TL;DR
This paper tackles the challenge of limited semantic resources for French by extending the WOLF (WordNet Libre du Français) lexicon with adverbial information. They then couple this enriched semantic resource with the Lefff syntactic lexicon, creating a more comprehensive linguistic tool for French NLP applications.
What Problem Does This Paper Address?
Despite advancements in morphological and syntactic resources for French, high-quality semantic resources remain scarce, especially when compared to English. This gap hinders the development of sophisticated Natural Language Processing (NLP) applications. Specifically, the authors identify two key issues: 1) the WOLF, a valuable French WordNet, has limited coverage for adverbs, and 2) there's a lack of coupling between existing syntactic and semantic resources. This paper aims to address these limitations by focusing on French adverbs, a morphologically rich and open class, to enhance the WOLF and bridge the gap with the Lefff syntactic lexicon.
Methods
The authors employ two primary techniques to extend the WOLF's adverbial coverage:
- Extension by Derivation: This method leverages the morphological relationship between adjectives and adverbs. They exploit the "derived" relation in the Princeton WordNet, which links adverbial synsets to adjectival synsets. By applying French derivational rules (adding "-ment" to the feminine singular of adjectives), they generate new adverbial lexemes. These generated adverbs are then checked against the Lefff lexicon, and if found, are added to the corresponding WOLF synset.
- Extension by Synonymy: This technique utilizes the DicoSyn database, a French synonym dictionary. They construct an adverbial synonym graph from DicoSyn, specifically focusing on adverbs related to the "-ment" suffix. Using a concept called "cliques" (fully connected subgraphs) within this synonym graph, they identify groups of synonymous adverbs. These cliques are then used to add new adverbs to existing WOLF synsets, thereby increasing the number of lexemes per synset.
Datasets
- WOLF (WordNet Libre du Français): The primary semantic resource used and extended in this paper.
- Lefff (Lexique des Formes Fléchies du Français): A comprehensive French morphological and syntactic lexicon used for validation and coupling.
- DicoSyn: A French electronic synonym dictionary used to build the adverbial synonym graph.
- Princeton WordNet: The English WordNet used as a basis for the WOLF and for identifying derivational relationships.
- Molinier & Levrier (2000) tables: Lexicon-grammar tables for "-ment" adverbs, used to inform the Lefff and the coupling process.
Key Insights and Ideas
- Leveraging Derivation for Semantic Expansion: The paper demonstrates that the morphological relationship between adjectives and adverbs, which is often parallel across languages, can be effectively exploited to expand semantic lexicons like WordNet. This is a cost-effective way to increase coverage.
- Synonymy Graphs as a Source of Semantic Enrichment: The use of DicoSyn and the concept of "cliques" to identify groups of synonyms provides a structured way to add more lexemes to existing semantic concepts, improving the richness of the lexicon.
- The Importance of Lexicon Coupling: The work highlights that combining syntactic and semantic resources (Lefff and WOLF) is crucial for building more robust NLP systems. This coupling allows for richer annotations and more accurate linguistic analysis.
- Adverbs as a Rich Domain for Lexical Research: The paper's focus on adverbs, particularly those ending in "-ment," showcases their morphological regularity and semantic interconnectedness, making them a good testbed for lexical resource development.
Key Results
- The extension by derivation increased the number of adverbial lexeme-synset relations in WOLF by 56% and the number of non-empty adverbial synsets by 43%.
- The extension by synonymy further increased the number of adverbial lexeme-synset relations by 28.5%.
- A manual evaluation of a significant portion of the extended WOLF showed that approximately 68% of the lexeme-synset associations were correct (OK).
- The coupling process successfully assigned synset identifiers from the WOLF to numerous adverbial entries in the Lefff.
Limitations and Future Work
- Limitations:
- The derivation process relies on a heuristic for adjective-adverb conversion, which is not always perfect (e.g., "courante/couramment").
- DicoSyn does not explicitly provide part-of-speech information, requiring additional steps to construct adverb-specific graphs.
- The manual evaluation covered only 20% of the extended WOLF.
- Future Work:
- Explore other derivational relationships and semantic relations (e.g., hyperonymy, antonymy).
- Extend the coupling to other parts of speech, such as nouns ending in "-ité" and verbs ending in "-ifier" and "-iser."
- Investigate the use of corpus analysis for syntagmatic relations.
- Further explore the potential of graph-based lexical resources for automatic semantic information access.
Summary updated