IN SEARCH OF THE BEAUTY AND THE BEAST: ANTONYMY IN SYNTACTIC CONSTRUCTIONS

HDPL Osijek 2021: XXXV. međunarodni znanstveni skup / 35th International Conference

JEZIK U DIGITALNOM OKRUŽENJU / LANGUAGE IN THE DIGITAL ENVIRONMENT

​

dr. sc. Benedikt Perak, docent,

Odsjek za kulturalne studije, Filozofski fakultet Sveučilišta u Rijeci

dr. sc. Janja Čulig Suknaić, postdoc,

Odsjek za anglistiku, Filozofski fakultet Sveučilišta u Zagrebu

​

OVERVIEW

1. What is ‘antonymy’?
2. How does antonymy help us understand emotions and sentiment from a lexical/semantic perspective?
3. What is our research method?
4. Did it work?
5. What can we do in the future?

​

WHAT IS ANTONYMY

• „Antonymy is the standard technical term for oppositeness of meaning between lexemes”. (Lyons, 1977)
• „Oppositeness is perhaps the only sense relation to receive direct lexical recognition in everyday language.” (Cruse, 2004)
• „Unlike synonymy, everyone agrees that antonymy exists, and it is robustly evident in natural language. Unlike hyponymy and meronymy, it can be as much a relation among words as it is a relation among concepts or denotata” (Murphy 2003)
• „Antonymy is unique among lexical semantic relations in that it requires ono-to-one relations, rather than one-to-many or many-to-many” (Jones et al. 2012)
• The contemporary view of ANT (Jones 2002, Murphy 2003, Paradis 2011, Jones et al. 2012) suggests that:
• Oppositeness is created primarily on the conceptual level – two semantically opposed concepts
• Oppositeness is based on the process of construal – we establish a boundary between concepts
• E.g. love-hate – speakers determine where the boundary is drawn based on their experience and conceptual knowledge

​

• ANT is a conceptual prototype-based category of semantic opposition (Čulig Suknaić 2020)

​

CONCEPTUAL-SEMANTIC UNITS (POJMOVNO/ZNAČENJSKE CJELINE)

• Regardless of the type of speech or logical relation, each pair of ANT has the same fundamental structure: two semantic poles divided by an established boundary.
• The knowledge about each pole depends on the knowledge about the unit as a whole, and vice versa.

happy

sad

Paradigmatic structure!

ANTONYMYC CONSTRUCTIONS (ANTONIMIJSKE KONSTRUKCIJE)

• Specific constructions that hold ANT in language use = syntagmatic structure:
• E. g. ‘X and Y’ / ‘X i Y’, ‘X or Y’ / ‘X ili Y’, ‘more X than Y’ / ‘više X nego Y’, ‘neither X nor Y’ / ‘niti X niti Y’
• The schematic form of ANT constructions is embedded in the speaker’s conceptual knowledge about opposition, which includes conceptual-semantic units – therefore, it belongs to the category of ANT
• Linguistic expressions of C-S units are based on, and work together with, ANT constructions:
• E. g. ‘love and hate’ / ‘ljubav i mržnja’, ‘love or hate’ / ‘ljubav ili mržnja’

SO, WHAT CAN WE DO WITH THIS KIND OF ANTONYMY?

• Antonymy is a cognitive mechanism for categorizing knowledge.
• It offers insight into the semantic and syntactic structure of language (ENG and CRO)
• Through ANT constructions (‘X and/or Y’) we can get to algorithms for extracting conceptually similar lexemes (~synonymy)
• SYN lead to ANT expansion
• assigning psychological hedonic valency values to lexemes in a sense cluster can be used to further distinguish antonymic lexemes based on the affective dimension
• This provides access to possible overarching patterns of antonymy, especially emotion and sentiment conceptualization!

RESEARCH METHODS FOR IDENTIFYING ANTONYMY

1. Using the semantic knowledge database (WordNet)
2. Extending the semantic knowledge database with the coordination collocates (Wordnet + ConGraCNet)
3. Filtering constructions from corpora using NLP morphosyntactic tags (NLP tags for antonym constructions)
4. Using prompts in Transformers model to elicit appropriate tokens using <mask> (Transformer models: Bert...)��
5. Filtering antonyms from a set of coordination candidates using lexical sentiment values (2 + Sentiment dictionary)

SYMBOLIC

SUBSYMBOLIC

syntagmatic

syntagmatic - paradigmatic

paradigmatic - syntagmatic

paradigmatic

syntagmatic - paradigmatic

1) Antonyms from the semantic knowledge database (WordNet)

Advantages:

- curated

- multilingual (34 languages, http://compling.hss.ntu.edu.sg/omw/)

​

​

Weakness:

• sparse:

hrWac lemmas that have antonyms: # n 1240,

# a 1538, 5/120 = 4%

# v 1492

http://compling.hss.ntu.edu.sg/omw/

Hr: Raffaelli, Ida; Bekavac, Božo; Agić, Željko; Tadić, Marko. (2008), ; Oliver A., Šojat, K., Srebačić, M. (2015)

1) Antonyms from the semantic knowledge database (WordNet)

WORDNET IMPLEMENTATION IN ConGraCNet

http://emocnet.uniri.hr/congracnet2/

1) Antonyms from the semantic knowledge database (WordNet)

WORDNET IMPLEMENTATION IN ConGraCNet

http://emocnet.uniri.hr/congracnet2/

WORDNET IMPLEMENTATION IN ConGraCNet

antonyms belong to same category

II) Extending the semantic knowledge database with the coordination collocates (Wordnet + ConGraCNet)

SOURCE LEXEME = ljepota - n HAS ANTONYMS [y1…..yx] WordNET

each lexical node in a SOURCE LEXEME FoF coordination graph is conceptually similar to SOURCE node ConGraCNET

therefore:

each lexical node in a SOURCE LEXEME FoF coordination graph can be considered antonymous to ANTONYMS [y1…..yx] set

II) Extending the semantic knowledge database with the coordination collocates (Wordnet + ConGraCNet)

SOURCE LEXEME = ljepota - n HAS ANTONYMS [y1…..yx] WordNET

each lexical node in a SOURCE LEXEME FoF coordination graph is conceptually similar to SOURCE node ConGraCNET

therefore:

each lexical node in a SOURCE LEXEME FoF coordination graph can be considered antonymous to ANTONYMS [y1…..yx] set

II) Extending the semantic knowledge database with the coordination collocates (Wordnet + ConGraCNet)

for each lexeme l in corpus create a dataframe where: each x in l+fof has antonym a for each a in l-antonyms

​

III) Filtering constructions from corpora using NLP morphosyntactic tags (NLP tags for antonym constructions)

HrEngRI

• focus shifts from paradimatic to syntagmatic

Using prompts in Transformers model to elicit appropriate tokens using <mask> (Transformer models: Bert...)

Filtering antonyms from a set of coordination candidates using lexical sentiment values (2 + Sentiment dictionary)

• ​
• ​

SYMBOLIC

SUBSYMBOLIC

DID IT WORK?

HOW DOES ANT HELP US UNDERSTAND EMOTIONS AND SENTIMENT FROM A LEXICAL/SEMANTIC PERSPECTIVE?

• ANT helps us reveal the connection and interrelatedness of semantic domains for emotions because it encompasses the entire spectrum (both poles of the C-S unit)!

​

WHAT CAN WE DO IN THE FUTURE?

We are grateful for your attention!

We are ungrateful for your attention!

BIBLIOGRAPHY

​

• Čulig Suknaić, Janja (2020). Antonymy as a conceptual category of semantic opposition in English and Croatian (doctoral dissertation)
• Goldberg, Adele (2006). Constructions at work: the nature of generalization in language. Oxford: oxford University Press
• Jones, Steven, M. Lynn Murphy, Carita Paradis, Caroline Willners (2012). Antonyms in English. Cambridge: Cambridge University Press
• Lakoff, George (1987). Women, Fire and Dangerous Things: What Categories Reveal about the Mind. Chicago - London: The University of Chicago Press
• Lüdtke, Ulrike M. (ed.) (2015). Emotion in Language. Amsterdam - Philadelphia: John Benjamins Publishing Company
• Murphy, M. Lynn (2006). “Antonyms as lexical constructions: or, why paradigmatic construction is not an oxymoron”. In: Constructions, 1-8
• Paradis, Carita (2011). “A dynamic construal approach to antonymy”. In: Selected papers from the 19th International Symposium on Theoretical and Applied Linguistics. 33-42
• Perak, B., & Ban Kirigin, T. (2020). Corpus-Based Syntactic-Semantic Graph Analysis: Semantic Domains of the Concept Feeling. Rasprave: Časopis Instituta za hrvatski jezik i jezikoslovlje, 46(2), 493-532.
• Oliver A., Šojat, K., Srebačić, M. (2015) Automatic Expansion of Croatian Wordnet In Proceedings of the 29th CALS international conference: Applied Linguistic Research and Methodology Zadar (Croatia)
• Raffaelli, Ida; Bekavac, Božo; Agić, Željko; Tadić, Marko. (2008) Building Croatian WordNet. In Proceedings of the Fourth Global WordNet Conference pp349-359
• ​

​