Good Nouns, Bad Nouns

1
Good Nouns, Bad Nouns

• What the corpus says
• and what native speakers think
• Philip Dilts, University of Alberta Department of
  Linguistics
• AACL 2008, Brigham Young University, Provo, UT
• March 14, 2008

2
Overview

• Semantic Orientation
• Psychological measure of how good/bad a word is
• Semantic Preference
• Corpus-based measure of collocational preference
• How Semantic Orientation and Preference relate
• The wrong story (abstract version)
• The right story (maybe?)
• How to get the right story

3
Semantic Orientation

• Term from Hatzivassiloglou McKeown (1997)
• How good or bad people feel about a word

4
Semantic Orientation

• fun

5
Semantic Orientation

• jail

6
Semantic Orientation

• How good or bad people feel about a word

• Out of context
• Psychologically real
• Osgood et al (1957) Evaluative Factor
• Experimentally determined
• E.g., ANEW (Bradley Lang 1999)
• rate how you felt while reading each word
  Happy vs. Unhappy

7
ANEW Questionnaire
8
Semantic Preference

• Collocational preference for a semantic category
• See, e.g, Partington (2004)
• The focus here
• Preference for good/bad adjectives in
  adjective-noun sequences

9
Semantic Preference
thoughts thoughts thoughts thoughts thoughts thoug
hts thoughts
10
Semantic Preference
i.e. nouns that collocate with many adjectives of
one orientation
thoughts thoughts thoughts thoughts thoughts thoug
hts thoughts
sad cruel morbid cold dirty
11
Semantic Preference
i.e. nouns that collocate with many adjectives of
one orientation
thoughts thoughts thoughts thoughts thoughts thoug
hts thoughts
sad cruel morbid cold dirty
sweet soothing
...and relatively few adjectives of the other
orientation
12
Semantic Preference

• In this case, we would say thoughts has a
  negative semantic preference
• How closely do preference and orientation
  correspond?
• Do good nouns prefer to collocate with good
  adjectives?
• Can good nouns prefer to collocate with bad
  adjectives?

13
Does Preference Match Orientation?

• Dilts Newman (2006)
• Start with adjectives rated as good/bad
• Pairs of adjectives from Osgood et al. (1957)
• Find nouns that follow many good or bad
  adjectives in a corpus (BNC)
• Approximately, nouns that are consistently
  modified by good or bad evaluations

14
Does Preference Match Orientation?

• Surprising results
• joke, grin, jokes, beast, thoughts, mistakes,
  laughter, head, shoulder, dreams
• 10 strongest negative preferences in the BNC
• Are these words consistently judged as negative
  by native speakers?

15
Does Preference Match Orientation?

• Pilot orientation study (Dilts Newman 2008)
• Participants 5 native English speakers
• Stimuli - 121 nouns
• 25 nouns with most positive preference in BNC
• 25 nouns with most negative preference in BNC
• 71 filler nouns (strong preference by other
  methods)
• Method
• ask participants to rate each noun as
  good/neutral/bad

16
Does Preference Match Orientation?
Positive
Collocational preference
Negative
Bad
Speaker ratings
Good
17
Does Preference Match Orientation?
community
setting
range
programme
Positive
family
conversation
assessment
development
support
source
alternative
response
home
Collocational preference
cough
delusion
logic
dreams
beast
conditions
truths
joke
mistakes
moods
Negative
legacy
grin
bastard
lump
jokes
gods
injuries
bruise
moment
laughter
blows
killer
Bad
Speaker ratings
Good
18
Does Preference Match Orientation?
Reinforcement
community
setting
range
programme
Positive
family
conversation
assessment
development
support
source
alternative
response
home
Collocational preference
cough
delusion
logic
dreams
beast
conditions
truths
joke
mistakes
moods
Negative
legacy
grin
bastard
lump
jokes
gods
injuries
bruise
moment
laughter
blows
killer
Bad
Speaker ratings
Good
19
Does Preference Match Orientation?
community
setting
range
programme
Positive
family
conversation
assessment
development
support
source
alternative
response
home
transformation
Collocational preference
cough
delusion
logic
dreams
beast
conditions
truths
joke
mistakes
moods
Negative
legacy
grin
bastard
lump
jokes
gods
injuries
bruise
moment
laughter
blows
killer
Bad
Speaker ratings
Good
20
Does Preference Match Orientation?
community
setting
range
programme
Positive
family
conversation
assessment
development
support
source
alternative
transformation
response
home
Collocational preference
cough
delusion
logic
dreams
beast
conditions
truths
joke
mistakes
moods
Negative
legacy
grin
bastard
lump
jokes
gods
injuries
bruise
moment
laughter
blows
killer
Bad
Speaker ratings
Good
21
Tentative Conclusions

• Nouns that speakers rate as good collocate with
  good adjectives, and vice versa
• i.e., nouns often collocate with adjectives that
  reinforce their orientation
• Many good nouns attract bad adjectives
• Good nouns collocate with adjectives that
  transform their orientation
• BUT few bad nouns attract good adjectives
• Bad nouns dont collocate (much) with adjectives
  that transform their orientation

22
Orientation DOES NOT Match Preference

• Is this the right story?
• We looked at only 50 nouns
• with extreme collocational preferences
• Noun orientation ratings came from a small,
  biased sample
• 5 linguistics grad students
• How can we get a better story?
• Better measure of orientation

23
ANEW
24
ANEW

• 1034 words rated for pleasure
• Do they agree with our raters?

25
ANEW ratings vs. Collocational Preference

• Did our participants agree with ANEW?

26
ANEW ratings vs. Collocational Preference

• Did our participants agree with ANEW?

27
ANEW ratings vs. Collocational Preference

• Did our participants agree with ANEW?

transformation
28
ANEW ratings vs. Collocational Preference
community
setting
range
programme
Positive
family
conversation
assessment
development
support
source
alternative
transformation
response
home
Collocational preference
cough
delusion
logic
dreams
beast
conditions
truths
joke
mistakes
moods
Negative
legacy
grin
bastard
lump
jokes
gods
injuries
bruise
moment
laughter
blows
killer
Bad
Speaker ratings
Good
29
ANEW ratings vs. Collocational Preference

• but only 10 of our 50 nouns were in ANEW
• For Dilts Newman 2006, we established
  collocational preferences for gt30,000 nouns
• More than 700 were rated in ANEW
• Does the mismatch hold for these nouns too?

30
ANEW ratings vs. Collocational Preference

• 10 nouns from Pilot study

31
ANEW ratings vs. Collocational Preference

• 734 nouns from Dilts Newman (2006)

32
ANEW ratings vs. Collocational Preference
33
ANEW ratings vs. Collocational Preference
34
ANEW ratings vs. Collocational Preference

• Everything is drawn upwards
• Nouns tend to collocate with positive adjectives

35
ANEW ratings vs. Collocational Preference

• Many bad nouns attracting good adjectives

More transformation here
36
ANEW ratings vs. Collocational Preference

• Opposite to the previous mismatch!

community
setting
range
programme
Positive
family
conversation
assessment
development
support
More transformation here
source
alternative
Collocational preference
response
home
cough
delusion
logic
dreams
beast
conditions
truths
joke
mistakes
moods
Negative
legacy
grin
bastard
lump
jokes
gods
injuries
bruise
moment
laughter
blows
killer
Bad
Speaker ratings
Good
37
ANEW ratings vs. Collocational Preference

• Focusing on the extreme values selected outliers
• Negative preference nouns had less extreme values
• e.g. minimum preference was -4, maximum
  preference was 8

38
Conclusions

• The right story
• General tendency for positive collocational
  preference
• Noun orientations are more often reinforced by
  collocating adjectives than transformed
• More bad nouns than good nouns are transformed by
  collocating adjectives

39
Future Work

• Is this the right story?
• Preference based on adjective pairs
• Negative adjectives were usually lower frequency
• e.g., pure (3301) vs. impure (68) (BNC)
• We controlled for low frequency, but not 0
  frequency
• ie. nouns that collocated with pure but not
  impure got full positive preference scores for
  that pair
• But nouns are more likely to collocate only with
  pure because of its higher frequency

40
Future Work

• How to get the right story
• Select pairs of good/bad adjectives
• ANEW? MPQA?
• Matched for frequency
• Matched for goodness/badness
• Find nouns with strong collocational preferences
  for these adjectives
• Use Collostructional Analysis (Stefanowitsch
  Gries 2003)
• Profit

41
References

• Bradley, M.M., Lang, P.J. (1999). Affective
  norms for English words (ANEW) Stimuli,
  instruction manual and affective ratings.
  Technical report C-1, Gainesville, FL. The Center
  for Research in Psychophysiology, University of
  Florida.
• Dilts, P. J. Newman. 2008. Good Nouns, Bad
  Nouns, and the Company they Keep. Presentation
  at the 2008 Annual meeting of the LSA, Chicago,
  IL.
• Dilts, P. J. Newman. 2006. A Note on
  Quantifying 'good' and 'bad' prosodies. Corpus
  Linguistics and Linguistic Theory 2(2)233-242.
• Osgood, C. E., G.J. Suci P.H. Tannenbaum. 1957.
  The Measurement of Meaning. Urbana University of
  Illinois Press.
• Partington, A. 2004. Utterly Content in Each
  Others Company. International Journal of Corpus
  Linguistics, 9131-156
• Stefanowitsch, A. S.Th. Gries. 2003.
  Collostructions. International Journal of Corpus
  Linguistics, 8209-243

42
Thanks!

• Members of the University of Alberta Department
  of Linguistics
• You!