Linguistic Gender and Spoken-Word Recognition in French

Title :Linguistic Gender and Spoken-Word Recognition in French

Name of researcher :DelphineDahan, Daniel Swingley, Michael K. Tanenhaus, and James S. Magnuson

Journal Name :Memory and Language

Volume: 42

Year: 2000

Page: 465-480

Background:

Monitored from the eye movements of French participants following oral instructions to use a computer mouse to click on one of the four displayed images. Experiment 1 shows that, in the presence of grammatical gender in the context before the referral name [eg, cliquezsur les boutons (click on the (plural neut.) Button (masc.))], Participants glued to the image with the initial share of the voice with the target [ Bouteilles (bottle (fem.))] Is more than a phonological image of an unrelated name, a replication of the "cohort" effect previously found with this paradigm. When the gendermarked article immediately precedes a noun [eg, cliquezsur le bouton (click on (masc.) Button)], the initial activation of inconsistent gender groups is completely omitted (Experiment 2). This indicates that the initial set of candidates considered for the recognition of such objects is limited by the gendered articles marked. Two alternative accounts of these results, based on grammar-level processing and others based on statistical-based forms, are discussed.

Theory:

This research focuses on gender in the pronunciation of words in French. All French nouns are either feminine or masculine. A nominee of this noun is phonologically characterized in various ways, including the article form. The masculine noun preceded the exact chapter le, and the feminine noun preceded by la. The question we examined is how gender marking on a particular article affects the recognition of the next noun in the oral understanding of the language.

While many details on how to speak represented and accessible words remain controversial, it is also established that as a word unfolds, the recognition takes place against a partially activated alternate background that competes for recognition. The most active alternatives are those that best fit the input (eg, Marslen-Wilson, 1987; Zwisterlood, 1989). Thus, as a French listener hears a bouton (button), the lexical representation of words with a similar sound, such as bouteille (bottle), will be activated along with the lexical representation of the word itself.

Marslen-Wilson and colleagues (Marslen-Wilson, 1987, 1993; Marslen-Wilson & Welsh, 1978) first used this idea in the Model Cohort. According to this model, a "group" of lexical candidates is activated by the emergence of a word. The Activated Candidate becomes inconsistent with further information breaking the cohort. Recognition is achieved when only one candidate remains in the group or when the activation of one candidate is considerably greater than that of the other candidate. The following models, such as TRACE (McClelland & Elman, 1986), Shortlists (Norris, 1994), and NAM (Luce &Pisoni, 1998), also consider that some candidates are enabled, but differ in their assumptions about how competitor sets are defined. In this model, the candidate set is not strictly defined by the incidence of the word; Any part of the acoustic input may contribute to the activation of the candidate (for further discussion, see Allopenna, Magnuson, &Tanenhaus, 1998; Conn., Blasko, &Titone, 1993; Marslen-Wilson &Zwitserlood, 1989).

Like other instances while ambi- tity, the ambiguity process of resolution for a while ambiguous words can, in principle, be influenced by the correlated constraints of the context. Differences are often made between the constraints affecting the lexically candidate initially enabled and the constraints that do not affect the initial activation but rather facilitate readings between enabled alternatives. For example, Zwitserlood (1989) conducts de- investigation of how the semantic context of influence activation for the lexical cohort of champions- itors. At various points in the presentation of the last word of the spoken sentence, the speech was cut off and the word targets associated with either the actual cut of words or cohort competitors were visually presented for lexical decisions. That's the sentential context before the word cut is neutral or semantic to the actual word. When the acoustic information in the word cut is ambiguous between the word itself and its cohort competitors, the lexical decision is facilitated to target as- sociated with either the actual word or the group of competitors, regardless of the contextual bias. How- ever, facilitation to re-linked semantic targets in the cohort lasts longer when the semantic context is neutral than when it is leaning to the actual word. On the basis of this time-only Analysis, Zwisterlood concluded that the semantic context of the type he studied did not affect the initial activation of the candidate set, but rather influenced the selection of processes among candidates.

In contrast to the thick literature on syntactic-semantic effects and the context of the word recognition, the literature on the effects of morphosyntacticcontext is rare (eg, Lukatela, Kostic, Feldman, &Turvey, 1983). Several studies that have focused specifically on gender have found strong effects. Cole and Segui (1994) found that lexical decisions for nouns are slowed when nouns are preceded by gender or odd gender determinants (see also Jakubowicz&Faussart, 1998). Grosjean, Dommergues, Cornu, Guillelmon, and Besson (1994) suggest that gender articles are marked preceding gated (cut) noun limiting alternative sets generated by participants; less acoustic information is required for subjects to recognize nouns when a gender article is indicated to exist than When it is present. In addition, lexical decisions are faster on nouns when they are preceded by articles than when the article is omitted.

Grosjean et al. S 'results indicate that the presence of a marked gender article can increase the recognition of a noun. This conclusion is supported by Bates, Devescovi, Hernandez, and Pizzamiglio (1996) in Italy, using word-repetition, gender monitoring, and grammatical assessment tasks. The performance of the three tasks is slower when object targets are preceded by congruent gender traits and faster when preceded by congruent gender adjectives (relative to baseline conditions using gender-neutral adjectives). Although important, this study does not directly investigate the role of gender information in the initially enabled candidate set. In particular, they do not assess whether candidates who share onset are the same as targets but morphosynthetic initial mismatches are activated.

In order to determine whether gender is marked Previous articles nouns that may affect candidate lexical enters a set of competitors, we examine inconsistent gender competitor activations that match the initial sound of the target word, but harmonize gender marks in the article. We assessed an activation competitor cohort using heads with an eye tracking paradigm, introduced by Tanenhaus, Spivey-Knowlton, Eberhard, and Sedivy (1995) to study spoken-language processes in visual contexts and further developed by Allopenna et al. (1998) to explore pronounced- speech recognition (see also Cooper, 1974; Tanenhaus& Spivey-Knowlton, 1996). In the study, participants followed oral instructions to manipulate either real objects or images displayed on a computer screen, while their eye movements were monitored using a lightweight camera mounted on a headband. Eye movement to objects in the workplace closely locked time to refer expressions in speech flow progresses, providing sensitive and nondisruptive measure of oral comprehension language.

In this study, French participants presented with a computer screen consisted of cross-centered fixation and four images and were given verbal instructions to click on one of the images using the mouse [eg, cliquezsur le bouton (click on (masc.) Button (masc.))] . The four images included targets (eg, bouton), cohort competitors (hereinafter, cohorts) [eg, bouteille (bottles (fem.))], And two distractors (see Fig. 1). By recording the eye fixation produced during the presentation of the target bouton word, we were able to assess the activation of the inconsistent bouteille gender group. This study examines whether gender articles are characterized affecting candidates initially considered for the recognition of the following noun.

Method:

· Free Variable = Gender Linguistics

· Dependent variable = Word Recognition

· Subject = Native French

Results

Test of cohort activation. For 300- to 700-ms window time, the proportion fixa- tions for the target were 47.7%, for the competitors cohort, 21.2%, for each distractor, 12.1%; And cross, 7.0%. Planned comparison (one tailed test) between fixation to competitor cohort and the average distractor performed. The competitor group is fixed-created longer than the distractor (t 1 (11) ϭ 2.72, p Ͻ .05, t 2 (31) ϭ 3.26, p Ͻ 005). These results indicate that cohort competitors are activated during the presentation of the target word as a result of phonological similarity to the target.

Figure 2 shows the possibility of fixating the target and the cohort and the average probability of being distracted from time to time, from 0-1000 ms after the onset noun. The graph shows that the probability of fixating the cohort begins to deviate from the possibility of a distractor glued about 300 ms after the onset noun and remains high until long after 700 ms. Comparisons to fixations for 300 - 500 - ms of windows and 500 - to 700 - ms window con - ted these observations (at 300 - to 500 - ms window, t 1 (11) ϭ 2.58, p Ͻ .05; t 2 (31 ) Ϭ 3.61, p Ͻ .001, at 500 - for 700-ms window, t 1 (11) ϭ 2.08, p Ͻ .05, t 2 (31) ϭ 1.88, p Ͻ .05). To test the 300-700-forward fixation difference to the ms-window, ANOVA was performed on target fixation, a competitor cohort, and the first 300 ms first trapping after target onset showed that the proportion of fixation was different (F1 (2, 22) ϭ 7.0, P Ͻ .01, MSE ϭ 0.0053, F 2 (2, 62) ϭ 4.61, p Ͻ .05, MSE ϭ 0,0243). The Newman Keuls test indicates that the proportion of fixation to Target is higher than that of other pic- build structures, but the fixation for the cohort is no different from those of the distractor. This initial advantage for fixed targeting should be attrib- uted to noise-given the rather small amount of observation and unbiased bias for the drawing, as each item of these alternative pair targets is used as a reference.

We also conducted contingent analyzes by selecting trials whose participants glued either the cross or one of the bullies at the beginning of the noun target (112 out of 192 experiments, 58.3%). Figure 3 presents the probability fixation for targets, cohorts, and ag- ristractors aged time. If the research subject is initially riveted to one distractor the same possibility to shift to another trap or group, fixation for the cohort will not be expected to rise above the fixation to the distractor. However, as shown in Fig. 3, the subject of research initially riveted by one distractor is more likely to shift to the cohort than continue to be glued to the trap. This con- tingent analysis shows that the "group" of Securities seen above the whole set of experiments can not be attributed to the basic differences in the subjects' initial fixation. Fixation for cohorts much higher than fixation to disturbed concentrations from 300 to several hundred milliseconds after target onset (300-500 ms, t 1 (11) ϭ 2.02, p Ͻ 0.5, t 2 (31) ϭ 2.74 , P Ͻ 01, from 500 700 ms, t 1 (11) ϭ 2.43, p Ͻ .05, t 2 (31) ϭ 1.8, p Ͻ .05). The probability of fixing the targets did not differ significantly from the probability of gluing the cohort to 500 ms after the onset target (300-500 ms, t 1 Ͻ 1, t 2 Ͻ 1, from 500 to 700 ms, t 1 (11) ϭ 6.97, p .01, t 2 (31) ϭ 5.08, p Ͻ 01).

Discussion

This study investigates the use of morphology of word-spoken recognition contexts. In particular, it is focused on whether French listeners use the Information gender marked before the noun to limit the set of active lexical candidates to those who match this gender information. In order to assess the inconsistent gender word activation, our French subjects are presented with four dis- play images and the oral instruction asks the subject to click on one of the images (eg, the image of the button, bouton, which is a masculine noun). Pre-sented with the target image is a pure "group" whose name shares the initial sound of the target word but a different gender (for example, bottle image, bouteille, feminine noun) and two distractors whose names are not phonologically similar to the target or word cohort . The eye movements for the photographs are interpreted as evidence for the activataata according to the drawings. Thus, greater fixation with cohort images rather than to distractors as the target word is folded over time while activation indicates the gender-mis- matching cohort of competitors. Cohort activation was found when the target word was preceded by gender-neutral articles, tutoring (Experiment 1), how- ever, no cohort activation was found when the der- der-derived gene immediately preceded the tar-word (Experiment 2).

This result proves that the initial set of lexical candidates can be limited byPrevious context: While cohort activation is found when the context does not contain the gender of the information, no activation as observed when a gender article is marked immediately pre- submits the target word. To understand how Gender articles are marked affects the spoken-word of recognition, it is important to distinguish between possible notional, grammatical, and effect-based forms. First, the activation of the article may activate the notional concept of masculine or feminine, thus biasing the lexical concept of gender. The word is consistent. However, this seems very unlikely, given the absence of a clear and correlated concept of gender perception in French. Some entities have obviously masculine or feminine names based on their gendered attachment, as nouns are used to refer to men and women, whereas other nouns tend to be masculine or feminine by shape (eg, round objects tend to have feminine names with gender , For example, Bem, 1981). However, for some entities, gender can not be predicted from physical misdeeds or referral functions, such as shape or size.

A second possibility is that the listener uses a marked grammatical gender in the article to limit competitors set for words consistent with this grammatical gender. This constraint is a product of the grammatical structure of language. In noun phrases, noun heads, articles, and any adjectives that may have to agree on gender. This dependence may be expressed in terms of probability for a word to arise, given the gender grammatical of the preceding context, and perhaps partial information from the target word itself-for exams-for example, the probability of the word bouton in the context of acoustic information / bu / pre - submit by masculine gender. This constraint can be used by priming all objects consistent with the gender of the code in the article and / or inhibiting others. When the first voice of the target word arrives, words consistent with the gender mark have been primed, and competitors that are inconsistent with gender marking, such as competing cohorts, have hampered and therefore will not become active even though they are equal to the form of target -said. In this account, one would expect to see more fixation to the matched image gender rather than the mismatched image gender, since words related to the gender of the matching image would have been primed. However, neither Experiment 1 nor Experiment 2 show more visible evidence of gender distractors matched than gender incompatibility distractors.

However, we can not exclude the possibility that priming occurs but is too weak to be observed in the absence of a bot. The tom-up information is consistent with the gender matching of the lexical candidate. Alternatively, the grammatical effect of gender may be ex- pressed as a conditional probability, where the probability of the dependent occurs at the grammatical moment of gender in the context (eg, what is the probability of the word bouton, given the partial input / bu / and masculine gender information that precedes it) . Crucially, this regularity of distribution will be calculated by using catego-ries grammar (ie, feminine or masculine). In ac- count, lexical access will utilize information from the grammatical level of processing language.

A third possibility is that linguistic forms mark sex, rather than grammati-cal categories, the effect of access from successive to noun quent. The regularity of the distribution can be calculated between the word-starting form (or phoneme sequence) and the following word form: As the first sound of the objective word is heard following a marked Gender article (eg, / l bu /, lebou ...), Probability of The target word becomes bouton in a high le context, whereas the probability of the target word being bouteille in this context is very low, based on voice-based language stats. The co-occurrence article in France is very high: There are ten strong dency for nouns happening with accompanying-Nying determinants, and clothing is used even in cases where no determinant is semantically required (eg, I like cheese ,J'aime le fromage, Harris, 1990). In addition, although adjectives can be inserted between determinants and nouns, they often occur after the noun. In this account, the processing system will track the policy-based vote contingency and use this contingency during the recognition word.

The difference between form-based and grammar-based effects is subtle, but important. French grammar specifies that the mas-culine chapter passes the bouton, which is the masculine. Grammatical results in these statistics Facts about / l / and bouton. But the sensitivity of listeners' POSSI ble to this fact in speech recognition may reflect an insensitivity to grammatical rules, defined in terms of the "masculine" and "feminine" categories, or the sensitivity to co-occurrence of / l / and / Them-self.

Both accounts may have very different consequences on the pronounced architecture of the word-recognition and processing system. This study does not allow us to decide whether the grammatical gender is brought about by an article, or high-co-based co-occurrence with targeted, constrained competitors defined by excluding the inconsistent gender inconsistency of a competing cohort. To tempt beyond this alternative, it is necessary to create a con- text in which grammatical information of gender precedes the noun, but the forms that carry the gender and target information have a rela- tively low co-occurrence. This can be done using a lower frequency of gender-marked articles, with the expression of one or more words between the gender marked definite article and noun, or by using lower-frequency words, such as adjectives, to provide gender information. This research using this strategy is currently in progress in our laboratories.

In most speech-recognition models, lexical activation is primarily a function of the degree of compatibility between spoken words and stored lexical representations, modulated by the char acteristics of the word itself, such as frequency. This study shows that early lexical activation can be limited by external speech to the word itself. A full speech recognition report is continuous speech so it should refer to the context before the word.