35 years of Cognitive Linguistics Session 4: Polysemy Martin Hilpert

35 years of Cognitive LinguisticsSession 4: Polysemy

Martin Hilpert

“normal” homonymy sign synonymy & polysemy

form form form form

meaning meaning meaning meaning

“normal” homonymy sign synonymy & polysemy

form form form form

meaning meaning meaning meaning

pants trousers

“normal” homonymy sign synonymy & polysemy

form form form form

meaning meaning meaning meaning

homonymy (Tuggy: ambiguity)

bank ‘sloping margin of a river’bank ‘financial institution’

race ‘competition’race ‘genetically defined population group’

polysemy summit ‘top of mountain’

‘meeting at the highest political level’

paper ‘writing material’ ‘document written on paper’

one form, two related meanings

ambiguity polysemy vagueness

form form form

meaning meaning meaning - - - - meaning meaning

e.g. bat e.g table e.g reporter

• two different words

• unrelated meanings

• one word

• related meanings

• one word

• highly similar meanings

ambiguity polysemy vagueness

form form form

meaning meaning meaning - - - - meaning meaning

e.g. bat e.g table e.g reporter

• two different words

• unrelated meanings

• one word

• related meanings

• one word

• highly similar meanings

ambiguity polysemy vagueness

form form form

meaning meaning meaning - - - - meaning meaning

e.g. bat e.g table e.g reporter

• two different words

• unrelated meanings

• one word

• related meanings

• one word

• highly similar meanings

tests for ambiguity

• the ‘logical’ test:– This is an X but not an X.– * This is a banana, not a banana. >> not

ambiguous– This is a bank, not a bank. >> ambiguous

tests for ambiguity

• the ‘definition’ test:– Is there a common meaning?– if yes >> not ambiguous– if no >> ambiguous

tests for ambiguity

• ‘linguistic’ tests:– X did Y and so did Z

• I have a bat and so does Bill.

– A did B and C D• I lost my wallet and Argentina the world cup final.

– if not zeugmatic >> not ambiguous– if zeugmatic >> ambiguous

tests for ambiguity

• puns of varying quality:– if a pun works >> ambiguous

problems with these tests

• clear yes-or-no answers only in clear cases– I painted a portrait and Bob the living room.– zeugmatic, funny, a little funny?

• sometimes in mutual conflict– I’m a painter, but not, you know, a painter.– logical test suggests ambiguity– common meaning suggests vagueness, not

ambiguity

same formtwo clearly distinct meaningsweakly related through a common meaning

summit‘mountain top’‘meeting of decision makers’

same formtwo distinct meaningsclearly related through a common meaning

string‘thin rope’‘part of a guitar’

same formvery similar meaningsobviously related through a common meaning

wing‘body part of a bird’‘part of an airplane’

same formbarely distinguishable meaningscommon meaning more accessible than the different meanings

grandmother‘mother’s mother’‘father’s mother’

advantages of the continuum view

• gradience is found in many other linguistic categories

• the traditional tests of ambiguity and vagueness correctly identify cases at the end points of the continuum

• in-between cases are naturally accounted for

How do we know where a word is on the continuum?

nosewave

sinceover

What is the basic meaning of over?Draw a picture!

The bird flew over the house.

John walked over the bridge.

The cathedral is over the bridge.

The paper is over the hole.

The dots are all over the carpet.

The fence fell over.

basic and extended senses• Polysemous words have a basic, central sense that is

– conceptually simple– frequent– learned early by children– conceptually connected to many other senses

• New senses emerge when a word is used in new contexts that invite a slightly different understanding of the situation

• The result of meaning extension is a ‘radial category’, where members share some properties, but not necessarily all of them

a strong hypothesis

• Each node in the network of a polysemous item corresponds to a representation in the speaker’s mental lexicon

a too strong hypothesis?

• Perhaps the network reflects linguistic analysis, and not what speakers actually think!

Sandra & Rice 1995

• How do ordinary speakers distinguish between different senses of the same word?

• Sorting task:– 29 participants received cards with 20

sentences each for the prepositions at, on, in– Task: Sort the cards according to how the

prepositions are used, make as many or as few groups as you like

I can say ‘How are you’ in Italian.Make sure to get that in writing.I saw him in my dreams.There is a hole in your sweater.Are you putting onions in the stew?Don’t put that in your mouth.My pen is in the drawer.I looked him straight in the eyes.In Japan, they eat raw fish.Please come in.What would you have done in my place?What line of work are you in?

sorting task

• How often are two particular sentences of the 20 put into the same group?– Don’t put that in your mouth.– My pen is in the drawer.– In Japan they eat raw fish.

Often together)Rarely together)

spatial

time

abstract

Sandra & Rice 1995

• Language users are clearly sensitive to the different meanings of polysemous items.

• The sorting results show that speakers make fine-grained distinctions

• However, it is not clear whether these distinctions are made because the task requires it, or because they are already there in the mental lexicon.

meaning varies with context

• The meaning of words is tough to analyze.– predicative construction (X is tough)– followed by a to-infinitive– the subject is inanimate– …

• Tough guys never dance.– attributive construction (a tough X)– followed by a verb phrase– the subject is animate– …

difficult

rough

Gries (2006): the many meanings of run

– run a mile– run for office– run a small business– you can run, but you cannot hide– ooh you make my motor run, my motor run– …

behavioral profile analysis

1. extract all examples of a word from a corpus 2. annotate each example for its rough meaning (e.g.

run a mile >> fast pedestrian motion)3. identify contextual features

– collocates– morpho-syntactic properties of the word– semantic features of surrounding elements– …

4. annotate all examples for all of these features5. apply a statistical test that investigates correlations

between word sense and contextual features

features of run

1. morphological – past vs. present, simple vs. progressive aspect, active vs.

passive2. syntactic

– intransitive vs. transitive vs. complex transitive– declarative vs. interrogative vs. imperative– main clause vs. subordinate clause

3. semantic characteristics – subject: human vs. animate vs. inanimate– object: countable vs. mass, concrete vs. abstract

4. collocates – Left 1, Left 2, Right 1, Right 2

• …

an annotated example

• My friend Bob runs a model agency.– sense: manage– present, simple, active– transitive, declarative, main clause– human subject, abstract countable object– L2: friend, L2: Bob, R1: a, R2: model– …

• 800 examples with this kind of information

senses have structural profiles

• manage– by default transitive (100%)– subject often an organization (55%)– often past participle (43%)– often followed by a preposition (34%) – …

• fast pedestrian motion– almost never transitive (7%)– subject never an organization (0%)– rarely past participle (12%)– sometimes followed by a preposition (25%) – …

----------------------- features ------------------------

----

----

-sen

ses-

----

--

transitive human subject followed prep subordinate clausefastpedestrianmotion 0.06 0.82 0.67 0.30escape 0.69 0.86 0.72 0.45function 0.39 0.27 0.06 0.51manage 0.21 0.29 0.08 0.35meet 0.97 0.29 0.5 0.51

summing up

ambiguity polysemy vagueness

form form form

meaning meaning meaning - - - - meaning meaning

e.g. bat e.g table e.g reporter

• two different words

• unrelated meanings

• one word

• related meanings

• one word

• highly similar meanings

polysemy

• the presence of various related meanings associated with a single linguistic unit

• central and extended senses– central: conceptually simple, first learned, frequent, highly

connected to other senses

• linguists’ analyses of polysemy should correspond to speakers’ mental representations!– psycholinguistic experiments – corpus-based behavioral profile analyses

next time: conceptual blending

See you next time!

martin.hilpert@unine.ch