Challenges in finding metaphorical connections Katy Gero and Lydia - - PowerPoint PPT Presentation

▶
challenges in finding metaphorical connections
SMART_READER_LITE
LIVE PREVIEW

Challenges in finding metaphorical connections Katy Gero and Lydia - - PowerPoint PPT Presentation

Jul 26, 2023 247 likes •435 views

Challenges in finding metaphorical connections Katy Gero and Lydia Chilton C OLUMBIA U NIVERSITY NAACL Workshop on Figurative Language June 6, 2018 1 Example human written poem for: anger is wood the anger grew , like a tree this large,

slide-1
SLIDE 1

Challenges in finding metaphorical connections

Katy Gero and Lydia Chilton COLUMBIA UNIVERSITY NAACL Workshop on Figurative Language June 6, 2018

  • 1
slide-2
SLIDE 2

the anger grew, like a tree
 this large, immovable object had taken root 
 casting shade on even the happiest parts of my life 
 I could let it consume me, or cut it down

Example human written poem for: anger is wood

  • 2
slide-3
SLIDE 3

Poetry requires a conceptual message.

“…meaningfulness [in computer generated poetry] is not always explicitly considered and is often only softly satisfied.” Oliveira (2017) We are interested in the content of poetry generation.

3

Oliveira, Hugo Gonçalo. "A survey on intelligent poetry generation: Languages, features, techniques, reutilisation and evaluation." Proceedings of the 10th International Conference on Natural Language Generation. 2017.

slide-4
SLIDE 4

early dew the water contains teaspoons of honey

Netzer, Yael, et al. "Gaiku: Generating haiku with word associations norms." Proceedings of the Workshop on Computational Approaches to Linguistic Creativity. 2009.

Computer generated poems:

4

slide-5
SLIDE 5

Can you remember when it started raining! I had a stomach full of blood and sweat, The pins an arrow through the barrel aging, Nothing like a pile of hot and wet.

Ghazvininejad, Marjan, et al. "Generating topical poetry." Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing. 2016.

Computer generated poems:

5

slide-6
SLIDE 6

Why care about intentional conceptual messages?

  • 1. Helps generate meaning for longer texts.
  • 2. Easier to evaluate than “meaning exists’’.
  • 3. Can demonstrate improvement.

6

slide-7
SLIDE 7

How do we make sure poems have intentional meaning?

7

slide-8
SLIDE 8

We introduce a poetry writing task:

Instead of a topic, let’s use a metaphor as the prompt.

anger is wood

We evaluate if the meaning of the poem is semantically consistent with the metaphor. Ensures intention.

8

slide-9
SLIDE 9

god is a breath compassion is blood death is a rose surrender is a book anger is wood peace is a rock immortality is a room hate is a mist grace is a garden hope is a ship

Randomly generate 10 prompts.

Concrete Nouns bed horse bell book ship wing room mouth storm town silver stream dust color side state ear sand grass wood rose blood girl ring wine garden brain wave mist dawn breath spring nation finger hair rock breast window snow body ground stone flame shadow line path king darkness Poetic Themes loss confusion faith freedom grace hate jealousy spring unity love consciousness soul melancholy calmness death fear friendship anger gratitude hope joy nature religion sadness suffering vanity happiness surrender compassion envy forgiveness god grief immortality life peace remembrance silence spirituality truth war bitterness violence

9

Gagliano, Andrea, et al. “Intersecting Word Vectors to Take Figurative Language to New Heights." Proceedings of the Fifth Workshop on Computational Linguistics for Literature. 2016.

slide-10
SLIDE 10

Why have people do this task?

  • 1. We don’t know how hard this task is.
  • 2. People give strategies and failure points.
  • 3. Show us if and where support is needed.

10

slide-11
SLIDE 11

Experiment on Mechanical Turk

11

slide-12
SLIDE 12

Evaluation of 200 short poems

  • 14 were plagiarized and removed from dataset.
  • Two evaluators read all poems.

12

97% agreement on which were successful. 24% were found to be successful.

slide-13
SLIDE 13

24% of poems were successful.

hate is a mist

He spits at me with fiery tongues, his fists with betrayal, his eyes with loathing. The love we knew turned to thick, toxic vapor, and now we sit in its mist.

hate mist poem death rose poem

13

death is a rose

Alone I cried, my tears went unseen. Within I died no shoulder to lean. God tricked my life and closed the doors, I wish my death to be as soft as a rose.

slide-14
SLIDE 14

Evaluation of unsuccessful poems.

  • Used Grounded Theory to develop categories.
  • Placed unsuccessful poems into categories.

14

75% agreement on categorization. 5 different categories.

slide-15
SLIDE 15

Categories of unsuccessful poems:

anger wood poem

7% off-topic

mothers

poem anger wood poem

  • r

41% no connection

anger wood poem

17% attributional “the wood is angry”

anger wood poem fire

15% offset “anger is fire”

anger wood poem

26% incoherent
 “anger is drying like wood”

15

slide-16
SLIDE 16

Reasons for ‘other’ connections:

anger wood poem

17% attributional “the wood is angry”

anger wood poem fire

15% offset “anger is fire”

anger wood poem

26% incoherent
 “anger is drying like wood”

16

People want to relate the two words, but when difficult they back-off to related prompts.

slide-17
SLIDE 17

What to work on next

  • 1. Continue to use this prompt to address

intentional meaning in poetry.

  • 2. Support people doing this task:
  • Generate suggestions
  • Give feedback with detection

17

Dataset is available online: github.com/kgero/metaphorical-connections

slide-18
SLIDE 18

Summary

  • To ensure meaningfulness in poetry, we introduce a

poetry writing task that uses a metaphorical prompt.

  • We ran a study in which people wrote poems on 10

randomly generated prompts.

  • We show high agreement in evaluation.
  • We propose computational methods to support

people in task.

Questions?

18