Semantic ambiguitythe role of number of meanings and relatedness of meanings in word processing

  1. HARO RODRÍGUEZ, JUAN
Supervised by:
  1. Pilar Ferré Romeu Director

Defence university: Universitat Rovira i Virgili

Fecha de defensa: 15 June 2018

Committee:
  1. José Eugenio García-Albea Ristol Chair
  2. Isabel Fraga Carou Secretary
  3. Montserrat Comesaña Vila Committee member

Type: Thesis

Abstract

Semantic ambiguity is a key feature of language, as many of the words that we read or hear every day have more than one meaning (i.e., ambiguous words; e.g., banco, muñeca, pico, etc.). Given the prevalence of this phenomenon, any complete theory of language comprehension must account for how these words are processed and represented. For example, it should be able to provide answers to questions such as: What happens when someone reads a string of letters that refers to more than one meaning? How is represented in the mind the one-to-many mapping between the orthographic representation and the semantic representations of an ambiguous word? Taking that into account, the aim of the present doctoral thesis was to address some unresolved questions about how ambiguous words are processed in isolation. There is a large amount of evidence showing that ambiguous words are recognized faster than unambiguous words in the lexical decision task (LDT) (e.g., Azuma & Van Orden, 1997; Borowsky & Masson, 1996; Hino & Lupker, 1996; Jastrzembski, 1981; Lin & Ahrens, 2010; Rubenstein, Garfield, & Milikan, 1970). The explanation for this so-called ambiguity advantage is that ambiguous words benefit from having as many semantic representations as meanings. However, some studies conducted during the last two decades obtained results that are incompatible with this explanation, as they found that number of meanings (NOM) does not facilitate word recognition, but rather inhibits it (Armstrong & Plaut, 2008; Rodd, Gaskell, & Marslen-Wilson, 2002). In addition, these studies suggest that relatedness of meanings (ROM) has a significant effect in ambiguous word recognition. Namely, they observed a disadvantage for words with unrelated meanings (i.e., homonyms; e.g., banco), along with an advantage for words with related meanings (i.e., polysemes; e.g., bueno), with respect to unambiguous words. Nevertheless, some other studies reported a similar facilitation for homonyms and polysemes in comparison to unambiguous words (e.g., Hino, Kusunose, & Lupker, 2010; Pexman, Hino, & Lupker, 2004), indicating that NOM, regardless of ROM, facilitates word recognition. One of the main objectives of the present thesis was to shed some light regarding the inconsistencies about the effect of NOM and ROM in word recognition. Our working hypothesis was that the conflicting results regarding NOM and ROM found in the literature could be due to methodological differences between studies. This is supported by the fact that all the studies that reported a NOM disadvantage used number of dictionary meanings as a NOM measure (e.g., Armstrong & 2008; Rodd et al., 2002), while those that observed a NOM advantage employed NOM ratings provided by participants (e.g., Hino et al., 2010; Pexman et al., 2004). The hypothesis was addressed in a series of LDT experiments. But, before doing so, we collected several objective and subjective measures of semantic ambiguity for a large set of words (Study 1). These measures were included in a database of ambiguous and unambiguous Spanish words made up of 530 words. The data obtained in the Study 1 allowed us to conduct the Study 2, where we manipulated the measure employed for NOM estimation in a series of LDT experiments. The results showed that ambiguous words were recognized slower and with more errors than unambiguous words when the measure for NOM estimation was based on dictionary meanings (Experiment 1). In contrast, we observed a facilitation for ambiguous words when the measure was based on NOM ratings provided by participants (Experiments 2 and 3). Hence, the findings of the Study 2 showed that the method for NOM estimation influences ambiguous word recognition. This could be because there is not a direct match between the words that have one/many entries in the dictionary and the words that have one/multiple meanings for the speakers. Accordingly, NOM seems to facilitate word recognition but only when subjective NOM measures are used. In the Study 3 we examined the effect of ROM in word recognition. This was addressed in an LDT experiment in which we compared polysemes and homonyms. Subjective measures of ROM and NOM were used to select the experimental stimuli. In addition to behavioral measures (i.e., RTs and error rates), we also registered EEG data (i.e., ERPs); in particular, we focused on an ERP component associated with semantic processing (i.e., N400). The results showed a similar facilitation for polysemes and homonyms with respect to unambiguous words. Moreover, no differences were observed between polysemes and homonyms regarding N400 amplitudes. Therefore, the lack of a ROM effect in both behavioral and EEG data suggested that ROM does not influence word processing. So far, the results of the Studies 2 and 3 indicated that NOM, regardless of ROM, facilitates word recognition. With this in mind, the following objective was to determine the mechanism by which NOM facilitates recognition. To do so, we tested the predictions of the main models that account for the ambiguity advantage: the model of Kawamoto, Farrar, and Kello (1994), the model of Borowsky and Masson (1996) and the semantic feedback account (Balota, Ferraro, & Connoer, 1991; Hino & Lupker, 1996). First, we compared the model of Kawamoto et al. (1994) with respect the other two. The main difference between them is that Kawamoto et al. assume that the one-to-many inconsistency between orthography and semantics for ambiguous words would cause that these words develop strong orthographic links. As such, this would speed up orthographic settling for ambiguous words, facilitating thus their recognition. In contrast, Borowsky and Masson (1996)’s model and the semantic feedback account (Balota et al., 1991; Hino & Lupker, 1996) suggest that ambiguous words benefit from triggering a large amount of semantic activation during word processing (i.e., semantic enhanced activation accounts). To compare these two points of view, we examined the EEG correlates of ambiguous and unambiguous words during a LDT. Namely, we examined the N200 component (an ERP component associated with orthographic processing) and the N400 component (an ERP component associated with semantic processing). The results showed no differences between ambiguous and unambiguous words regarding the N200. In contrast, we found differences between both types of words in the N400: ambiguous words elicited larger N400 amplitudes than unambiguous words. Accordingly, the results of this experiment are in line with semantic enhanced activation accounts of the ambiguity advantage (Borowsky & Masson, 1996; Hino & Lupker, 1996). However, semantic enhanced activation accounts differ in how this large amount of semantic activation triggered by ambiguous words facilitates their recognition. The semantic feedback account (e.g., Hino & Lupker, 1996) holds that such activation feedbacks to the orthographic level, boosting orthographic processing and thus making these words to reach faster the orthographic recognition threshold. In contrast, Borowsky and Masson's (1996) model assumes that this semantic activation does not boost orthographic processing, so that ambiguous words are recognized faster because they increase the global semantic activation level. To test whether ambiguous words benefit from an orthographic boost or not, in the Study 4 we compared ambiguous and unambiguous words in a task that taps orthographic processing, in particular, in a two-alternative forced-choice task (2AFC). The results of this task showed significant differences between ambiguous and unambiguous words, providing further support for the hypothesis that ambiguous words benefit from an orthographic boost (e.g., Hino & Lupker, 1996). Therefore, according to the evidence obtained in the present thesis, the best explanation for the ambiguity advantage is that ambiguous words triggers a large amount of semantic activation during processing, affecting orthographic processing levels via semantic feedback and thus speeding up word recognition (e.g., Hino & Lupker, 1996).