March 31, 2005

Absolutely is totally not definitely

I'm a big fan of dictionaries, but sometimes they let you down badly. Here's an example: a systematic difference between absolutely and definitely that is missed by all the English dictionaries I've checked.

According to the American Heritage Dictionary, absolutely means "Definitely and and completely; unquestionably". The AHD doesn't give definitely a separate entry, but (the relevant sense of) definite is given as as "Indisputable; certain". Merriam-Webster's unabridged (3rd edition) says that definitely means "distinctly, unmistakably, positively" while absolutely means "independently, unconditionally, entirely, positively". The OED gives "Without doubt or condition" as the basic (relevant) sense of absolutely, and "In a definite manner; determinately, precisely" as the meaning of definitely. Encarta defines absolutely as "totally", and definitely as "certainly; finally and unchangeably; exactly; clearly; absolutely".

I don't see any way predict from these entries that writers on the web are 198 times more likely to "absolutely adore" something than to "definitely adore" it, while they are 97 times more likely to "definitely prefer" something than to "absolutely prefer" it. Details by Google (in all the tables in this post, the words heading the colums precede the words or word sequences labelling the rows: thus the first entry in the second row gives 289,000 as the count returned for the query {"absolutely adore|adores|adored"}):

  absolutely definitely absolutely/definitely ratio
overall 34.9M 29.3M  
adore|adores|adored 289K 1.46K
198/1
love|loves|loved 905K 51K
18/1
like|likes|liked 16.2K 158K
1/10
prefer|prefers|preferred 644 62.6K
1/97

The fact that love and like grade in between adore and prefer suggests that these are not random lexical associations, but rather reflect some sytematic difference in the emotional or attitudinal loading of the terms. This idea is confirmed if we look at similar counts and ratios for verbs expressing negative attitudes. Netizens are 179 times more likely to "absolutely abhor" something than to "definitely abhor" it, while they are 3 times more likely to "definitely dislike" something than to "absolutely dislike" it:

  absolutely definitely absolutely/definitely ratio
overall 34.9M 29.3M  
abhor|abhors|abhorred 7.15K 40
179/1
loathe|loathes|loathed 14.7K 91
162/1
despise|despises|despised 21.3K 242
88/1
hate|hates|hated 185K 5.04K
38/1
detest|detests|detested 14K 370
38/1
can't stand 4.38K 305
14/1

don't like

2.08K 6.25k
1/3
dislike|dislikes|disliked 603 1.7K
1/3

There's a similar effect for verbs where the emotion is associated with a proposition under consideration or a contemplated outcome:

  absolutely definitely absolutely/definitely ratio
overall 34.9M 29.3M  
insist|insists|insisted 45.5K 1.36K
33/1
believe|believes|believed 40.8K 38.4K
1/1
suspect|suspects|suspected 200 968
1/5
think|thinks|thought 16.4K 215K
1/13
feel|feels|felt that 12K 234K
1/20
suggest|suggests|suggested 1.25K 29.1K
1/23

The interaction with modals is complex, because of the different meanings of the modals and the presence or absence of negation, but I think you can see some related effects emerging from the great variation of ratios in this small sample of relevant comparisons:

  absolutely definitely absolutely/definitely ratio
overall 34.9M 29.3M  
must be 37.7K 4.34K
9/1
must not 7.15K 654
11/1
must be/not ratio
5/1
7/1
 
might be 61 806
1/13
might not 64 149
1/2
might be/not ratio
1/1
5/1
 
should be 6.29K 30.6K
1/5
should not 8.44K 22.6K
1/3
should be/not ratio
1/1.3
1.4/1
 
could be 869 14.1K
1/16
could not 21.4K 6.89K
3/1
could be/not ratio
1/25
2/1
 
may be 113 1.15K
1/10
may not 38.7K 447
87/1
may be/not ratio
1/342
3/1
 

This all has something to do with what absolutely and definitely mean, and how their meaning interacts with the meanings of other words. But this dimension of meaning seems to be completely (though perhaps not absolutely or definitely?) missing from the dictionary definitions.

[Update: John Lawler emailed a proposal for a solution:

It's long been a standard example in GSish semantics that absolute(ly) subcategorizes to modify only polar attributes, adjectives (for example) that stand at one end or another of a local semantic cline. Thus the asterisk distribution in absolutely frozen/freezing/*cold/*cool/*lukewarm/*hot/boiling/steaming/burning.

The killer example is absolutely mad; mad has two senses, one synonymous with angry, and one synonymous with insane. So He's mad is ambiguous. But He's absolutely mad isn't, since only the insane sense of mad is polar, and is selected by absolute(ly).

This fact about absolute(ly) pretty much distinguishes the data in first 3 tables; definitely outranks absolutely with non-polar predicates, and absolutely greatly outranks definitely with polar predicates (or with believe, which seems to be a predicate that swings different ways in the reality-based and faith-based communities).

As for the modals, I haven't done an extensive analysis, but by eyeball, I'd guess Square/L modals and negative Diamond/M modals would be treated as polar and would prefer absolute(ly). And that Diamond/M modals and negative Square/L modals would prefer definitely.

When John says "GSish" he's abbreviating "generative semantics-ish", I think. To make a long story short, what that means in this case is that he's expressing an aspect of semantics (what things mean) in an essentially syntactic way (in terms of constraints on local structures). In particular, he's saying that absolutely "subcategorizes for" polar adjectives, in the same way that try "subcategorizes for" an infinitive in phrases like "try to go". I'm not sure whether there is a story here about definitely, or whether it's hypothesized to be just operating at baseline rates in the polar cases.

I'm not entirely satisfied by this sort of account -- it seems a bit like saying that extract of poppy makes you sleepy because it has a dormative principle, or that a child who gets a rash in the winter has hibernal eczema. Still, even that much would be better than what the dictionaries now give us for this sort of phenomenon, which is nothing. John was not surprised that dictionaries don't discuss this stuff (as he put it: "Well, yes. Duh. Wudja expect?"). Actually, I'd expect the dictionaries to provide a description (i.e. "absolutely modifies polar adjectives", or something of the kind), and linguists to provide an explanation. ]

Posted by Mark Liberman at March 31, 2005 06:44 AM