Modeling Out-of-Vocabulary Words via Grammatical Fusion

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F26%3ABDTRMP6E" target="_blank" >RIV/00216208:11320/26:BDTRMP6E - isvavai.cz</a>

Výsledek na webu

<a href="http://dx.doi.org/10.1007/978-3-031-92471-2_3" target="_blank" >http://dx.doi.org/10.1007/978-3-031-92471-2_3</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/978-3-031-92471-2_3" target="_blank" >10.1007/978-3-031-92471-2_3</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Modeling Out-of-Vocabulary Words via Grammatical Fusion

Popis výsledku v původním jazyce

Text embedding is a crucial element in computational linguistics, allowing computers to represent human language semantics numerically. A key approach is the Vector Space Model (VSM), which maps text into high-dimensional vectors and supports tasks like classification, clustering, and semantic search. Building on this, techniques like word embeddings (e.g., word2vec) and document embeddings have advanced Natural Language Processing (NLP), enabling sentiment analysis, machine translation, and summarization. However, challenges persist in morphologically rich languages (MRLs) such as Arabic, Hebrew, Russian, and Turkish. These languages feature complex structures with prefixes, suffixes, and roots, complicating text representation and increasing out-of-vocabulary (OOV) words. Addressing these challenges is essential for improving NLP’s applicability across diverse linguistic contexts. This study proposes a novel methodology to overcome existing limitations by utilizing probabilistic corpus decomposition. The corpus is processed in three forms: words are represented as wholes, decomposed into grammatical components, or broken down into letters. The Word2Vec algorithm is applied to this “triple corpus”, generating vectors for words, their components, and letters, ensuring consistent semantic alignment across all forms. Experiments with English and Hebrew corpora demonstrate the effectiveness of this approach. The fused vectors of grammatical components achieve high-quality representations, often ranking first in cosine similarity, thereby reducing out-of-vocabulary (OOV) issues and expanding the lexicon. By improving the representation of morphological structures, this methodology enhances embeddings for both English and Semitic languages, advancing NLP capabilities in a variety of linguistic contexts. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Název v anglickém jazyce

Modeling Out-of-Vocabulary Words via Grammatical Fusion

Popis výsledku anglicky

Text embedding is a crucial element in computational linguistics, allowing computers to represent human language semantics numerically. A key approach is the Vector Space Model (VSM), which maps text into high-dimensional vectors and supports tasks like classification, clustering, and semantic search. Building on this, techniques like word embeddings (e.g., word2vec) and document embeddings have advanced Natural Language Processing (NLP), enabling sentiment analysis, machine translation, and summarization. However, challenges persist in morphologically rich languages (MRLs) such as Arabic, Hebrew, Russian, and Turkish. These languages feature complex structures with prefixes, suffixes, and roots, complicating text representation and increasing out-of-vocabulary (OOV) words. Addressing these challenges is essential for improving NLP’s applicability across diverse linguistic contexts. This study proposes a novel methodology to overcome existing limitations by utilizing probabilistic corpus decomposition. The corpus is processed in three forms: words are represented as wholes, decomposed into grammatical components, or broken down into letters. The Word2Vec algorithm is applied to this “triple corpus”, generating vectors for words, their components, and letters, ensuring consistent semantic alignment across all forms. Experiments with English and Hebrew corpora demonstrate the effectiveness of this approach. The fused vectors of grammatical components achieve high-quality representations, often ranking first in cosine similarity, thereby reducing out-of-vocabulary (OOV) issues and expanding the lexicon. By improving the representation of morphological structures, this methodology enhances embeddings for both English and Semitic languages, advancing NLP capabilities in a variety of linguistic contexts. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

—

Návaznosti

—

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Lect. Notes Bus. Inf. Process.

ISBN

978-3-031-92470-5

ISSN

—

e-ISSN

—

Počet stran výsledku

14

Strana od-do

38-51

Název nakladatele

Springer Science and Business Media Deutschland GmbH

Místo vydání

—

Místo konání akce

Seville

Datum konání akce

1. 1. 2026

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—