Диссертация (Experimental study of several core concepts of theoretical morphology (on the material of russian) - regularity, syncretism, markedness), страница 27

The goal of the experiment was to find outwhether subject–predicate gender agreement is processed differently dependingon the declension and gender the subject noun belongs to. 33 native speakers of Russian aged 18–55 (12 women) took part in the experiment.It is impossible to test all potentially interesting combinations of gender and declension in one experiment, so we selected the following three groups of inanimate nounsas subjects: masculine nouns of the 2nd declension (prototypical masculine, 2DM), feminine nouns of the 1st declension (prototypical feminine, 1DF), and feminine nouns of the3rd declension (non-prototypical feminine, 3DF).

The materials included 36 sets of targetsentences in six conditions exemplified in (3a-f). All sentences had the same structure:a subject noun, byl/byla ‘wasM/F’, an adjective or participle, and a three-word PP.(3) a. 2DM G(rammatical):XalatbylrobeM.NOM.SGwasMb. 2DM U(ngrammatical): XalatrobeM.NOM.SGc. 1DF G(rammatical):KurtkajacketF.NOM.SGd. 1DF U(ngrammatical): KurtkajacketF.NOM.SGbylawasFbylawasFbylwasMpotrepannym otmnogoletnej noski.shabbyMfromyears-long wearpotrepannoj...shabbyFpotrepannoj…shabbyFpotrepannym…shabbyM69365Slioussar N. A.e. 3DF G G(rammatical):Šinel’byla potrepannoj…overcoatF.NOM.SG wasF shabbyFf. 3DF U(ngrammatical): Šinel’bylpotrepannym…overcoatF.NOM.SG wasM shabbyM‘The robe / jacket / overcoat was shabby from being worn for many years.’Half of the sentences contained gender agreement errors on the predicate becausetaking previous studies of agreement into account (primarily agreement attraction experiments), the effects could be expected to be different in grammatical and ungrammaticalsentences6.

Subject nouns in the three declension groups were balanced for frequency andlength using the StimulStat lexical database (http://stimul.cognitivestudies.ru, [Alexeevaet al., 2018]). Frequency information in this database is taken from the Frequency Dictionary of Modern Russian Language [Lyashevskaya & Sharoff, 2009].Target sentences were distributed into six experimental lists so that each participant saw only one sentence from each set. The lists also contained 80 grammatically correct filler sentences.

The sentences were presented on a PC using Presentation software(http://www.neurobs.com). We used the word-by-word self-paced reading methodology.Each trial began with a sentence in which all words were masked with dashes while spacesand punctuation marks remained intact. Participants were pressing the space bar to reveala word and re-mask the previous one.

One third of the sentences were followed by forcedchoice comprehension questions to ensure that the participants were reading properly.We analyzed participants’ question-answering accuracy and reading times.On average, participants answered 12% questions to target sentences incorrectly,no participants made more than 3 errors. Reading times that exceeded a thresholdof 2.5 standard deviations, by region and condition, were excluded [Ratcliff, 1993].In total, 2.0% of the data were excluded as outliers.

Average reading times per regionin different conditions are presented in Figure 1.Figure 1. Average reading times per region (in ms)in different experimental conditions6We selected predicates that consisted of a copula and an adjective or participle because suchpredicates were used in the previous experimental studies of subject—predicate genderagreement in Russian.69466Gender, Declension and Stem-final ConsonantsFor each region, we made pairwise comparisons between the three conditionsusing a 2 × 2 Repeated Measures ANOVA with grammaticality and declension as factors. Analyses by participants (F1) and by items (F2) were performed.

In region 1 (thesubject noun), there were no significant results, which means that nouns in differentconditions were properly balanced and that nouns of a particular gender or declension are not intrinsically more difficult to process.Region 2 contains the verb byl / byla ‘wasM/F’—this is where agreement errorsappear in ungrammatical sentences.

Figure 1 suggests that participants’ reactionto these errors was different depending on the declension of the subject noun. In theconditions 1DF and 2DM (with prototypical feminine and masculine subjects), reading times in ungrammatical sentences are longer than in grammatical ones, whileno such difference can be seen in the 3DF conditions (with non-prototypical femininesubjects), which indicates that the error tends to remain undetected in the latter case.Statistical analyses support this intuition. In the comparison between 1DF and3DF conditions, grammaticality and the interaction between declension and grammaticality are significant (F1(1,32) = 8.13, p < 0.01, F2(1,35) = 4.20, p = 0.05;F1(1,32) = 7.41, p = 0.01, F2(1,35) = 4.05, p = 0.05), while the main effect of declension does not reach significance. This means that the influence of grammaticality depends on the declension of the subject.

In the comparison between 2DM and3DF conditions, grammaticality reaches significance, while the interaction betweendeclension and grammaticality is marginally significant (F1(1,32) = 8.01, p < 0.01,F2(1,35) = 4.09, p = 0.05; F1(1,32) = 3.98, p = 0.05, F2(1,35) = 3.17, p = 0.08). When1DF and 2DM are compared, only the grammaticality factor is significant (F1(1,32) =18.66, p < 0.01, F2(1,35) = 10.21, p < 0.01).In region 3 that contains an adjective or participle, differences between grammatical and ungrammatical sentences become visible in all conditions.

Accordingly,the grammaticality factor is significant in all pairwise comparisons (F1(1,32) = 15.90,p < 0.01, F2(1,35) = 21.24, p < 0.01 for 1DF vs. 2DM; F1(1,32) = 11.98, p < 0.01,F2(1,35) = 6.20, p = 0.02 for 1DF vs. 3DF; F1(1,32) = 9.73, p < 0.01, F2(1,35) = 7.83,p < 0.01 for 2DM vs. 3DF). No other factors or interactions reach significance.Regions 4–6 contain a three-word PP. In region 4, a tendency that can be alreadydetected in region 3 becomes statistically significant: the error-related delay in reading times is more pronounced in the 2DM conditions (with masculine subjects) thanin the 1DF and 3DF conditions (with feminine subjects). In the comparison between1DF and 2DM conditions, grammaticality and the interaction between declension andgrammaticality are significant (F1(1,32) = 36.95, p < 0.01, F2(1,35) = 15.91, p < 0.01;F1(1,32) = 9.77, p < 0.01, F2(1,35) = 6.45, p = 0.02), while declension is not significant.

The same is true for the comparison between 3DF and 2DM (F1(1,32) = 50.11,p < 0.01, F2(1,35) = 13.17, p < 0.01; F1(1,32) = 11.38, p < 0.01, F2(1,35) = 5.51, p =0.03). When 1DF and 3DF are compared, only the grammaticality factor is marginallysignificant (F1(1,32) = 12.34, p < 0.01, F2(1,35) = 3.65, p = 0.07).In region 5, only the grammaticality factor is significant in all pairwise comparisons(F1(1,32) = 18.51, p < 0.01, F2(1,35) = 17.67, p < 0.01 for 1DF vs.

2DM; F1(1,32) = 14.78,p < 0.01, F2(1,35) = 6.10, p = 0.02 for 1DF vs. 3DF; F1(1,32) = 18.07, p < 0.01, F2(1,35) =10.07, p < 0.01 for 2DM vs. 3DF). In region 6, there are no significant differences.69567Slioussar N. A.Finally, let us note that when we planned the experiment, we did not considerassessing the role of the final consonant of 3DF nouns. But 10 out of 36 nouns we selected ended in -l’ or -r’, which is more characteristic for masculine, while other nounshad final consonants characteristic for feminine. So the role of this factor could be estimated, and there were no hints of any relevant differences.4.

Pilot experiment 2This pilot experiment was included in a study we conducted together withVarvara Magomedova (SUNY, Stony Brook) and Natalia Chuprasova, an MA studentat Saint-Petersburg State University. The main goal of the study was to find out howRussian speakers determine the gender of real and nonce nouns with diminutive andaugmentative suffixes. However, to make the materials more diverse, other nouns hadto be included, and we selected 12 real and 12 nonce nouns ending in palatalized consonants (as well as some indeclinable nouns etc.).Participants were 30 native Russian speakers (17 women), aged 19–30.

They received a list of seven adjectives and then were presented with nouns one by one. Theywere asked to pick a matching adjective and pronounce the resulting phrase. Adjectives had meanings like ‘big’, ‘small’, ‘cool’, ‘bad’ etc., to make participants think thatthe experiment was about semantic connotations of different nouns.Analyzing the gender of the adjectives selected by the participants, for 12 realnouns ending in palatalized consonants (6 masculine and 6 feminine) we found only7 errors out of 360 responses. There were three errors with žen’šen’ ‘ginsengM’, twoerrors with stupen’ ‘stepF’, and one error with kisel’ ‘starch drink M’ and with prorub’‘ice-holeF’.

The low number of errors agrees with the previous findings by Rusakova(2013): adult native speakers of Russian do not experience particular difficulties determining the gender of such nouns.As for 12 nonce nouns, we had two examples with each of the following endings:-b’, -d’, -s’, -n’, -l’ and -r’. The number of answers with masculine adjectives in thesegroups was 22 (out of 60), 26, 38, 30, 44 and 51, respectively.