# Speaking Assessment Criteria and Rubric Design

## Holistic or analytic?

| Type | How | Use when | Trade-off |
|------|-----|----------|-----------|
| Holistic | One band for the whole performance | Quick placement, large numbers, low stakes | Fast, but no diagnostic information; raters weight features differently |
| Analytic | A band per criterion, combined | Progress and achievement tests, feedback, training | Diagnostic and more reliable; slower; halo effect between criteria |
| Checklist (can-do) | Yes/no for task features | Classroom, young learners, formative | Very clear; not for proficiency |

## Core criteria (used in most major speaking tests, wording paraphrased)

| Criterion | Looks at | Band features (low -> high) |
|-----------|----------|-----------------------------|
| Fluency and coherence | Speed, hesitation, length of turns, linking, organisation | Long pauses, isolated words -> some hesitation, simple linkers -> extended speech, range of cohesive devices, hesitation only to plan content |
| Lexical resource | Range, precision, collocation, paraphrase | Basic words, frequent gaps -> enough vocabulary for familiar topics, some paraphrase -> wide range, idiomatic, precise, effective paraphrase |
| Grammatical range and accuracy | Range of structures, complex sentences, error frequency and impact | Memorised phrases, frequent basic errors -> simple forms mostly correct, some complex forms with errors -> wide range, errors rare and minor |
| Pronunciation | Intelligibility, individual sounds, word and sentence stress, intonation, connected speech | Often unintelligible -> generally intelligible, L1 features cause some strain -> easy to understand throughout, features used to convey meaning |
| Interactive communication | Initiating, responding, turn-taking, developing the conversation, repair | Relies on partner, minimal responses -> maintains simple exchanges -> initiates, links contributions, negotiates towards an outcome |
| Task achievement / content | Did the speaker do what the task asked? | Task partly addressed -> all parts addressed -> fully developed, relevant, well supported |

Choose 3-5 criteria that match the construct of the test (an interview for a job course may need task achievement;
a pronunciation course weights pronunciation).

## Writing band descriptors

1. Start with the middle band (the expected standard), then the top and bottom, then the bands in between.
2. Describe observable features: what the speaker does, how often, with what effect on the listener.
3. Change one or two features between adjacent bands; keep the same features across bands so raters can
   compare ("errors in complex structures" -> "occasional errors in complex structures").
4. Use frequency words consistently: rarely < occasionally < sometimes < often < frequently.
5. Avoid native-speaker benchmarks; use intelligibility and effect on the listener.
6. Avoid purely evaluative words (good, adequate, poor) without features.
7. Make the lowest band describe what is present, not only what is missing.
8. Pilot with recordings; revise descriptors raters disagree on.

## Example: 4-band analytic rubric for a B1 speaking task

| Criterion | 4 (above B1) | 3 (B1 standard) | 2 (approaching B1) | 1 (below) |
|-----------|--------------|-----------------|--------------------|-----------|
| Fluency and coherence | Speaks at length with little hesitation; links ideas with a range of connectors | Keeps going with some pauses to plan; links ideas with simple connectors (and, but, because, so) | Frequent pauses; short turns; few connectors | Mostly single words or memorised phrases; long pauses |
| Vocabulary | Enough range to vary expression on the topic; some less common words | Enough vocabulary for the familiar topic; occasional gaps managed by paraphrase | Limited vocabulary; gaps cause breakdown | Very basic words; cannot express most ideas |
| Grammar | Simple forms accurate; attempts complex forms with some success | Simple forms mostly accurate; errors do not stop communication | Frequent errors in simple forms; meaning sometimes unclear | Errors in basic forms make meaning hard to follow |
| Pronunciation | Easy to understand; uses stress and intonation to show meaning | Generally intelligible; L1 features cause occasional strain | Some words unclear; listener effort often needed | Often unintelligible |

## Task design matters as much as the rubric

| Task | Elicits | Level |
|------|---------|-------|
| Interview (personal questions) | Familiar topics, simple description | A1-B1 |
| Picture description / comparison | Description, speculation | A2-C1 |
| Long turn (1-2 minutes on a prompt) | Extended discourse, organisation | B1-C2 |
| Paired discussion / decision task | Interaction, negotiation | A2-C2 |
| Role play with a problem | Functional language, repair | A2-B2 |
| Presentation + questions | Prepared and spontaneous speech | B1-C2 |

Use at least two task types for decisions that matter; one task gives too little evidence.