Contrastive Sets¶
Generate research-based phonological intervention word lists using three evidence-based approaches.
Overview¶
The Contrastive Sets tool provides three clinical intervention methods: 1. Minimal Pairs - Single-feature contrast (target vs. substitute) 2. Maximal Opposition - Two unknown phonemes with major class difference (Gierut 1989-1992) 3. Multiple Opposition - Global collapse treatment with minimal sets (Gierut 1989-1992, Storkel 2022)
Vocabulary: the ~48K-word canonical content vocabulary, drawing on 642K precomputed minimal pairs
A "Picture pairs only" toggle restricts results to pairs (or sets) where every member has a picture card, and results can be sent straight into a therapy pack with Add to Editor (see the Therapy Packs guide).
Minimal Pairs¶
Phonological intervention using single-feature contrasts between a target phoneme (correct production) and substitute phoneme (child's production) at the same position within words.
Key constraint: The phoneme difference must occur at the same position in both words (e.g., initial, medial, or final position).
Usage¶
- Select "Minimal Pairs"
- Enter Target phoneme (correct production, e.g., /θ/)
- Enter Substitute phoneme (child's production, e.g., /t/)
- Choose position: initial, medial, final, or any
- Optional: Toggle "Picture pairs only" to require imagery on both words
- Generate pairs
Algorithm¶
Data source: 642K precomputed minimal-pair relationships (queried server-side in D1; browse results are restricted to canonical content-word pairs)
Search process:
1. Filter pairs where phoneme1 = target AND phoneme2 = substitute
2. Filter by position (initial/medial/final) if specified
3. If "Picture pairs only": require a picture card on both words
4. Return matched pairs, sorted by combined word frequency
Example¶
Input: - Target: /θ/ (correct: "think") - Substitute: /t/ (child says: "tink") - Position: Initial
Expected output: - thin /θɪn/ → tin /tɪn/ - thick /θɪk/ → tick /tɪk/ - theme /θim/ → team /tim/
Research Context¶
Typical clinical application: - Single phoneme error pattern - Traditional phonological intervention approach - Focused practice on specific contrast
Research findings on generalization: - Addresses single error pattern - Generalization to other phonemes varies by individual - May require extended intervention for broader phonological change
Maximal Opposition¶
Targets two unknown phonemes (neither produced correctly) that differ maximally in phonological features, including major class differences (obstruent vs. sonorant). Words contrast these phonemes at the same position.
Key constraint: Like minimal pairs, maximal opposition requires phoneme contrast within the same position in word pairs.
Theory (Gierut 1989-1992)¶
Research basis: - Pairing two unknown sounds promotes broader phonological learning than one unknown + one known - Major class differences (obstruent vs. sonorant) show greater generalization - Maximal feature differences highlight phonological diversity - System-wide changes extend beyond trained sounds
Clinical evidence: - More efficient than minimal pairs for moderate-severe SSD - Promotes generalization to untrained phonemes - Requires fewer treatment sessions
Ranking Algorithm¶
PhonoLex ranks phoneme pairs by their continuous feature distance — the L2 distance between the two phonemes' learned 26-dimensional articulatory feature vectors, precomputed in the pairs database — together with a sonorant difference value that measures how far apart the pair sits on the sonorant dimension:
feature_distance = L2 distance between the two phonemes'
learned 26-d feature vectors (precomputed)
sonorant_diff = difference on the sonorant dimension
Major class difference:
Sonorants: /m, n, ŋ, l, ɹ, w, j/ (and vowels)
Obstruents: /p, t, k, b, d, g, f, v, s, z, θ, ð, ʃ, ʒ, h, ʧ, ʤ/
Major class difference = the pair crosses the obstruent–sonorant
boundary (sonorant_diff > 0)
Ranking examples:
| Pair | Major Class? | Interpretation |
|---|---|---|
| /θ/ - /l/ | ✓ Yes | Suggested (obstruent vs. sonorant); ranked by feature distance |
| /s/ - /l/ | ✓ Yes | Suggested (obstruent vs. sonorant); ranked by feature distance |
| /ʃ/ - /ɹ/ | ✓ Yes | Suggested (obstruent vs. sonorant); ranked by feature distance |
| /s/ - /ʃ/ | ✗ No | Not suggested (both obstruents) |
| /l/ - /ɹ/ | ✗ No | Not suggested (both sonorants) |
Threshold: PhonoLex only suggests pairs that cross the obstruent–sonorant boundary; among those, pairs with the largest feature distance rank first
Usage¶
- Select "Maximal Opposition"
- Enter unknown phonemes (comma-separated, e.g., /s, ʃ, θ, l, ɹ/)
- Click Generate Phoneme Pairs
- System automatically ranks all pairs by feature distance
- Review top-ranked pairs
- Select a pair
- Click Generate Word Lists to find minimal pairs
Algorithm¶
Step 1: Generate all pairs
For each phoneme pair (i, j) where i < j:
Look up the precomputed feature_distance and sonorant_diff
Keep the pair only if it crosses the obstruent–sonorant boundary
Sort kept pairs by feature_distance, highest first
Step 2: Find word lists
For selected pair (phoneme1, phoneme2):
Load precomputed minimal pairs
Filter pairs where BOTH phonemes occur at the SAME position:
(word1 has phoneme1 at position X AND word2 has phoneme2 at position X)
OR
(word1 has phoneme2 at position X AND word2 has phoneme1 at position X)
Group by position (initial/medial/final)
Return pairs
Note: Position X must be identical in both words (e.g., both at position 0,
or both at position 2). This ensures true opposition within the same context.
Complexity: O(n) where n = minimal pairs count - ~5-10 ms lookup
Example¶
Input: - Unknown phonemes: /s, ʃ, θ, l, ɹ/
System output (suggested pairs, ranked by feature distance):
| Pair | Major Class | Contrast |
|---|---|---|
| /θ/ - /l/ | ✓ | Crosses obstruent–sonorant; differs in manner, voicing, laterality |
| /s/ - /l/ | ✓ | Crosses obstruent–sonorant; differs in manner, voicing, stridency |
| /ʃ/ - /ɹ/ | ✓ | Crosses obstruent–sonorant; differs in manner, voicing, place |
Selected pair: /θ/ - /l/
Word lists: - Initial: thin/Lynn, think/link, thumb/lumb - Final: mouth/mole, bath/ball, math/mall
Articulatory Features¶
PhonoLex ranks contrasts using 26-dimensional learned Bayesian feature vectors — a Hayes (2009)-based articulatory prior refined against perceptual-confusability and acoustic evidence. The discrete feature labels below are shown for interpretation (which dimensions a contrast spans):
Major features (commonly different):
| Feature | Values | Description | Examples |
|---|---|---|---|
| consonantal | + / - | Constriction in vocal tract | +: /t, s, l/ -: /a, i/ |
| sonorant | + / - | Spontaneous voicing | +: /l, m, a/ -: /t, s/ |
| continuant | + / - | Airflow continues | +: /s, f, l/ -: /t, p/ |
| voice | + / - | Vocal fold vibration | +: /b, d, z/ -: /p, t, s/ |
| nasal | + / - | Nasal cavity open | +: /m, n, ŋ/ -: /p, t/ |
| lateral | + / - | Air flows around tongue sides | +: /l/ -: all others |
| strident | + / - | High-frequency noise | +: /s, z, ʃ, ʒ/ -: /f, θ/ |
Place features:
| Feature | Values | Description | Examples |
|---|---|---|---|
| labial | + / - | Lips involved | +: /p, b, m, f, v/ -: /t, k/ |
| coronal | + / - | Tongue blade/tip | +: /t, d, s, l/ -: /p, k/ |
| dorsal | + / - | Tongue body | +: /k, g, ŋ/ -: /t, p/ |
| anterior | + / - | Front of mouth | +: /t, s/ -: /k, ʃ/ |
| distributed | + / - | Broad tongue contact | +: /ʃ, ʒ/ -: /s, z/ |
Vowel features:
| Feature | Values | Description | Examples |
|---|---|---|---|
| syllabic | + / - | Forms syllable nucleus | +: /a, i, o/ -: /t, k/ |
| high | + / - | Tongue raised | +: /i, u/ -: /a/ |
| low | + / - | Tongue lowered | +: /a, æ/ -: /i, u/ |
| front | + / - | Tongue forward | +: /i, e/ -: /u, o/ |
| back | + / - | Tongue back | +: /u, o/ -: /i, e/ |
| tense | + / - | Muscular tension | +: /i, u/ -: /ɪ, ʊ/ |
| round | + / - | Lips rounded | +: /u, o/ -: /i, a/ |
Complete feature list: See Articulatory Features Reference for the full feature system with definitions and examples.
Clinical Research Context¶
Assessment considerations: - Maximal opposition typically targets phonemes not produced in any context - Research studies used assessment data from single-word naming and connected speech - Stimulability testing may inform phoneme selection
Phoneme pair selection: - Research prioritized pairs with major class differences (obstruent vs. sonorant) - Studies balanced feature maximization with learnability - Feature distance alone doesn't predict treatment success
Word list characteristics in research: - Studies typically used high-frequency words - Imageability considerations appeared in studies with young children - Initial position frequently targeted due to perceptual salience - Study word lists ranged from 8-20 pairs per position
Generalization patterns in research: - Studies documented system-wide phonological changes - Generalization to untrained phonemes varied by individual - Some studies reported generalization within 8-12 weeks
Multiple Opposition¶
Treats global phoneme collapse by contrasting the substitute phoneme with multiple target phonemes simultaneously using minimal sets (triplets, quadruplets, quintuplets). All phonemes in a set contrast at the same position.
Key constraint: All words in a minimal set must differ at exactly the same position (e.g., all at initial position, or all at position 2). This creates a clear phonological contrast point.
Theory (Gierut 1989-1992, Storkel 2022)¶
When to use: - Child collapses multiple phonemes to one substitute - Example: /t, d, k, g/ all produced as [t] - Global phonological patterns need addressing
Research basis: - Addresses entire collapse pattern simultaneously - Minimal sets force attention to multiple contrasts - Faster generalization than sequential minimal pairs - More efficient than treating each phoneme individually
Clinical evidence: - Effective for severe phonological disorders - Promotes awareness of phonological system - Requires cognitive engagement (more complex than minimal pairs)
Algorithms¶
PhonoLex uses two complementary algorithms:
1. Maximal Classification¶
Purpose: Select representative target phonemes from the collapsed set
Algorithm:
Given: substitute phoneme, list of target phonemes
Goal: Select subset that maximizes phonological diversity
For each target phoneme:
Compute feature distance from substitute
Compute feature distance from other targets
Select targets that:
1. Differ maximally from substitute
2. Differ maximally from each other
Example: - Substitute: /t/ - Candidates: /d, k, g, b, p/ - Selected: /d/ (voice), /k/ (dorsal), /g/ (voice + dorsal) - Rationale: Maximizes feature spread (voice, place)
2. Maximal Distinction¶
Purpose: Find minimal sets where all phonemes contrast at the same position
Algorithm:
Given: substitute + selected targets
Goal: Find words that form minimal sets (differ at the SAME POSITION only)
1. Index words by position and phoneme
For each word:
For each position P:
index[P][phoneme].add(word)
2. Find minimal sets
For each position P:
For each phoneme in [substitute + targets]:
candidates[phoneme] = words with this phoneme at position P
If all phonemes have candidates at position P:
Create minimal set by selecting one word per phoneme
All words differ at position P only
Example: tie, die, kite, guy (all differ at position 0)
Critical constraints: - All words must have same phoneme length - Must differ at exactly one position (the same position for all) - That position can be initial, medial, or final - Counter-example: tie /taɪ/, die /daɪ/, sky /skaɪ/ is NOT valid because /k/ in "sky" is at a different position than /t/ and /d/
Complexity: O(n) where n = vocabulary size (indexing is linear) - ~30-80 ms for full vocabulary
Usage¶
- Select "Multiple Opposition"
- Enter substitute phoneme (what child says, e.g., /t/)
- Enter target phonemes (what should be said, e.g., /d, k, g/)
- Optional: Choose position (initial/medial/final/any)
- Click Generate Sets
- System uses Maximal Classification + Maximal Distinction
- Review minimal sets (triplets, quadruplets, quintuplets)
Example¶
Input: - Substitute: /t/ - Targets: /d, k, g/ - Position: Initial
System process:
Step 1: Maximal Classification - All targets selected (small set, all differ from /t/)
Step 2: Maximal Distinction - Find 4-word minimal sets (substitute + 3 targets)
Output (minimal sets):
| Set | /t/ (substitute) | /d/ (target) | /k/ (target) | /g/ (target) |
|---|---|---|---|---|
| Set 1 | tie | die | kite* | guy |
| Set 2 | tore | door | core | gore |
| Set 3 | tear | dear | care | gear |
*Note: "kite" has different structure but contrasts /t/ vs /k/ in different positions - system will flag non-perfect sets
Clinical use: - Present all 4 words simultaneously - Child must produce correct phoneme for each word - Contrasts highlight phonological differences - Promotes system-level awareness
Set Size in Research¶
| Set Size | Name | Typical Application |
|---|---|---|
| 3 words | Triplet | Substitute + 2 targets |
| 4 words | Quadruplet | Substitute + 3 targets |
| 5 words | Quintuplet | Substitute + 4 targets |
Research findings: - Studies used various set sizes depending on collapse pattern - Larger sets (4-5 words) appeared in studies with older children - Set size often determined by availability of appropriate word sets rather than predetermined choice
Word Selection in Research¶
Common characteristics in published studies: - Words typically matched on phoneme length - High-frequency words often preferred - Imageability considered in some studies with younger populations
Factors noted in research: - Word familiarity relevant to treatment success - Some studies used semantically related sets - Phonological complexity varied across studies
Research Comparison¶
| Approach | Typical Application | Participant Profile | Generalization Findings | Task Complexity |
|---|---|---|---|---|
| Minimal Pairs | Single phoneme contrast | Varied | Focused on trained contrast | Single contrast |
| Maximal Opposition | Two unknown phonemes | Moderate-severe SSD | Broader system changes | Dual contrast |
| Multiple Opposition | Phoneme collapse | Severe SSD | System-wide changes | Multiple contrasts |
Application context from research:
- Minimal Pairs in research
- Studies targeted specific phoneme contrasts
- Example: /s/ vs. /θ/ contrast training
-
Traditional approach with extensive research base
-
Maximal Opposition in research
- Studies selected phoneme pairs with major class differences
- Example: /θ/ - /l/ pairing in Gierut (1989)
-
Research showed broader generalization than minimal pairs
-
Multiple Opposition in research
- Studies addressed global collapse patterns
- Example: /t, d, k, g/ → [t] collapse
- Research found efficient system-wide phonological change
Performance Characteristics¶
| Operation | Time | Notes |
|---|---|---|
| Minimal pairs lookup | 1-5 ms | Precomputed database |
| Maximal opposition pairs | 20-50 ms | O(p²) phoneme comparisons |
| Word list generation | 5-10 ms | Filtered minimal pairs |
| Multiple opposition sets | 30-80 ms | O(n) vocabulary indexing |
Data Sources¶
Minimal pairs: 642K precomputed relationships (browse surfaces filter to canonical content-word pairs at query time)
Features: 26-d learned Bayesian phoneme vectors (Hayes 2009 prior, refined with perceptual-confusability and acoustic evidence)
Research basis: Clinical studies by Gierut (1989, 1990, 1992) and Storkel (2022)
References¶
Clinical Research:
- Gierut, J. A. (1989). Maximal opposition approach to phonological treatment. Journal of Speech and Hearing Disorders, 54(1), 9-19.
- Gierut, J. A. (1990). Differential learning of phonological oppositions. Journal of Speech and Hearing Research, 33(3), 540-549.
- Gierut, J. A. (1992). The conditions and course of clinically induced phonological change. Journal of Speech and Hearing Research, 35(5), 1049-1063.
- Gierut, J. A., & Neumann, H. J. (1992). Teaching and learning /θ/: A non-confound. Clinical Linguistics & Phonetics, 6(3), 191-200.
- Storkel, H. L. (2022). Minimal, Maximal, or Multiple Oppositions: A review of phonological intervention approaches. Language, Speech, and Hearing Services in Schools, 53(2), 421-437.
Phonological Features:
- Hayes, B. (2009). Introductory Phonology. Wiley-Blackwell.
See Also¶
- Practical Examples - Worked examples for all three approaches
- Articulatory Features Reference - Complete 38-feature system
- Technical Architecture - Algorithm implementation details