Assertions¶
The table below shows the 49 assertions usable by AssertionAnnotator and ClassificationPipeline.
Each assertion has a corresponding SetFit classifier on HuggingFace.
The labeled training dataset is also available on HuggingFace at StanfordSCALE/assertions_llm_annotated_talkmoves.
Model performance varies
All available assertions are listed here, but some models have poor performance, poor inter-rater reliability, or low support and therefore unreliable evaluation metrics. Consider assertion performance before using annotators to make claims about your data.
| Assertion | Model | Test set support | Krippendorff's alpha | Test F1 |
|---|---|---|---|---|
sentence_has_negation_or_denial |
StanfordSCALE/assertion_sentence_has_negation_or_denial | 148 (6.9%) | 0.902 | 0.941 |
sentence_includes_student_name |
StanfordSCALE/assertion_sentence_includes_student_name | 156 (7.3%) | 0.930 | 0.933 |
sentence_has_a_question |
StanfordSCALE/assertion_sentence_has_a_question | 466 (21.7%) | 0.867 | 0.922 |
sentence_has_math_terms |
StanfordSCALE/assertion_sentence_has_math_terms | 989 (46.1%) | 0.755 | 0.900 |
sentence_calls_on_student_by_name |
StanfordSCALE/assertion_sentence_calls_on_student_by_name | 97 (4.5%) | 0.918 | 0.889 |
sentence_is_a_declarative_statement_or_description |
StanfordSCALE/assertion_sentence_is_a_declarative_statement_or_description | 793 (37.0%) | 0.815 | 0.885 |
sentence_has_number |
StanfordSCALE/assertion_sentence_has_number | 560 (26.1%) | 0.755 | 0.874 |
sentence_has_fraction_terms |
StanfordSCALE/assertion_sentence_has_fraction_terms | 238 (11.1%) | 0.876 | 0.868 |
sentence_has_acknowledgment |
StanfordSCALE/assertion_sentence_has_acknowledgment | 454 (21.2%) | 0.709 | 0.840 |
sentence_has_agreement_or_affirmation |
StanfordSCALE/assertion_sentence_has_agreement_or_affirmation | 363 (16.9%) | 0.586 | 0.801 |
sentence_has_apology |
StanfordSCALE/assertion_sentence_has_apology | 6 (0.3%) | 0.869 | 0.800 |
sentence_addresses_the_whole_class |
StanfordSCALE/assertion_sentence_addresses_the_whole_class | 1403 (65.4%) | 0.546 | 0.799 |
sentence_has_comparison_terms |
StanfordSCALE/assertion_sentence_has_comparison_terms | 134 (6.2%) | 0.657 | 0.796 |
sentence_has_a_directive_or_instruction |
StanfordSCALE/assertion_sentence_has_a_directive_or_instruction | 430 (20.0%) | 0.735 | 0.790 |
sentence_is_a_short_utterance |
StanfordSCALE/assertion_sentence_is_a_short_utterance | 711 (33.2%) | 0.438 | 0.789 |
sentence_has_praise_or_encouragement |
StanfordSCALE/assertion_sentence_has_praise_or_encouragement | 77 (3.6%) | 0.604 | 0.780 |
sentence_uses_collaborative_or_inclusive_language |
StanfordSCALE/assertion_sentence_uses_collaborative_or_inclusive_language | 341 (15.9%) | 0.559 | 0.765 |
sentence_has_politeness_marker |
StanfordSCALE/assertion_sentence_has_politeness_marker | 46 (2.1%) | 0.731 | 0.760 |
sentence_invites_participation |
StanfordSCALE/assertion_sentence_invites_participation | 532 (24.8%) | 0.610 | 0.750 |
sentence_has_counting_sequence |
StanfordSCALE/assertion_sentence_has_counting_sequence | 17 (0.8%) | 0.757 | 0.732 |
sentence_has_time_reference |
StanfordSCALE/assertion_sentence_has_time_reference | 363 (16.9%) | 0.603 | 0.728 |
sentence_references_classroom_materials_or_visuals |
StanfordSCALE/assertion_sentence_references_classroom_materials_or_visuals | 319 (14.9%) | 0.686 | 0.719 |
sentence_references_task_procedure_or_logistics |
StanfordSCALE/assertion_sentence_references_task_procedure_or_logistics | 745 (34.7%) | 0.404 | 0.703 |
sentence_references_student_behavior_or_work |
StanfordSCALE/assertion_sentence_references_student_behavior_or_work | 743 (34.7%) | 0.382 | 0.672 |
sentence_checks_for_understanding_or_agreement |
StanfordSCALE/assertion_sentence_checks_for_understanding_or_agreement | 217 (10.1%) | 0.324 | 0.664 |
sentence_has_explanation_or_reasoning |
StanfordSCALE/assertion_sentence_has_explanation_or_reasoning | 214 (10.0%) | 0.593 | 0.615 |
sentence_expresses_personal_stance_or_thinking_aloud |
StanfordSCALE/assertion_sentence_expresses_personal_stance_or_thinking_aloud | 203 (9.5%) | 0.447 | 0.579 |
sentence_manages_classroom_behavior_or_attention |
StanfordSCALE/assertion_sentence_manages_classroom_behavior_or_attention | 219 (10.2%) | 0.460 | 0.566 |
sentence_poses_a_hypothetical_or_scenario |
StanfordSCALE/assertion_sentence_poses_a_hypothetical_or_scenario | 65 (3.0%) | 0.373 | 0.550 |
sentence_has_measurement_terms |
StanfordSCALE/assertion_sentence_has_measurement_terms | 189 (8.8%) | 0.180 | 0.548 |
sentence_quotes_or_reads_text_aloud |
StanfordSCALE/assertion_sentence_quotes_or_reads_text_aloud | 74 (3.4%) | 0.661 | 0.529 |
sentence_seeks_or_gives_clarification |
StanfordSCALE/assertion_sentence_seeks_or_gives_clarification | 357 (16.6%) | 0.270 | 0.506 |
sentence_narrates_ongoing_action |
StanfordSCALE/assertion_sentence_narrates_ongoing_action | 71 (3.3%) | 0.594 | 0.495 |
sentence_has_informal_language |
StanfordSCALE/assertion_sentence_has_informal_language | 614 (28.6%) | -0.120 | 0.494 |
sentence_has_answer_to_a_math_problem |
StanfordSCALE/assertion_sentence_has_answer_to_a_math_problem | 94 (4.4%) | 0.409 | 0.464 |
sentence_shows_uncertainty |
StanfordSCALE/assertion_sentence_shows_uncertainty | 50 (2.3%) | 0.503 | 0.459 |
sentence_expresses_certainty_or_emphasis |
StanfordSCALE/assertion_sentence_expresses_certainty_or_emphasis | 210 (9.8%) | 0.195 | 0.441 |
sentence_expresses_emotion_or_humor |
StanfordSCALE/assertion_sentence_expresses_emotion_or_humor | 59 (2.7%) | 0.275 | 0.421 |
sentence_references_prior_learning_or_lesson |
StanfordSCALE/assertion_sentence_references_prior_learning_or_lesson | 80 (3.7%) | 0.442 | 0.393 |
sentence_evaluates_a_student_response |
StanfordSCALE/assertion_sentence_evaluates_a_student_response | 150 (7.0%) | 0.433 | 0.391 |
sentence_has_disagreement_or_challenge |
StanfordSCALE/assertion_sentence_has_disagreement_or_challenge | 123 (5.7%) | 0.166 | 0.332 |
sentence_repeats_or_revoices_prior_speech |
StanfordSCALE/assertion_sentence_repeats_or_revoices_prior_speech | 238 (11.1%) | 0.419 | 0.313 |
sentence_is_incomplete_or_trails_off |
StanfordSCALE/assertion_sentence_is_incomplete_or_trails_off | 127 (5.9%) | 0.369 | 0.292 |
sentence_shows_realization_or_insight |
StanfordSCALE/assertion_sentence_shows_realization_or_insight | 77 (3.6%) | 0.080 | 0.229 |
sentence_answers_a_question |
StanfordSCALE/assertion_sentence_answers_a_question | 132 (6.2%) | 0.101 | 0.218 |
sentence_has_a_rhetorical_question |
StanfordSCALE/assertion_sentence_has_a_rhetorical_question | 24 (1.1%) | 0.180 | 0.176 |
sentence_expresses_confusion_or_requests_help |
StanfordSCALE/assertion_sentence_expresses_confusion_or_requests_help | 21 (1.0%) | 0.218 | 0.160 |
sentence_summarizes_or_reviews |
StanfordSCALE/assertion_sentence_summarizes_or_reviews | 60 (2.8%) | 0.229 | 0.128 |
sentence_grants_or_requests_permission |
StanfordSCALE/assertion_sentence_grants_or_requests_permission | 26 (1.2%) | 0.500 | 0.118 |
Krippendorff's alpha is the agreement between the two LLM annotators that labelled the 3,217 utterances in the dataset. Test F1 is the positive-class F1 of the published classifier on the held-out test set.