Hierarchical Embeddings for Hypernymy Detection and Directionality

Nguyen, Kim Anh; Köper, Maximilian; Walde, Sabine Schulte im; Vu, Ngoc Thang

doi:10.18653/v1/d17-1022

Cited by 65 publications

(31 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The recent trend of learning efficient representations for lexical entailment has moved to supervised learning. In particular, pre-trained word embeddings are retrained to distinguish a hypernymy relation from other relations (Vulić and Mrkšić, 2018;Nguyen et al, 2017;Alsuhaibani et al, 2019). Hierarchical structures defined in taxonomies and ontologies (e.g., WordNet (Miller, 1995)) are commonly used for the retraining (Nguyen et al, 2017;Alsuhaibani et al, 2019 the loss functions.…”

Section: Supervised Representation Methodsmentioning

confidence: 99%

“…We used the two thresholds introduced in Section 3.3.3 when evaluating on the WBLESS and BIB-LESS datasets. Following Vulić and Mrkšić (2018) and Nguyen et al (2017), we tuned the thresholds with 2% randomly chosen from the datasets and evaluated our method on the remaining 98%. We repeated this procedure 1,000 times and report the average accuracy.…”

Section: Task Overviewmentioning

confidence: 99%

See 1 more Smart Citation

Visual Objects As Context: Exploiting Visual Objects for Lexical Entailment

Muraoka

Nasukawa

Bhattacharjee

2020

Findings of the Association for Computational Linguistics: EMNLP 2020

View full text Add to dashboard Cite

We propose a new word representation method derived from visual objects in associated images to tackle the lexical entailment task. Although it has been shown that the Distributional Informativeness Hypothesis (DIH) holds on text, in which the DIH assumes that a context surrounding a hyponym is more informative than that of a hypernym, it has never been tested on visual objects. Since our perception is tightly associated with language, it is meaningful to explore whether the DIH holds on visual objects. To this end, we consider visual objects as the context of a word and represent a word as a bag of visual objects found in images associated with the word. This allows us to test the feasibility of the visual DIH. To better distinguish word pairs in a hypernym relation from other relations such as co-hypernyms, we also propose a new measurable function that takes into account both the difference in the generality of meaning and similarity of meaning between words. Our experimental results show that the DIH holds on visual objects and that the proposed method combined with the proposed function outperforms existing unsupervised representation methods. * Note that the number of search trials differs among the datasets because hrel(x, y), the function for detecting word pairs in a hypernym relation, is applied only to the WBLESS and BIBLESS datasets.

show abstract

Section: Supervised Representation Methodsmentioning

confidence: 99%

Section: Task Overviewmentioning

confidence: 99%

Visual Objects As Context: Exploiting Visual Objects for Lexical Entailment

Muraoka

Nasukawa

Bhattacharjee

2020

Findings of the Association for Computational Linguistics: EMNLP 2020

View full text Add to dashboard Cite

show abstract

“…Joint learning models. Many joint learning models introduce semantic lexicons or linguistic structures as additional constraints to the representation learning models (Liu et al 2015;Ono, Miwa, and Sasaki 2015;Nguyen et al 2017). Yu and Dredze (2014) integrate word2vec (Mikolov et al 2013) with synonym constraints that the distances between synonym representations shall be close.…”

Section: Word Representation Specializationmentioning

confidence: 99%

“…Both post-processing and joint learning models require high-quality, manually constructed, semantic lexicons or linguistic structures (Nguyen et al 2017;Liu et al 2018;Mrkšić et al 2016;Glavaš and Vulić 2018). It should be noted that structured data on the web contains rich semantic information, which has the advantage that it can be easily extracted (Zhang et al 2013).…”

Section: Semantic Knowledge Constructionmentioning

confidence: 99%

Leveraging Web Semantic Knowledge in Word Representation Learning

Liu

Fang

Lou

et al. 2019

AAAI

View full text Add to dashboard Cite

Much recent work focuses on leveraging semantic lexicons like WordNet to enhance word representation learning (WRL) and achieves promising performance on many NLP tasks. However, most existing methods might have limitations because they require high-quality, manually created, semantic lexicons or linguistic structures. In this paper, we propose to leverage semantic knowledge automatically mined from web structured data to enhance WRL. We first construct a semantic similarity graph, which is referred as semantic knowledge, based on a large collection of semantic lists extracted from the web using several pre-defined HTML tag patterns. Then we introduce an efficient joint word representation learning model to capture semantics from both semantic knowledge and text corpora. Compared with recent work on improving WRL with semantic resources, our approach is more general, and can be easily scaled with no additional effort. Extensive experimental results show that our approach outperforms the state-of-the-art methods on word similarity, word sense disambiguation, text classification and textual similarity tasks.

show abstract

“…A generalization of that second approach may be more promising. Perhaps different features are involved in a feature comparison process depending upon the particular semantic discrimination required in a particular condition 1 (e.g., Baroni et al, 2012;Lenci and Benotto, 2012;Roller et al, 2014;Weeds et al, 2014;Nguyen et al, 2016Nguyen et al, , 2017aShwartz et al, 2016). More specifically, in the Anomalous condition, as is the case for more traditional feature comparison theories, an overall similarity comparison is sufficient to distinguish true from false stimuli.…”

Section: Introductionmentioning

confidence: 99%

The Representation of Coordinate Relations in Lexical Semantic Memory

Gruenenfelder

2020

Front. Psychol.

View full text Add to dashboard Cite

Two experiments examined the size of the typicality effect for true items in a category verification task as a function of the type of false item used. In Experiment 1, compared to the case where false items paired unrelated concepts ("carrot-vehicle"), the typicality effect was much larger when false items paired an exemplar with a category coordinate to its proper category ("carrot-fruit"). In contrast, when false items paired coordinate concepts ("carrot-pea") or reversed the ordering of subject and predicate terms ("All vegetables are carrots"), the typicality effect did not change in size. Further, the time to verify true sentences did not increase monotonically with the semantic similarity of the two terms used in false sentences. Experiment 2 showed that the pattern of results for coordinate items reflected semantic processing, not simply task difficulty. A combined analysis examined data across multiple experiments, increasing the power of the statistical analysis. The size of the typicality effect when coordinate false items were used was again the same as when false items paired unrelated concepts. The most straightforward explanation of this pattern of results seems to be in terms of a sparse semantic network model of lexical semantic memory, in which labeled links are used to indicate the semantic relation that exists between pairs of words.

show abstract

Hierarchical Embeddings for Hypernymy Detection and Directionality

Cited by 65 publications

References 31 publications

Visual Objects As Context: Exploiting Visual Objects for Lexical Entailment

Visual Objects As Context: Exploiting Visual Objects for Lexical Entailment

Leveraging Web Semantic Knowledge in Word Representation Learning

The Representation of Coordinate Relations in Lexical Semantic Memory

Contact Info

Product

Resources

About