Processing of Morphological and Semantic Cues inRussian and GermanVera KempeState University of New York at OswegoBrian MacWhinneyCarnegie Mellon UniversityThis study examines the on-line processing of morphological cues to sentenceinterpretation in Russian and German with the goal of evaluating the relativeimpacts of cue availability and cue reliability. Both Russian and German usethe cues of word order, animacy, case-marking, and subject–verb agreementto identify the agent of active transitive sentences. However, the availabilityof the case-marking cue is higher in Russian than in German. Using apicture-choice paradigm, we contrasted case-marking and animacy inRussian and German. The reaction times showed larger effects of case-marking in Russian than in German and effects of animacy in German, butnot in Russian. These results suggest that the higher the availability of a cue,the larger the processing benets associated with the presence of this cue andthe smaller the impact of other converging information. A recurrentcascaded backpropagation network was designed to simulate these effects.The network succeeded in capturing the essential language differences in thereaction times, thereby illustrating how the statistical properties of cues in alanguage can affect the time-course of activation of alternative interpreta-tions during sentence processing.INTRODUCTIONCrosslinguistic studies of sentence processing have documented pervasivedifferences between languages (Cuetos & Mitchell, 1988; Frazier &d’Arcais, 1989; Vigliocco, Butterworth, & Semenza, 1994). The Competi-Requests for reprints should be addressed to Vera Kempe, Department of Psychology,SUNY Oswego, Oswego, NY 13126, USA. E-mail: [email protected] research was supported by a fellowship from the German Academic Exchange Board(DAAD) to the rst author. We would like to thank Don Mitchell, Roman Taraban, and veanonymous reviewers for helpful comments on an earlier draft of this paper.c1999 Psychology Press LtdLANGUAGE AND COGNITIVE PROCESSES, 1999, 14 (2), 129–171130KEMPE AND MACWHINNEYtion Model (MacWhinney & Bates, 1989) has been proposed as a way ofrelating the observed processing differences to variations in languagestructure. In particular, the Competition Model focuses on the role ofsurface cues such as word order, noun animacy, subject–verb agreement,and noun case-marking. The strength of a cue is viewed as depending onthree factors: (i) its availability, dened as the proportion of times a cue ispresent and can be used for accessing the underlying function, (ii) itsreliability, dened as the proportion of times a cue signals the correctinterpretation given that it was present, and (iii) its cost, which depends onthe perceptual salience of the cue and the load it places on workingmemory.Studies within the Competition Model framework have examinedprocessing in over 15 languages (Bates, McNew, MacWhinney, Devescovi,& Smith, 1982; Kail, 1989; MacWhinney, Bates, & Kliegl, 1984;MacWhinney, Ple´h, & Bates, 1985; McDonald, 1986; Sokolov, 1989). Ina typical Competition Model experiment, subjects are presented withsimple transitive sentences and are asked to decide which noun refers tothe agent of the sentence. A consistent nding from these studies is that, ifthe cues for agentivity are placed in competition with each other, thechoice of one of the nouns as the agent can be modelled by multiplicativecue integration (McDonald & MacWhinney, 1989). In these off-linestudies, it has been found that the primary determinant of cue strength iscue reliability. In other words, when subjects are given enough time topermit full, deliberate consideration of all competing cues and interpreta-tions, they integrate cues in a way that maximises the probable correctnessof their nal interpretation.Newer work in the Competition Model framework (Hernandez, Bates,& Avila, 1994; Kilborn, 1989; Li, Bates, & MacWhinney, 1993; McDonald& MacWhinney, 1995; Mimica, Sullivan, & Smith, 1994) has attempted toextend the earlier off-line models to the study of on-line sentenceprocessing. The paradigm used in many of these studies is a speeded cross-modal picture-choice task where participants listen to a transitive sentencewhile making a choice between two candidate agents that are presentedvisually. Both candidates are named in the sentence, but only one can beselected as the agent. Participants are encouraged to respond even beforethe sentence is nished, if they are condent of their interpretation. Theuse of this speeded picture-choice task provides insights into how thecomplex interaction between various cues unfolds over time. There aretwo ways in which the results from this on-line technique further supportthe emphasis on cue reliability that emerged from the earlier off-linestudies. First, it has been found that the strongest and most reliable cueslead to the fastest reaction times. Second, it has been found that any typeof competition or disagreement between cues results in inhibition andLANGUAGE DIFFERENCES IN ON-LINE PROCESSING131slower reaction times. These two effects are well-predicted by a simplemultiplicative cue integration model in which cue strength is largelypredicted by reliability.However, there are two other effects found in these on-line studies thatpoint towards the need for a more complicated model. First, certain cueshave a more pronounced effect on the reaction times than on the choice ofone of the nouns as an agent. Second, reaction times are not as clearlyaffected by cue convergence as are patterns of agent choice. It appears thatstrong cues tend to saturate the on-line processing system, so thatproviding additional evidence when a strong cue is already present haslittle additional effect on reaction times. There may also be a trade-offbetween the benets gained from obtaining more evidence from aconvergent cue and the costs associated with processing this cue (Kail,1989; Mimica et al., 1994).We noted earlier that the Competition Model relates cue strength to twoseparate cue validity factors: reliability and availability. Reliability is ameasure of cue consistency and dependability, whereas availability is ameasure of frequency. There is some reason to believe that the frequencyof a syntactic structure has a stronger effect upon on-line role assignmentprocesses than upon off-line interpretation. In related work, frequencyeffects have been demonstrated for relative clause attachment (Cuetos
or
We will never post anything without your permission.
Don't have an account? Sign up