Research map / Label-efficient and robust learning / Recognition and long-tail
Quantum-inspired Non-homologous Representation Constraint Mechanism for Long-tail Senses of Word Sense Disambiguation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes QiWSD, a quantum-inspired WSD system whose superposition-state and interference-term constraint on BERT embeddings of glosses and example sentences improves long-tail sense disambiguation.
Problem
Long-tail word sense disambiguation suffers from few training samples and unclear, hard-to-distinguish sense definition representations learned from limited glosses.
Approach
Treats gloss and example-sentence embeddings as normalized quantum states combined into a superposition, and uses the quantum-measurement interference term as a geometric constraint in loss, with a parallel traditional BERT-based scorer for high-frequency senses.
Key results
- QiWSD-large reaches 81.8 F1 on the concatenation of all standard test sets, versus 80.8 for QiWSD-base and 80.4 for the quantum-inspired DRWSD.
- On LTS subsets, the full QiWSD-base scores 49.3 on ALL versus 34.9 without the quantum recognition branch.
- In the data-enhanced setting, QiWSD-large reaches 82.1 ALL, compared with 80.7 for ESR.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use QiWSDlarge instead of DRWSD | English all-words WSD under standard evaluation · ALL | F1-score (%): 80.4 -> 81.8 | Helped |
| Use QiWSDlarge instead of ESR | English all-words WSD under data-enhanced evaluation · ALL | F1-score (%): 80.7 -> 82.1 | Helped |
| Add the quantum recognition method (ablation) | QiWSDbase on SemCor WSD evaluation · ALL | F1-score (%): 77.7 -> 80.8 | Helped |
| Add the quantum recognition method (ablation) | QiWSDbase on the LTS dataset · ALL | F1-score (%): 34.9 -> 49.3 | Helped |
Surprising findings
- Performance gap to competing systems is small, which the authors attribute to the small number of long-tail senses.
- Quantum-inspired geometric constraint is claimed to help only with non-homologous data; homologous data yields only representation combination and continuous-space constraints.
Limitations
- Gains over strong baselines are modest on standard test sets; authors note the gap is small due to few LTS samples.
- Weaker performance on adjective and adverb subsets, attributed to few long-tail senses there.
- Internal mechanism and applicable fields are not yet clarified by the authors.
Experimental setup
Datasets: SemCor WordNet 3.0 glosses WNGT (GlossTag) SemEval-07 Senseval-2 Senseval-3 SemEval-13 SemEval-15 BabelNet cross-lingual XL-WSD
Benchmarks: Unified WSD evaluation framework (SE2, SE3, SE13, SE15, ALL) SE7 dev set LTS subsets XL-WSD cross-lingual datasets
Metrics: F1
Baselines: EWISE LMMS SREF ARES BEM EWISER SyntagRank COF ESR Z-Reweighting QWSD DRWSD
Builds on: BERT EWISE BEM QWSD DRWSD
Closest papers by meaning
- Quantum-Inspired Representation for Long-Tail Senses of Word Sense Disambiguation (AAAI 2023 · similarity 0.93)
- Quantum Interference Model for Semantic Biases of Glosses in Word Sense Disambiguation (AAAI 2024 · similarity 0.90)
- XL-WSD: An Extra-Large and Cross-Lingual Evaluation Framework for Word Sense Disambiguation (AAAI 2021 · similarity 0.85)
- Retrieval-based Disentangled Representation Learning with Natural Language Supervision (ICLR 2024 · similarity 0.82)
- Enriching Disentanglement: From Logical Definitions to Quantitative Metrics (NeurIPS 2024 · similarity 0.82)
- QPEN: Quantum Projection and Quantum Entanglement Enhanced Network for Cross-Lingual Aspect-Based Sentiment Analysis (AAAI 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.