Here we investigate how hierarchical structure may affect parsing decisions. In complex noun phrases like the servant of the actress who was on the balcony speakers of English prefer to parse the relative clause as attached to the noun closest to it (the actress). We have conducted a series of offline experiments and found that in similar constructions, speakers of Slovenian prefer to attach the relative clause to the more remote noun (the servant). We discuss implications of this finding for the nature of the universal parsing principles.
COBISS.SI-ID: 3218683
In this work we address an interesting phenomenon concerning change of the pattern of agreement in Bulgarian numeral phrases from the prescriptively correct “count form” of masculine nouns to a simple plural form. The phenomenon is observed in everyday speech and writing and has been noted by traditional Bulgarian grammarians. We propose to treat it on a par with agreement errors previously observed in the language production literature on English (e.g. The editor of the books are…) and Romance languages. We have conducted a corpus study of agreement errors in Bulgarian the results of which support a class of theoretical models of language production which hold, in particular, that structural hierarchy is relevant for computation of agreement in language production.
COBISS.SI-ID: 2910459
We set to determine whether success in segmenting L-grammar sequences is due to a particular parsing strategy aimed at detecting constituents in the linear stream. This strategy draws on the self-similar property of context-free L-grammars. To test this hypothesis, we conducted a dual task experiment whereby groups of adult participants performed a visual l(exical) d(ecision) task focusing their attention thereon, while passively listening to a synthesized sequence of syllables generated either by Fibonacci or Morse-Thue L-grammar or a random sequence in the headphones. A separate control group performed the LD task alone (total N=40). We found that 1)listening to L-grammar sequences leads to significant increase of r(esponse) t(imes) on the LD task, as opposed to the random sequence; 2)there were no significant differences in RTs between the two L-grammar conditions. These results support the hypothesis and further suggest that constituent detection is performed outside the focus of attention and thus largely an automatic process.
COBISS.SI-ID: 2866939
In Bulgarian numeral phrases like “four cars” the numeral assigns a special “count form” on the noun which, however, is often replaced with the regular plural in colloquial speech. We have conducted a series of offline and online experiments on processing these numeral phases asking whether and how the structural distance between the numeral and noun affects affects the pattern of these speech “errors”. We found that recursive addition of adjectives (e.g. “four (red (big (old))) cars”) increases the likelihood of such errors because adjectives, which themselves are marked for regular plural, provide a competing source of feature compatible with the noun.
COBISS.SI-ID: 2867195