The polysemy effect across inflected word forms
Apstrakt
Polysemes are ambiguous words with multiple related senses (PAPER – ‘writing paper’ and ‘scientific paper’; Rodd et al., 2002). Previous research established that polyseme processing is also affected by sense uncertainty (Filipović Đurđević & Kostić, 2017). However, Serbian words appear in multiple inflected forms where each form has multiple syntactic meanings and functions and are therefore also ambiguous. The aim of this paper was twofold. The first was to explore the uncertainty effect throughout the inflected forms. We tested whether inflected word forms are equally ambiguous, considering that syntactic properties of a sentence can affect word meaning (Kostić et al., 2003). The second aim was to explore whether the potential (in)consistency of the polysemy effect could be a consequence of the dynamics of mapping cues to outcomes in a discriminative learning (DL) network.
We collected lexical decision data and compared the predictive power of information-theoretic (IT) measures of... uncertainty and the measures derived from the DL model (Baayen et al., 2011). We presented 35 polysemous nouns of masculine gender to 124 participants. Each noun was presented in seven inflected forms in a Latin-square design. Reaction times were predicted by two sets of variables – entropy of the sense frequency distribution (Filipović Đurđević & Kostić, 2017). DL predictors were derived from two cue-outcome weight matrices (Milin et al., 2017). Grapheme-to-lexome (G2L) matrix was calculated from a network mapping trigraphs to lemmata. Lexome-to-lexome (L2L) matrix from a network mapping 1000 co-occurring context words as cues to lemmata.
Diversity measures from G2L and L2L (r = .36, p < .001; r = .25, p < .001), and G2L prior measure (r = .27, p < .001) correlated with entropy. We modelled lexical decision latencies with GAMMs (Wood, 2006) and compared models with entropy and DL predictors interacting with the inflected form. The entropy model revealed differing entropy effects across word forms, with marginal significance in the nominative form. DL model revealed a similar variation in DL predictor effect but had a better fit than the IT model.
Inconsistent entropy effects across word forms suggest that there might be an interaction between semantics and syntax. Correlations between DL and entropy measures reveal that discrimination of orthographic cues might be a process behind this observed complexity. This research opens the question of the need for estimating ambiguity separately across different inflected forms of the same lemmata.
Ključne reči:
polysemy / word form / entropy / naive discrimination learningIzvor:
Book of Abstracts, XXVII Scientific Conference Empirical Studies in Psychology, May 13-16, Faculty of Philosophy, University of Belgrade, 2021, 37-38Izdavač:
- Institute for Psychology and Laboratory for Experimental Psychology, Faculty of Philosophy in Belgrade
Finansiranje / projekti:
- Ministarstvo nauke, tehnološkog razvoja i inovacija Republike Srbije, institucionalno finansiranje - 200163 (Univerzitet u Beogradu, Filozofski fakultet) (RS-MESTD-inst-2020-200163)
Institucija/grupa
Psihologija / PsychologyTY - CONF AU - Mišić, Ksenija AU - Filipović Đurđević, Dušica PY - 2021 UR - http://reff.f.bg.ac.rs/handle/123456789/5142 AB - Polysemes are ambiguous words with multiple related senses (PAPER – ‘writing paper’ and ‘scientific paper’; Rodd et al., 2002). Previous research established that polyseme processing is also affected by sense uncertainty (Filipović Đurđević & Kostić, 2017). However, Serbian words appear in multiple inflected forms where each form has multiple syntactic meanings and functions and are therefore also ambiguous. The aim of this paper was twofold. The first was to explore the uncertainty effect throughout the inflected forms. We tested whether inflected word forms are equally ambiguous, considering that syntactic properties of a sentence can affect word meaning (Kostić et al., 2003). The second aim was to explore whether the potential (in)consistency of the polysemy effect could be a consequence of the dynamics of mapping cues to outcomes in a discriminative learning (DL) network. We collected lexical decision data and compared the predictive power of information-theoretic (IT) measures of uncertainty and the measures derived from the DL model (Baayen et al., 2011). We presented 35 polysemous nouns of masculine gender to 124 participants. Each noun was presented in seven inflected forms in a Latin-square design. Reaction times were predicted by two sets of variables – entropy of the sense frequency distribution (Filipović Đurđević & Kostić, 2017). DL predictors were derived from two cue-outcome weight matrices (Milin et al., 2017). Grapheme-to-lexome (G2L) matrix was calculated from a network mapping trigraphs to lemmata. Lexome-to-lexome (L2L) matrix from a network mapping 1000 co-occurring context words as cues to lemmata. Diversity measures from G2L and L2L (r = .36, p < .001; r = .25, p < .001), and G2L prior measure (r = .27, p < .001) correlated with entropy. We modelled lexical decision latencies with GAMMs (Wood, 2006) and compared models with entropy and DL predictors interacting with the inflected form. The entropy model revealed differing entropy effects across word forms, with marginal significance in the nominative form. DL model revealed a similar variation in DL predictor effect but had a better fit than the IT model. Inconsistent entropy effects across word forms suggest that there might be an interaction between semantics and syntax. Correlations between DL and entropy measures reveal that discrimination of orthographic cues might be a process behind this observed complexity. This research opens the question of the need for estimating ambiguity separately across different inflected forms of the same lemmata. PB - Institute for Psychology and Laboratory for Experimental Psychology, Faculty of Philosophy in Belgrade C3 - Book of Abstracts, XXVII Scientific Conference Empirical Studies in Psychology, May 13-16, Faculty of Philosophy, University of Belgrade T1 - The polysemy effect across inflected word forms EP - 38 SP - 37 UR - https://hdl.handle.net/21.15107/rcub_reff_5142 ER -
@conference{ author = "Mišić, Ksenija and Filipović Đurđević, Dušica", year = "2021", abstract = "Polysemes are ambiguous words with multiple related senses (PAPER – ‘writing paper’ and ‘scientific paper’; Rodd et al., 2002). Previous research established that polyseme processing is also affected by sense uncertainty (Filipović Đurđević & Kostić, 2017). However, Serbian words appear in multiple inflected forms where each form has multiple syntactic meanings and functions and are therefore also ambiguous. The aim of this paper was twofold. The first was to explore the uncertainty effect throughout the inflected forms. We tested whether inflected word forms are equally ambiguous, considering that syntactic properties of a sentence can affect word meaning (Kostić et al., 2003). The second aim was to explore whether the potential (in)consistency of the polysemy effect could be a consequence of the dynamics of mapping cues to outcomes in a discriminative learning (DL) network. We collected lexical decision data and compared the predictive power of information-theoretic (IT) measures of uncertainty and the measures derived from the DL model (Baayen et al., 2011). We presented 35 polysemous nouns of masculine gender to 124 participants. Each noun was presented in seven inflected forms in a Latin-square design. Reaction times were predicted by two sets of variables – entropy of the sense frequency distribution (Filipović Đurđević & Kostić, 2017). DL predictors were derived from two cue-outcome weight matrices (Milin et al., 2017). Grapheme-to-lexome (G2L) matrix was calculated from a network mapping trigraphs to lemmata. Lexome-to-lexome (L2L) matrix from a network mapping 1000 co-occurring context words as cues to lemmata. Diversity measures from G2L and L2L (r = .36, p < .001; r = .25, p < .001), and G2L prior measure (r = .27, p < .001) correlated with entropy. We modelled lexical decision latencies with GAMMs (Wood, 2006) and compared models with entropy and DL predictors interacting with the inflected form. The entropy model revealed differing entropy effects across word forms, with marginal significance in the nominative form. DL model revealed a similar variation in DL predictor effect but had a better fit than the IT model. Inconsistent entropy effects across word forms suggest that there might be an interaction between semantics and syntax. Correlations between DL and entropy measures reveal that discrimination of orthographic cues might be a process behind this observed complexity. This research opens the question of the need for estimating ambiguity separately across different inflected forms of the same lemmata.", publisher = "Institute for Psychology and Laboratory for Experimental Psychology, Faculty of Philosophy in Belgrade", journal = "Book of Abstracts, XXVII Scientific Conference Empirical Studies in Psychology, May 13-16, Faculty of Philosophy, University of Belgrade", title = "The polysemy effect across inflected word forms", pages = "38-37", url = "https://hdl.handle.net/21.15107/rcub_reff_5142" }
Mišić, K.,& Filipović Đurđević, D.. (2021). The polysemy effect across inflected word forms. in Book of Abstracts, XXVII Scientific Conference Empirical Studies in Psychology, May 13-16, Faculty of Philosophy, University of Belgrade Institute for Psychology and Laboratory for Experimental Psychology, Faculty of Philosophy in Belgrade., 37-38. https://hdl.handle.net/21.15107/rcub_reff_5142
Mišić K, Filipović Đurđević D. The polysemy effect across inflected word forms. in Book of Abstracts, XXVII Scientific Conference Empirical Studies in Psychology, May 13-16, Faculty of Philosophy, University of Belgrade. 2021;:37-38. https://hdl.handle.net/21.15107/rcub_reff_5142 .
Mišić, Ksenija, Filipović Đurđević, Dušica, "The polysemy effect across inflected word forms" in Book of Abstracts, XXVII Scientific Conference Empirical Studies in Psychology, May 13-16, Faculty of Philosophy, University of Belgrade (2021):37-38, https://hdl.handle.net/21.15107/rcub_reff_5142 .