Displayed are triplets annotated with human referential utterances. A neural listener, trained with chairs, is used to color-code attention-wise the utterances and score each object. Underlined tokens where out-of-the-vocabulary and were ignored by the listener.
| Distractor-A | Distractor-B | Target |
|
|
|
|
Human Speaker: lame with rectangular base Correctly guessed: True Confidences: 0.05, 0.07, 0.88 |
||
| Distractor-A | Distractor-B | Target |
|
|
|
|
|
Human Speaker:
rectangular table lamp with rectangular shade Correctly guessed: True Confidences: 0.00, 0.00, 0.99 |
||
| Distractor-A | Distractor-B | Target |
|
|
|
|
Human Speaker: modern stand that alternates feet Correctly guessed: False Confidences: 0.69, 0.13, 0.18 |
||
| Distractor-A | Distractor-B | Target |
|
|
|
|
|
Human Speaker: irregular cube on two legs Correctly guessed: True Confidences: 0.00, 0.00, 0.99 |
||
| Distractor-A | Distractor-B | Target |
|
|
|
|
Human Speaker: boxy , box shade , table lamp Correctly guessed: True Confidences: 0.01, 0.01, 0.98 |
||
| Distractor-A | Distractor-B | Target |
|
|
|
|
|
Human Speaker: rectangle cube with trapezoid on top Correctly guessed: True Confidences: 0.02, 0.01, 0.97 |
||
| Distractor-A | Distractor-B | Target |
|
|
|
|
Human Speaker: cone shaped shade Correctly guessed: True Confidences: 0.03, 0.04, 0.93 |
||
| Distractor-A | Distractor-B | Target |
|
|
|
|
|
Human Speaker: traditional table lamp with solid base Correctly guessed: True Confidences: 0.00, 0.02, 0.98 |
||
| Distractor-A | Distractor-B | Target |
|
|
|
|
Human Speaker: tall and rectangular Correctly guessed: True Confidences: 0.01, 0.01, 0.98 |
||
| Distractor-A | Distractor-B | Target |
|
|
|
|
|
Human Speaker: tall rectangular tower Correctly guessed: True Confidences: 0.02, 0.02, 0.96 |
||