Skip to content

dimensions.md

Latest commit

 

History

History
211 lines (154 loc) · 10.3 KB

dimensions.md

File metadata and controls

211 lines (154 loc) · 10.3 KB
diataxis_type reference
diataxis_describes Voice profile dimensions, scoring tiers, and semantic differential pairs

Dimensions Reference

The voice profile measures writing voice across 8 gold standard dimensions, 10 gap dimensions, and 20 semantic differential pairs. Dimension scores range from 0-100. Semantic differential ratings range from 1.0-7.0.

Configuration files:

  • question-bank/scoring/dimension-item-mapping.json -- item-to-dimension mappings
  • question-bank/scoring/scoring-weights.json -- per-item weights and integration formula
  • question-bank/modules/SD-semantic-differential.json -- semantic differential items

Gold Standard Dimensions

The 8 core voice dimensions are measured with triangulated self-report and behavioral observation. Each produces a composite score blending self-report and observed values, weighted by confidence.

Dimension Description Contributing Modules SD Pairs Total Items Min Items Tier
formality Degree of formal vs. informal register M03 (6 items) formal_casual, accessible_specialized 8 5 1
emotional_tone Emotional valence and intensity, from reserved to expressive M04 (6 items) emotional_analytical, warm_detached 8 5 1
personality Degree to which the writer's personality emerges through prose M02 (5 items), M11 (3 items) personal_impersonal, earnest_ironic 10 6 2
complexity Syntactic and conceptual complexity M06 (5 items) elaborate_minimalist, concise_expansive 7 4 2
audience_awareness Degree of adaptation to and addressing the audience M07 (5 items) (none) 5 3 1
authority Level of authoritative, expert, or confident voice M08 (4 items), M02 (2 items) assertive_tentative, authoritative_collaborative 8 5 2
narrativity Tendency toward narrative and storytelling M05 (5 items) narrative_expository 6 4 1
humor Use of humor, wit, irony, and playfulness M04 (2 items), M02 (2 items) serious_playful 5 3 1

Contributing Items

formality

Source Items
M03 M03-Q01, M03-Q02, M03-Q03, M03-Q04, M03-Q05, M03-Q06
SD formal_casual, accessible_specialized

emotional_tone

Source Items
M04 M04-Q01, M04-Q02, M04-Q03, M04-Q04, M04-Q05, M04-Q06
SD emotional_analytical, warm_detached

personality

Source Items
M02 M02-Q01, M02-Q02, M02-Q03, M02-Q04, M02-Q05
M11 M11-Q01, M11-Q02, M11-Q03
SD personal_impersonal, earnest_ironic

complexity

Source Items
M06 M06-Q01, M06-Q02, M06-Q03, M06-Q04, M06-Q05
SD elaborate_minimalist, concise_expansive

audience_awareness

Source Items
M07 M07-Q01, M07-Q02, M07-Q03, M07-Q04, M07-Q05

authority

Source Items
M08 M08-Q01, M08-Q02, M08-Q03, M08-Q04
M02 M02-Q06, M02-Q07
SD assertive_tentative, authoritative_collaborative

narrativity

Source Items
M05 M05-Q01, M05-Q02, M05-Q03, M05-Q04, M05-Q05
SD narrative_expository

humor

Source Items
M04 M04-Q07, M04-Q08
M02 M02-Q08, M02-Q09
SD serious_playful

Gap Dimensions

Secondary voice dimensions that extend the profile beyond the 8 gold-standard measures. Each is scored 0-100 with a source field indicating measurement origin (self_report, observed, or both).

Dimension Description Contributing Items Min Items
precision Attention to grammatical precision, word choice exactness, and mechanical correctness M09-Q01, M09-Q02, M09-Q03, M09-Q04, M09-Q05 3
revision_orientation Approach to revision and editing in the writing process M10-Q01, M10-Q02, M10-Q03, M10-Q04 2
contextual_flexibility Ability and willingness to shift voice across different writing contexts M11-Q04, M11-Q05, M11-Q06 2
risk_tolerance Willingness to take creative risks and experiment with unconventional voice choices M02-Q10, M05-Q04 1
self_awareness Metacognitive awareness of one's own writing voice and its effects M12-Q01, M12-Q02, M12-Q03 2
consistency Consistency of voice across writing sessions and pieces M12-Q04, M12-Q05 1
influence_absorption Tendency to absorb and integrate influences from other writers M02-Q03, M10-Q03 1
vocabulary_range Breadth and register range of vocabulary usage M06-Q02, M09-Q03 1
sentence_rhythm Characteristic sentence length patterns and rhythmic variation M06-Q04, M09-Q06 1
genre_awareness Awareness of and adherence to genre-specific voice conventions M07-Q04, M08-Q03 1

Semantic Differential Pairs

20 bipolar adjective pairs rated on a 7-point scale (1.0 = first pole, 4.0 = neutral, 7.0 = second pole). All pairs are presented to all writer types.

ID Pole 1 (1) Pole 2 (7) Mapped Dimensions Weights Tier
SD-01 Formal Casual formality formality: 1.0 1
SD-02 Serious Playful humor humor: 1.0 1
SD-03 Elaborate Minimalist complexity complexity: 1.0 2
SD-04 Assertive Tentative authority authority: 1.0 2
SD-05 Warm Detached emotional_tone emotional_tone: 1.0 1
SD-06 Concrete Abstract complexity, narrativity complexity: 0.5, narrativity: 0.5 2
SD-07 Structured Flowing complexity complexity: 1.0 2
SD-08 Direct Diplomatic authority, audience_awareness authority: 0.5, audience_awareness: 0.5 2
SD-09 Accessible Specialized formality, audience_awareness formality: 0.5, audience_awareness: 0.5 1
SD-10 Emotional Analytical emotional_tone, narrativity emotional_tone: 0.5, narrativity: 0.5 1
SD-11 Authoritative Collaborative authority authority: 1.0 2
SD-12 Conventional Experimental personality personality: 1.0 2
SD-13 Precise Evocative personality, complexity personality: 0.5, complexity: 0.5 2
SD-14 Concise Expansive complexity complexity: 1.0 1
SD-15 Personal Impersonal personality, formality personality: 0.5, formality: 0.5 1
SD-16 Earnest Ironic humor, personality humor: 0.5, personality: 0.5 2
SD-17 Careful Bold authority, personality authority: 0.5, personality: 0.5 2
SD-18 Narrative Expository narrativity narrativity: 1.0 2
SD-19 Inclusive Exclusive audience_awareness audience_awareness: 1.0 1
SD-20 Spontaneous Deliberate personality personality: 1.0 2

SD Normalization

Semantic differential scores are normalized from the 1-7 scale to 0-100 before integration:

normalized = ((raw - 1) / 6) * 100

Self-Reportability Tiers

Self-reportability tiers indicate how accurately participants can self-assess a dimension. The tier determines how self-report data is weighted relative to observed (behavioral/computational) data.

Tier Label Reliability Description
1 High High Respondents can accurately self-assess. Self-report items are weighted at full value.
2 Moderate Moderate Respondents have partial insight. Projective and behavioral items are weighted more heavily when available.
3 Low Low Limited self-awareness on this dimension. Behavioral observation is the primary data source.
4 Very Low Very low Self-report is unreliable. Measurement relies entirely on observational and computational methods.

Tier Assignments

Tier Dimensions
1 (High) formality, emotional_tone, audience_awareness, narrativity, humor
2 (Moderate) personality, complexity, authority

Tiers 3 and 4 are not assigned to any gold standard dimension in the current question bank but are used for individual question items where indirect measurement is required.

Scoring Integration Formula

The final dimension score combines module item scores and semantic differential scores using a weighted blend:

dimension_score = (0.7 * module_items_mean) + (0.3 * sd_normalized_mean)
Component Weight Description
module_items_mean 0.70 Mean across all answered module items for the dimension. Missing items are excluded (not zero-filled), provided min_items_for_score is met.
sd_normalized_mean 0.30 Mean of normalized semantic differential scores (1-7 mapped to 0-100) for pairs assigned to the dimension.

All items within a dimension carry equal weight (1.0) unless otherwise specified in scoring-weights.json. This reflects the psychometric design where each item was calibrated to contribute equally to its target dimension.

Tier 2 Adjustment

For Tier 2 (moderate self-reportability) dimensions, projective item scores are preferred when self-report and projective scores diverge significantly. The self_perception_divergence deep-dive trigger (defined in deep-dive-triggers.json) detects this condition and injects additional measurement items.

Computational Validators

The NLP pipeline (voice-nlp) produces observed scores that serve as computational validators for self-report data. The mapping from writing analysis features to gold standard dimensions:

Dimension Key NLP Features
formality formality_f_score, latinate_germanic_ratio, contractions_per_100_words
emotional_tone liwc_equivalent.emotional_tone, positive_emotion_pct, negative_emotion_pct
personality liwc_equivalent.authenticity, first_person_singular_pct, self_mention_count
complexity flesch_kincaid_grade, mean_sentence_length, dependent_clause_ratio, mtld
audience_awareness liwc_equivalent.clout, second_person_pct, engagement_markers
authority liwc_equivalent.clout, hedge_boost_ratio, active_voice_ratio
narrativity liwc_equivalent.analytical_thinking (inverse), temporal_connective_density, referential_cohesion
humor Requires manual/LLM evaluation; no direct NLP proxy.

Calibration

When both self-report and observed scores are available, the scoring engine (voice-scoring score) produces a calibration report comparing the two. Dimensions where self-report and observed scores diverge by more than a threshold are flagged as blind spots or aspirational gaps in the voice profile's calibration section.