Bounded word reward with predicted length
Mathematical Methods AIMech:PredictedLengthBoundedReward DRAFT
A sequence rescoring method that adds a fixed word reward only up to an input-specific length predicted from encoder states, replacing a corpus-level generation-ratio cap with Lpred(x).
Identity Notes & Concept Scope
This is Yang et al.’s predicted-length bounded word-reward variant. It changes the cap, while retaining a separately tuned reward rate r. It is not the beam-statistic adaptive reward or the brevity-penalty normalized score, and it is not a standalone trained length-predictor instance. No normalized identity match was found across the 442 AIO and 146 prior native records; both checkout text searches included ignored and hidden files and returned no matches for the selected exact IDs, full labels, aliases and source titles.
Claims & Grounded Evidence
1 Claim
Mechanism
#method
Feed the mean source encoder hidden state to a two-layer multilayer perceptron to predict a target/source generation ratio, then multiply it by the source length to obtain Lpred(x). Score candidate y by S(x,y)+r·min(|y|,Lpred(x)), where S is cumulative log probability. The predictor is evaluated from the input and the reward cap is fixed for that input during decoding.
⚠️ Curated Limitations & Negative Results
- The method still requires a learned length predictor and a tuned reward rate; replacing the corpus ratio does not make the complete system parameter-free.
- Predicted length is a reward cap, not automatically an enforced maximum sequence length or an accurate target for every input.
- A usable nonnegative length prediction, tokenizer conventions and integration with the decoder must be specified. The paper passage does not provide a complete predictor training or calibration protocol.
- Finite beam width still prunes hypotheses. Neither accurate length prediction nor a stopping certificate implies globally optimal output or universally better translation quality.
- The empirical source scope is Chinese-to-English news translation; no numerical result is curated.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:NaturalLanguageProcessing
Natural Language Processing
|
related_match | The primary paper defines this sequence scoring, stopping or pruning method for natural-language generation, with neural machine translation as its central setting. The exact AIO CLASS describes the language-processing task. This is an method-to-application-domain relationship, not equivalence or an is-a assertion; the method is not a trained language-model instance and may also apply outside NLP. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
method-source
|
Yilin Yang, Liang Huang, Mingbo Ma
|
primary paper
full_text
|
ID: doi:10.18653/v1/D18-1342
sha256:019ba14d74ad…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T06:36:44Z
Updated: 2026-10-11T06:37:11Z
Source File:
data/curated/PredictedLengthBoundedReward.yaml