Keyboard shortcuts

/ or ⌘/Ctrl K
Find a note
j / k
Next / previous section or linked note
h / l
Collapse or go to parent / expand or enter
e or Alt-click
Read a linked note here
o
Open focused note on its own
g g / G
First / last section or linked note
g h / g a
Home / all notes
g b / g t
Backlinks / table of contents
t
Cycle System, Light, Dark
? / Esc
Show / close this reference

Search: ↑/↓ or Ctrl N/P, Enter to open. Shortcuts pause while typing.

Pavlick and Kwiatkowski: Inherent Disagreements in Human Textual Inferences [48d6ac77]

Tags: nlu, papers

1. Summary

  • Talks about how human annotations and disagreements from those are not derived from statistical noise
  • More context does not necessarily mean more agreement, once the input reaches sentence/passage, disagreements hold relatively steady
  • We assume what people write is what they mean, but actually there's a pretty big gap between what people mean <-> what they write <-> what they interpret <-> what they infer. NLP models currently only model the 2nd and 3rd arrows.
  • punted on precise definitions of "real world" and rather tried to have their models approximate "what humans do"

file:2021-03-10_09-29-33_screenshot.png

2. Annotations

  • Uses a combination of RTE2, SNLI, MNLI, JOCI, and DNC

    • RTE2 - premises/hypothesis combos
    • SNLI - premises from image captions, hypoteshis from existing NLI dataets paired with hypotehsis that were automatically generated
    • MNLI - Same as SNLI but with a range of text genres
    • JOCI - "commen sense" inferences
    • DNC - mostly naturally occuring premises paired with template generated hypothesis

file:2021-03-10_09-30-06_screenshot.png

2.1. Preprocessing

  • They had 500 workers complete and rank the response on a slider

2021-03-10_09-34-24_screenshot.png

  • Continous scale not always the best, they had to do some z-score normalization in order for all the data to make sense

3. Analysis

  • Ultimately wanted to judge how much "noise" exists in the annotation process
  • If there is a single truth, then the the "noise" should be generalizable with a single gaussian distribution
  • If there are "multiple" truths, then a gaussian mixture model should be correct

  • Assumption: if a single truth exists, the GMM woul be the exact same as the gaussian

    2021-03-10_09-39-16_screenshot.png

  • Model chooses to fit towards the GMM than the gaussian
  • Example: is the word "swat" forceful? Is "confess that" factive?

  • In the first section, they note that the annotations can be modeled by the GMM, implying that humans believe there are multiple "truths"

    • NLP models currently only believe there is one truth to model

3.1. Context

  • Sampled sentences from wikipedia, and considered each sentence to be a premise, and generated a hyptothesis by replacing the corresponding a word from the premise with a substiute word, where the substitue word is either a hypernym/hyponym, antonym, or co-hyponym

  • Collected ratings at 3 levels

    • word
    • sentence
    • passage
  • Disagreements among raters actually increase when more context is shown
  • Definitely some confounds abound

4. Model Predictions

  • Is this a problem at all?

    • what if the underlying distributions already reflect the distributions observed in human judgements, and the models already adaquetely capture that with softmax?

  • since nli is usually treated as a classification, they discretize (after experimenting with z-normalied human scores) by mapping into different bins

    • entailment~/~contradiction~/~neutral label bins
  • They used a pretrained bert and fine tuned it on the labels

  • Attempted to see how well bert captured the underlying multi-modal distribution with a softmax

    • found that the softmax is a poor approximation

      2021-03-10_09-55-29_screenshot.png