--- title: "ECNU at SemEval-2017 Task 7 — paper by Xiu et al. (2017)" description: "This paper presents the ECNU team's systems for the first two subtasks of SemEval-2017 Task 7… 3 references, cited by 5 papers here." authors: ["Yuhuan Xiu", "Man Lan", "Yuanbin Wu"] year: 2017 venue: "Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)" source: "https://aclanthology.org/S17-2078/" doi: "10.18653/v1/s17-2078" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/xiu-2017-ecnu-semeval-2017-task-7/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/xiu-2017-ecnu-semeval-2017-task-7/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2017](https://md.research.elamur.ai/free/en/papers/2017/page/5/) / ECNU at SemEval-2017 Task 7: Using Supervised and Unsupervised Methods to Detect and Locate English Puns # ECNU at SemEval-2017 Task 7: Using Supervised and Unsupervised Methods to Detect and Locate English Puns **Authors:** Yuhuan Xiu, [Man Lan](https://md.research.elamur.ai/free/en/authors/man-lan/) and [Yuanbin Wu](https://md.research.elamur.ai/free/en/authors/yuanbin-wu/) · **Year:** 2017 · **Venue:** [Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)](https://md.research.elamur.ai/free/en/venues/semeval/) ([SemEval 2017](https://md.research.elamur.ai/free/en/venues/semeval/2017/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/S17-2078/) · [PDF](https://aclanthology.org/S17-2078.pdf) · [DOI 10.18653/v1/s17-2078](https://doi.org/10.18653/v1/s17-2078) · [OpenAlex](https://openalex.org/W2752471326) **Citations:** 11 citations (OpenAlex, 2026-10-14) · [5](https://md.research.elamur.ai/free/en/papers/xiu-2017-ecnu-semeval-2017-task-7/cited-by/) in our library, of the 149,439 papers whose bibliographies we read ## Summary This paper presents the ECNU team's systems for the first two subtasks of SemEval-2017 Task 7: deciding whether an English sentence contains a pun (detection) and finding the pun word (location). The task supplied no official training data, so the authors built a small training set themselves. It combines puns from the Pun of the Day website with pun-free sentences collected online. The two pun types get different treatment. For homographic puns, Simplified Lesk over WordNet picks plausible senses for each content word. Each sense is embedded with Google word2vec, and six similarity kernels measure how far apart a word's senses are. For heterographic puns, all words are grouped with k-means, and the pun is assumed to sit in a semantically outlying cluster. Detection uses AdaBoost and RandomForest classifiers on pooled similarity features. Location is unsupervised: the word with the lowest similarity score is picked. The work shows that cheap, knowledge-based sense signals plus a tiny self-built dataset can give a workable baseline for wordplay recognition. ## Key points - Training data was self-collected: 60 homographic and 60 heterographic puns from Pun of the Day, plus 120 negative sentences, giving 180 sentences per pun type. The official test sets had 2,250 homographic sentences (1,607 with puns) and 1,780 heterographic sentences (1,271 with puns). - Preprocessing used Stanford CoreNLP POS tagging and stop-word removal. The remaining words are the "target words." - Similarity features come from six kernels: cosine, Manhattan, Euclidean, Pearson, Spearman, and a tanh-based "sigmoid" kernel. They produce a 6-d vector per word, and min/max/mean pooling turns these into an 18-d vector per sentence. - Sense vectors are 900-d: min, max and average pooling of 300-d word2vec vectors over the WordNet synonyms, gloss and examples. Heterographic processing uses k-means with k=100 over word vectors. - In 5-fold cross-validation on the training data, the AdaBoost+RandomForest ensemble did best on homographic puns (reported F1 0.7747). AdaBoost alone did best on heterographic puns (F1 0.8316). ## Interesting ideas - A word whose top candidate senses are semantically distant from each other in context is a likely pun, so the spread between senses serves as an ambiguity signal. - The Lesk filter keeps every sense scoring at least the second-best overlap, rather than only the single best one. This preserves deliberate ambiguity cheaply. - Clustering a sentence's content words and finding the outlier cluster localizes the word that does not fit the topic. - Several heterogeneous similarity kernels are compressed by min/max/mean pooling into a tiny fixed feature vector. This makes a small classifier trainable on just 180 examples. - When no training set exists, a small labelled corpus scraped from a themed website, plus generic negatives, can be enough to tune a system. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (frames): The irony/implicature module could flag words where several WordNet senses stay plausible yet unrelated, marking possible deliberate double meanings instead of forcing one sense. - **I2** (world): When grounding entities to the WordNet ontology, Mova could keep a short list of tied senses with explicit overlap evidence, so later stages can explain why a word stayed ambiguous. - **I3** (answer): Fable and tale readers could use an explainable "odd word out" check to spot wordplay or anomalous lexical choices before answering questions about meaning or humor. - **I4** (learn): Low-data diagnostic classifiers, such as one for detecting jokes in silver data, could use a few interpretable pooled features instead of large models, keeping each decision traceable. - **I5** (learn): Mova could bootstrap evaluation sets for rare phenomena (puns, sarcasm) the same way, pairing them with negative controls to check that detectors do not fire on ordinary text. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* ## Institutions - [East China Normal University](https://md.research.elamur.ai/free/en/universities/east-china-normal-university/) — University, Shanghai, China · 69 works · 2,490 citations 📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works 🧭 Topics (1): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,379 works ## Cite Cite as: Yuhuan Xiu, Man Lan and Yuanbin Wu. 2017. ECNU at SemEval-2017 Task 7: Using Supervised and Unsupervised Methods to Detect and Locate English Puns. Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017). The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{xiu2017ecnu, title = {{ECNU at SemEval-2017 Task 7: Using Supervised and Unsupervised Methods to Detect and Locate English Puns}}, author = {Yuhuan Xiu, Man Lan and Yuanbin Wu}, booktitle = {Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)}, year = {2017}, doi = {10.18653/v1/s17-2078}, url = {https://doi.org/10.18653/v1/s17-2078} } ``` - [← previous Core Arguments in Universal Dependencies](https://md.research.elamur.ai/free/en/papers/zeman-2017-core-arguments-universal-dependencies/) - [next → TEI-Lex0 Guidelines for the Encoding of Dictionary Information on Written and Spoken Forms](https://md.research.elamur.ai/free/en/papers/banski-2017-tei-lex0-forms/) - [↑ back to papers of 2017](https://md.research.elamur.ai/free/en/papers/2017/page/5/) ## References 3 references, 2 of them with a card of ours. ↗ leads to the work. 1. Adam Kilgarriff and Joseph Rosenzweig. 2000. Framework and results for english senseval. Computers and the Humanities 34(1):15–48. [↗](https://doi.org/10.1023/a:1002693207386) 2. Christopher D Manning, Mihai Surdeanu, John Bauer, Jenny Rose Finkel, Steven Bethard, and David McClosky. 2014. [The stanford corenlp natural language processing toolkit](https://md.research.elamur.ai/free/en/papers/manning-2014-stanford-corenlp-natural-language-processing/). In ACL (System Demonstrations). pages 55–60. F 1 0.6785 0.7900 0.8254 0.6828 0.7247 0.8439 0.8440 3. Tristan Miller, Christian F. Hempelmann, and Iryna Gurevych. 2017. [SemEval-2017 Task 7: Detection and interpretation of English puns](https://md.research.elamur.ai/free/en/papers/miller-2017-semeval-2017-task-7-detection/). In Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017). *Table 4: Performance of our systems and the top-ranked(ranked by P) systems. The numbers in the brackets are the official ranking* F 1 0.3373 0.6631 0.6523 0.5215 0.5681 0.7964 0.6845 0.6553 This research is supported by grants from Science and Technology Commission of Shanghai Municipality(14DZ2260800 and 15ZR1410700), Shanghai Collaborative Innovation Center of Trustworthy Software for Internet of Things (ZF1213) and NSFC (61402175). *Updated 1 November 2026*