A5074570184- Natural Language Processing Techniques
- Topic Modeling
- Syntax, Semantics, Linguistic Variation
- Multimodal Machine Learning Applications
- Text Readability and Simplification
- Speech and dialogue systems
- Language and cultural evolution
- Advanced Algebra and Logic
- Logic, Reasoning, and Knowledge
- Lexicography and Language Studies
- Linguistics and Discourse Analysis
- Advanced Text Analysis Techniques
- Authorship Attribution and Profiling
- Linguistic Variation and Morphology
- Linguistic and Sociocultural Studies
- Logic, programming, and type systems
- Gender Studies in Language
- Video Analysis and Summarization
- Speech Recognition and Synthesis
- Language, Linguistics, Cultural Analysis
- Phonetics and Phonology Research
- AI in cancer detection
- Semantic Web and Ontologies
- Categorization, perception, and language
- Linguistics and language evolution
University of Groningen
2023
Utrecht University
2018-2023
Language Science (South Korea)
2023
University of Helsinki
2023
Laboratoire Lorrain de Recherche en Informatique et ses Applications
2017-2020
University of Southern California
2020
Qinghai Normal University
2019
Minzu University of China
2019
Analyse et Traitement Informatique de la Langue Française
2019
University of Pisa
2018
Denis Paperno, Germán Kruszewski, Angeliki Lazaridou, Ngoc Quan Pham, Raffaella Bernardi, Sandro Pezzelle, Marco Baroni, Gemma Boleda, Raquel Fernández. Proceedings of the 54th Annual Meeting Association for Computational Linguistics (Volume 1: Long Papers). 2016.
Distributional semantic methods to approximate word meaning with context vectors have been very successful empirically, and the last years seen a surge of interest in their compositional extension phrases sentences. We present here new model that, like those Coecke et al. (2010) Baroni Zamparelli (2010), closely mimics standard Montagovian treatment composition distributional terms. However, our approach avoids number issues that prevented application earlier linguistically-motivated models...
Contextualized word embeddings, i.e. vector representations for words in context, are naturally seen as an extension of previous noncontextual distributional semantic models. In this work, we focus on BERT, a deep neural network that produces contextualized embeddings and has set the state-of-the-art several tasks, study coherence its embedding space. While showing tendency towards coherence, BERT does not fully live up to natural expectations particular, find position sentence which occurs,...
Corpus-based distributional semantic models capture degrees of relatedness among the words very large vocabularies, but have problems with logical phenomena such as entailment, that are instead elegantly handled by model-theoretic approaches, which, in turn, do not scale up. We combine advantages two views inducing a mapping from vectors (or sentences) into Boolean structure kind which natural language terms assumed to denote. evaluate this Distributional Semantic Model (BDSM) on recognizing...
Word embeddings have advanced the state of art in NLP across numerous tasks. Understanding contents dense neural representations is utmost interest to computational semantics community. We propose focus on relating these opaque word vectors with human-readable definitions, as found dictionaries This problem naturally divides into two subtasks: converting definitions embeddings, and definitions. task was conducted a multilingual setting, using comparable sets trained homogeneously.
This paper describes the SemEval 2018 Task 10 on Capturing Discriminative Attributes. Participants were asked to identify whether an attribute could help discriminate between two concepts. For example, a successful system should determine that ‘urine’ is discriminating feature in word pair ‘kidney’, ‘bone’. The aim of task better evaluate capabilities state art semantic models, beyond pure similarity. attracted submissions from 21 teams, and best achieved 0.75 F1 score.
Logical negation is a challenge for distributional semantics, because predicates and their negations tend to occur in very similar contexts, consequently vectors are similar. Indeed, it not even clear what properties “negated” vector should possess. However, when linguistic considered its actual discourse usage, often performs role that quite different from straightforward logical negation. If someone states, the middle of conversation, “This dog,” strongly suggests restricted set...
Distributional semantic models, deriving vector-based word representations from patterns of usage in corpora, have many useful applications (Turney and Pantel 2010 ). Recently, there has been interest compositional distributional which derive vectors for phrases their constituent words (Mitchell Lapata Often, the values are pointwise mutual information (PMI) scores obtained raw co-occurrence counts. In this article we study relation between PMI dimensions a phrase vector its components order...
Abstract Pretrained embeddings based on the Transformer architecture have taken NLP community by storm. We show that they can mathematically be reframed as a sum of vector factors and showcase how to use this reframing study impact each component. provide evidence multi-head attentions feed-forwards are not equally useful in all downstream applications, well quantitative overview effects finetuning overall embedding space. This approach allows us draw connections wide range previous studies,...
If lexical similarity is not enough to reliably assess how word vectors would perform on various specific tasks, we need other ways of evaluating semantic representations.We propose a new task, which consists in extracting differences using distributional models: given two words, what the difference between their meanings?We present proof concept datasets for this task and outline it may be performed.
Complex interactions among the meanings of words are important factors in function that maps word to phrase meanings.Recently, compositional distributional semantics models (CDSM) have been designed with goal emulating these complex interactions; however, experimental results on effectiveness CDSM difficult interpret because current metrics for assessing them do not control confound lexical information.We present a new method degree which capture semantic dissociates influences and then...
In this position paper we argue that an adequate semantic model must account for language in use, taking into how discourse context affects the meaning of words and larger linguistic units.Distributional models are very attractive mainly because they capture conceptual aspects automatically induced from natural data.However, need to be extended order use a or dialogue context.We discuss phenomena new generation distributional should capture, propose concrete tasks on which could tested.