the vision of Benedict's
abated, the ruling party stopped
abating, only hours before the
abates but growth continues moderate
abated in its intensity, trailed
abated in 1989 with a ceasefire in
abating rapidly, the evening sky
abated somewhat between 1931 and
abated though the sea was still angry
[[Event = Storm]] abate [NO OBJ]
dry kit and go again.The storm
ling.Thankfully, the storm had
sting his time until the storm
storm outside was beginning to
bag to the ocean.The storm was
after dawn the storm suddenly
t he wait until the rain storm
storm.By dawn the weather had
lcolm White, and the gales had
he rain, which gave no sign of
n became a downpour that never
ned away, the roar of the wind
abates a bit, and there is no problem in
abated, at least for the moment, and the
abated but also endangering his life, Ge
abate, but the sky was still ominously o
abating rapidly, the evening sky clearin
abated.Ruth was there waiting when the h
abated.She had her way and Corbett went
abated though the sea was still angry, i
abated: Yachting World had performed the
abating, knowing her options were limite
abated all day.My only protection was
abating as he drew the hatch closed behi
[[Event = Problem]] abate [NO OBJ]
‘soft landing’ in which inflation abates but growth continues modera
Fortunately, much of the fuss has
the threshold. The fearful noise
incessant noise and bustle had
ability. However, when the threat
the Intifada shows little sign of
h he declared the recession to be
he ferocity of sectarian politics
abated, but not before hundreds of
abated in its intensity, trailed
abated. It seemed everyone was up
abated in 1989 with a ceasefire in
abating. It is a cliche to say that
abating, only hours before the pub
abated somewhat between 1931 and 1
been arrested and street violence abated, the ruling party stopped b
the dispute showed no sign of abating yesterday. Crews in
[[Emotion = Negative]] abate [NO OBJ]
(selected lines)
ript on the table and his anxiety
that her initial awkwardness had
es if some inner pressure doesn't
Baker in the foyer and my anxiety
hained at the time.When the agony
self; the pain gradually began to
ght, after the shock had begun to
y calm, control it!) The fear was
his dark eyes. That fear did not
abated a little.This talented, if
abated # for she had never seen a
abate.He wanted to play at the fun
abated.He seemed disappointed and
abated he was prepared to laugh wi
abate spontaneously, a great relie
abate, the vision of Benedict's sn
abating, the trembling beginning t
abate when, briefly, he halted. For
isapproval, his kindlier feelings abated, to be replaced by a resurg
(“kindlier feelings” are normally positive, not negative.)
Part of the lexical set [[Event =
Problem]] as subject of ‘abate’
From BNC: {fuss, problem, tensions, fighting, price war, hysterical
media clap-trap, disruption, slump, inflation, recession, the Mozart
frenzy, working-class militancy, hostility, intimidation, ferocity of
sectarian politics, diplomatic isolation, dispute, …}
From AP: {threat, crisis, fighting, hijackings, protests, tensions, violence,
bloodshed, problem, crime, guerrilla attacks, turmoil, shelling,
shooting, artillery duels, fire-code violations, unrest, inflationary
pressures, layoffs, bloodletting, revolution, murder of foreigners,
public furor, eruptions, bad publicity, outbreak, jeering, criticism,
infighting, risk, crisis, …}
(All these are kinds of problem.)
Part of the lexical set [[Emotion =
Negative]] as subject of ‘abate’
From BNC: {anxiety, fear, emotion, rage, anger, fury, pain,
agony, feelings,…}
From AP: {rage, anger, panic, animosity, concern, …}
A domain-specific norm:
[[Person | Action]] abate [[Nuisance]]
(DOMAIN: Law. Register: Jargon)
o undertake further measures to
us methods were contemplated to
s specified are insufficient to
as the inspector is striving to
t practicable means be taken to
ll equipment to prevent, and or
rmation alleging the failure to
t I would urge you at least to
way that the nuisance could be
otherwise the nuisance is to be
ion, or the local authority may
abate the odour, and in Attorney Ge
abate the odour from a maggot farm
abate the odour then in any further
abate the odour, no action will be
abate any existing odour nuisance,
abate odour pollution would probabl
abate a statutory nuisance without
abate the nuisance of bugles forthw
abated, but the decision is the dec
abated.They have full jurisdiction
abate the nuisance and do whatever
Lexical sets are contrastive in
• Different lexical sets generate different meanings.
• Lexical sets are not like syntactic structures.
• In principle, lexical sets are open-ended, but most have
high-value best examples.
• In practice, a lexical set may have only 1 or 2 members,
e.g. take a {look | glance}.
• No certainties in word meaning; only probabilities.
• … but probabilities can be measured.
A more complicated verb: ‘take’
• 61 phrasal verb patterns, e.g.
[[Person]] take [[Garment]] off
[[Plane]] take off
[[Human Group]] take [[Business]] over
• 105 light verb uses (with specific objects), e.g.
[[Event]] take place
[[Person]] take {photograph | photo | snaps | picture}
[[Person]] take {the plunge}
• 18 ‘heavy verb’ uses, e.g.
[[Person]] take [[PhysObj]] [Adv[Direction]]
• 13 adverbial patterns, e.g.
[[Person]] take [[TopType]] seriously
[[Human Group]] take [[Child]] {into care}
• TOTAL: 204, and growing (but slowly)
A fine distinction: ‘take + place’
• [[Event]] take {place}: A meeting took place.
• [[Person 1]] take {[[Person 2]]’s place}:
– George took Bill’s place; Bill left
and George took his place.
• [[Person]] take {[REFLDET] place}: Wilkinson took
his place among the greats of the game.
• [[Person=Competitor]] take {[ORDINAL] place}: The
Germans took first place.
Another fine distinction:
‘break + [[Bone]]’
• “John broke his leg”
• Whose leg?
• “John broke his nose”
• Whose nose?
We need to distinguish ‘reflexive determiners’
from other kinds of possessive determiners
Noun norms
• Norms for nouns are different in kind from norms
for verbs.
– Adjectives and prepositions are more like verbs.
• A different analytical apparatus is required for
• Prototype statements for each true noun can be
derived from a corpus.
• Examples for the noun ‘storm’ follow.
Storm (literal meaning) (1)
• Storms blow.
• Storms rage.
• Storms lash coastlines.
• Storms batter ships and places.
• Storms hit ships and places.
• Storms ravage coastlines and other places.
Storm (literal meaning) (2)
• Before it begins, a storm is brewing, gathering, or
• There is often a calm or a lull before a storm.
• Storms last for a certain period of time.
• Storms break.
• Storms abate.
• Storms subside.
• Storms pass.
Storm (literal meaning) (3)
• People can weather, survive, or ride (out) a storm.
• Ships and people may get caught in a storm.
Storm (literal meaning) (4)
• There are thunder storms, electrical storms, rain
storms, hail storms, snow storms, winter storms,
dust storms, sand storms, tropical storms…
• Storms are violent, severe, raging, howling,
terrible, disastrous, fearful, ferocious…
Storm (literal meaning) (5)
• Storms, especially snow storms, may be heavy.
An unexpected storm is a freak storm.
The centre of a storm is called the eye of the storm.
A major storm is remembered as the great storm (of
hurricanes, gales, and floods.
Pattern Dictionary and FrameNet
CPA investigates syntagmatic criteria for distinguishing different meanings of
polysemous words, in a “semantically shallow” way.
• expresses the deep semantics of situations (frames);
• proceeds frame by frame, not word by word;
• analyses situations in terms of frame elements;
• studies meaning differences and similarities between different words in a
• does not explicitly study meaning differences of polysemous words;
• does not analyse corpus data systematically, but goes fishing in corpora for
examples in support of hypotheses;
• has problems grouping words into frames, and misses some;
• has no established inventory of frames;
• has no criteria for completeness of a lexical entry.
Theoretical consequences and
practical applications (1)
• Anyone acquiring a language must learn competence in two
kinds of rule-governed linguistic behaviour:
– How to use words normally
– How to exploit the norms (creative metaphors, ellipsis, etc.)
• A pattern dictionary gives comparative frequency of patterns.
– A lexical syllabus could select only primary norms.
– “Primary norms” are a) high-frequency norms and b) concrete norms.
• In error analysis: what norm was aimed at?
– If learners are exploiting norms creatively, do you (the teacher) really
want them to?
Theoretical consequences
and practical applications (2)
For theoretical linguistics:
• Are some grammars better than others for representing how
words are used to make meanings?
‘S  NP VP’: a confusion of language with predicate logic?
• ARG3 (sometimes called ‘adjunct’, ‘adverbial’):
– CPA shows that a new grammar of adverbials is needed.
• Metaphor analysis:
– CPA distinguishes conventional metaphors from exploitations.
– The relationship between a possible ontology of words in use and
scientific conceptual ontologies such as WordNet.
Theoretical consequences and
practical applications (3)
• For computational linguistics and AI:
• Improving machine translation
– Getting the pattern right is more likely to select the right translation.
• Parsing and word-class tagging:
– CLAWS achieves ~90% accuracy in word-class tagging in BNC
– CPA reveals some systematic errors in CLAWS tagging.
• Anaphora resolution:
– He found a glass of water on the table and drank it.
– ‘[[Animate]] drink [[Liquid]]’ selects water as a direct object
• Meanings are associated with patterns – normal
contexts, rather than with words in isolation.
• Normal contexts correlate statistically significant
collocations in different clause roles.
• The whole language system is probabilistic and
• The probabilities can be analysed in a new kind of
dictionary – a syntagmatic pattern dictionary.

Norms and Exploitations