Idioms and exceptionality Nik Gisborne and Dick Hudson

Idioms and exceptionality Nik Gisborne and Dick Hudson LAGB Leeds September 2010

Idioms are exceptional E. g. kick the bucket • exception to general compositionality: – it means ‘die’, not ‘kick the bucket’ • exception to general syntax: – no passive: *The bucket was kicked. – no tough movement: *The bucket was hard to kick – etc.

Our questions • Why are such exceptions possible? – Does default inheritance help? • How are idioms stored in relation to their constituent lexemes? – Do the ‘sub-lexemes’ of Word Grammar help? • How are idioms organized syntactically? – Does dependency structure help? – Do we need phrases?

Kinds of idiom Nunberg, Sag and Wasow 1994 distinguish: • ‘idiomatic phrases’(Id. P) – e. g. kick the bucket – rigid syntax • ‘idiomatically combining expressions’ (Id. CE) – e. g. bury the hatchet – some syntactic freedom • e. g. The hatchet was buried.

More recent research in linguistics Nunberg at al’s contrast has been explored: • George Horn (2003): Id. CEs are syntactically regular if the parts have regular theta roles. • Espinal & Mateu (2010): Id. P vs Id. CE is too rigid, e. g. laugh ones head off is part Id. P, part Id. CE. • Jackendoff (1997, 2008): accepts Id. P vs Id. CE. – Influential player. – Suggests a formal analysis.

Id. P or Id. CE? • Is it likely that there are just two kinds of idiom? • Maybe there are degrees of opacity? – most opaque, e. g. kick the bucket – less opaque, e. g. bury the hatchet – least opaque, e. g. laugh ones head off • But how to measure opacity? – Does a network analysis help?

Jackendoff: an Id. P Doesn’t contribute at all to meaning. NB special role of head word!

Jackendoff: An Id. CE NB This is the only link between bury and the hatchet.

Jackendoff’s analysis • Id. P: totally rigid syntax – but: kicked the proverbial bucket • Id. CE: totally free syntactic order – but: *They found the hatchet then buried it. – and excludes: buried the proverbial hatchet • Id. CE: requires meaning: syntax = 1: 1 – e. g. bury (‘reconcile’) the hatchet (‘a disagreement’) ‘Metaphorical semantic composition’

Metaphorical semantic composition • Isn’t sufficient for Id. CE – raise (‘cause’) hell (‘disturbance’) isn’t an Id. CE (Postal). • Isn’t necessary for Id. CE His example! – let (‘reveal’) the cat (‘the secret’) out of the bag (? ? ) • Isn’t necessary for literal meaning – do (‘cartwheel’) a cartwheel (‘cartwheel’)

Research in psycholinguistics (1) • How does activation affect idioms? • How are idioms represented? – NOT as single words – But as phrases with a single entry • ‘The hybrid theory’ – Cutting and Bock (1997) – Superlemma theory (Sprenger et al 2006)

Cutting and Bock (1997) Syntax is independent of words.

Superlemmas Shows syntactic relations among parts. (But how? ) NB model of activation, not structure.

Research in psycholinguistics (2) • We access conceptual metaphors in idioms. – e. g. ‘anger is heat’ for blow one’s stack , but not jump down someone’s throat – Gibbs, Bogdanovich, Sykes and Barr (1998) • We process idiom syntax normally. – Peterson, Burgess, Dell and Eberhard (2001) • Literal word meanings become active during idiom production. – Sprenger, Levelt and Kempen (2006).

To summarise, … • • Idioms have a single entry in memory. They contain ordinary lexemes. They involve ordinary metaphor. They have ordinary syntax. – But abnormal linkage to meaning – So syntax may be abnormally limited.

Word Grammar 1984 1990 2007 2010 Part 1: How the mind works Part 2: How language works Part 3: How English works 2010

What Word Grammar offers • Default inheritance – allows exceptions • Sub-lexemes – allows partial differences within a lexeme • Dependency structure – allows words to relate directly • Network structure – explains spreading activation and relatedness

For example: tall man height tall man > 1. 75 m referent height > typical height sense TALL 1. 75 m man sense ‘isa’ MAN stored type/lexeme dependent tall man token sense

Default inheritance allows exceptions I. e. instances may have exceptional properties • e. g. tall overrides the default height. • Typically, a dependent enriches the head’s sense. – and may override default properties. – Any property can be overridden. • e. g. fake diamonds just look like diamonds

So exceptionality ranges … • from zero – kick a ball [cf morphology: walked] • through partial – kick up a fuss [cf vowel-change: ran] • to total – kick the bucket [cf suppletion: went]

Theoretical point • Default inheritance is different from unification. – Unification is blocked by conflict. • But default inheritance is widely accepted in AI models of cognition. • And it explains the prototype effects found by psychologists.

Sub-lexemes • Lexemes are in a conceptual taxonomy: – e. g. TAKE isa full verb isa word • Each word token isa some lexeme. • So ‘sub-lexeme isa lexeme’ is permitted – e. g. TAKE/off isa TAKE isa verb … – like TAKE: TAKE/off inflects to took – unlike TAKE: TAKE/off is intransitive and …

Sub-lexemes in idioms • KICK/bucket isa KICK • Syntax: – like KICK: it needs an object – unlike KICK: this must be THE/bucket • whose complement must be BUCKET/the • Semantics: – unlike KICK, its sense is ‘die’.

pace Jackendoff … who rejects this kind of analysis: • 1997: – “a notational variant of listing a lexical VP” – “clumsy” … “collapses under its own weight” • 2008: – “no non-theory-internal reason to concentrate all the meaning in one morpheme” • but his own analysis locates meaning on head!

Jackendoff: an Id. P NB special role of head word!

NB ‘die’ is separated from ‘kick’ by many links. NB direct link from ‘create a disturbance’ to ‘kick’ Two idioms in Word Grammar die sense kick sense KICK create dust sense create a disturbance c KICK/up • o c NB link from ‘fuss’ to ‘disturbance’ (not o KICK/bucket • • shown) KICK/fuss THE BUCKET UP sense c • • A/fuss FUSS/a

Id. Ps and Id. CEs in WG • • • Id. Ps and Id. CEs use ordinary syntax. Their head words have exceptional senses. The network shows how close the idiomatic sense is to the literal sense. – • So there’s no need for any other Id. P/Id. CE contrast. We speakers can vary the syntax as we want. – But there’s no point in varying it if the parts are unrelated to the idiomatic meaning.

Conclusion • The Id. P/Id. CE contrast has no theoretical status. • The range of possibilities found is as expected given: – default inheritance – sub-lexemes for syntactic and semantic detail – dependents as semantic modifiers – network structure

Thank you • This slide show can be found at www. phon. ucl. ac. uk/home/dick/talks. htm