Negation in Nonliteral Sentences

Negation in Nonliteral Sentences Raluca Budiu (budiu@parc.com) Palo Alto Research Center, 3333 Coyote Hill Rd., Palo Alto, CA 94304, USA John R. Anderson (ja@cmu.edu) Carnegie Mellon University, 5000 Forbes Ave., Pittsburgh, PA 15213, USA Abstract was different (e.g., blue). After they saw the picture and described it verbally, participants had to fill in sen- tences of the type Circle No.3 is ... and Circle No.3 is not ... In support of his exceptionality hypothesis, Wason found that, subjects needed less time to fill in exceptional negatives (e.g., Circle No.3 is not red) than the implausible negatives (e.g., Circle No.3 is not blue). De Villiers and Tager Flusberg (1975) performed a re- lated experiment with young children, and also looked at “confusability”: how far away the discrepant item was from the similar items. They found that the more similar the discrepant items was to the other items, the easier was for the children to fill in the plausible nega- tive (e.g., a “six horses and one cow” context was easier as opposed to a “seven cars and one baby’s bottle” con- text). Valle Arroyo (1982) also manipulated similarity and confusability and showed empirical support for the exceptionality hypothesis, without reaching a conclusive result about confusability. We investigate how people process negation in seman- tically distorted and metaphoric sentences. We present three experiments in which participants judged the truth of affirmative and negative sentences that were either lit- eral or contained semantic illusions (Erickson & Matt- son, 1981) or metaphors. In all experiments, negation increases processing times; although for semantic illu- sions, negation preserves the ordering of judgment times for literal and nonliteral sentences, for metaphors this ordering is reversed, with nonliteral negatives taking less time than literal negatives. This result presents evidence against the traditional Clark and Chase (1972) model of negation. We propose the negation-as distortion model and discuss how this model can explain the data. Keywords: negation; Moses illusion; metaphor; sen- tence processing; computational modeling; INP; ACT-R Introduction Negation markers (“no” and “not”) are a pervasive fea- ture of natural language that has been only modestly addressed in the psycholinguistic literature. The classic attempt of a process model of negation belongs to Clark and Chase (1972), who assume that negation acts as a post-processing operator: the sentence as an affirma- tive is processed first, and then the negation is applied. Other studies (Kaup, 2001; MacDonald & Just, 1989; Lea & Mulligan, 2002) found that sometimes negation reduces the availability of the negated meaning. Giora, Balaban, Fein, and Alkabets (2004) propose the reten- tion hypothesis for negation; according to this model a “negation marker does not suppress salient meanings activated initially but only modifies them.” In this pa- per we investigate yet another possibility – namely that negation acts as a semantic distortion, lowering the sim- ilarity of the overall (negative) sentence to other items in memory. For a long time now, negation has been associated with similarity. In his 1965 paper, Wason has proposed the exceptionality hypothesis, which asserts that nega- tive sentences are processed more easily if they state an exception to a rule than if they state “obvious”, implau- sible information. For instance, if it is known that Bill drives to work everyday, Bill did not drive to work to- day should be understood faster than Bill did not walk to work today. In Wason (1965)’s original study, par- ticipants were shown display cards with eight circles; seven circles were of one color (e.g., red) and another We use a different paradigm to study how similarity affects negation: given a fact already existent in long term memory, we distort it (in varying degrees) and ex- amine how this distortion affects the processing of neg- atives. In other words, we further explore the issue of confusability by manipulating language similarity. Pre- vious experiments with semantic distortions in affirma- tive sentences have shown that people often tend to ig- nore them and extract the gist from the sentence. Er- ickson and Mattson (1981) were the first to document the so-called Moses illusion: when people were asked questions such as How many animals of each kind did Moses take on the ark, in about 80% of cases, they did not notice that the sentence incorrectly referred to Moses instead of Noah. Metaphors (and, more generally, nonlit- eral language) can also be thought of as another easy-to- process instance of semantic distortions. In fact, Budiu and Anderson (2004) have shown that both Moses illu- sion and metaphors can be explained through a unique sentence processing mechanism that involves similarity- based matching to concepts already existent in memory. In their INP (Interpretation-Based Processing) sentence- processing model, the input sentence is matched against other facts in memory and the proposition in memory that is most similar to the input is considered the cur- rent interpretation of the sentence. INP is able to find interpretations for nonliteral sentences (e.g., The sky was filled with drops of molten silver ) by assuming a parallel