R. Koolen - Academia.edu (original) (raw)

Papers by R. Koolen

Research paper thumbnail of Learning Preferences for Referring Expression Generation: Effects of Domain, Language and Algorithm

One important subtask of Referring Expression Generation (REG) algorithms is to select the attrib... more One important subtask of Referring Expression Generation (REG) algorithms is to select the attributes in a definite description for a given object. In this paper, we study how much training data is required for algorithms to do this properly. We compare two REG algorithms in terms of their performance: the classic Incremental Algorithm and the more recent Graph algorithm. Both rely on a notion of preferred attributes that can be learned from human descriptions. In our experiments, preferences are learned from training sets that vary in size, in two domains and languages. The results show that depending on the algorithm and the complexity of the domain, training on a handful of descriptions can already lead to a performance that is not significantly different from training on a much larger data set.

Research paper thumbnail of Need I say more? On factors causing referential overspecification

Proceedings of the CogSci workshop on the Production of Referring Expressions (PRE-CogSci 2009), Amsterdam, The Netherlands, Jul 29, 2009

We present the results of an elicitation experiment conducted to investigate which factors cause ... more We present the results of an elicitation experiment conducted to investigate which factors cause speakers to overspecify their referential expressions, where we hypothesized properties of the target and properties of the communicative setting to play a role. The results of this experiment show that speakers tend to provide more information when referring to a target in a more complex domain and when referring to plural targets. Moreover, written and spoken referring expressions do not differ in terms of redundancy, but do differ in terms of ...

Research paper thumbnail of The Effect of Scene Variation on the Redundant Use of Color in Definite Reference

Cognitive Science, 2013

This study investigates to what extent the amount of variation in a visual scene causes speakers ... more This study investigates to what extent the amount of variation in a visual scene causes speakers to mention the attribute color in their definite target descriptions, focusing on scenes in which this attribute is not needed for identification of the target. The results of our three experiments show that speakers are more likely to redundantly include a color attribute when the scene variation is high as compared with when this variation is low (even if this leads to overspecified descriptions). We argue that these findings are problematic for existing algorithms that aim to automatically generate psychologically realistic target descriptions, such as the Incremental Algorithm, as these algorithms make use of a fixed preference order per domain and do not take visual scene variation into account.

Research paper thumbnail of Reduction in gesture during the production of repeated references

In dialogue, repeated references contain fewer words (which are also acoustically reduced) and fe... more In dialogue, repeated references contain fewer words (which are also acoustically reduced) and fewer gestures than initial ones. In this paper, we describe three experiments studying to what extent gesture reduction is comparable to other forms of linguistic reduction. Since previous studies showed conflicting findings for gesture rate, we systematically compare two measures of gesture rate: gesture rate per word and per semantic attribute (Experiment I). In addition, we ask whether repetition impacts the form of gestures, by manual annotation of a number of features (Experiment I), by studying gradient differences using a judgment test (Experiment II), and by investigating how effective initial and repeated gestures are at communicating information (Experiment III). The results revealed no reduction in terms of gesture rate per word, but a U-shaped reduction pattern for gesture rate per attribute. Gesture annotation showed no reliable effects of repetition on gesture form, yet participants judged gestures from repeated references as less precise than those from initial ones. Despite this gradient reduction, gestures from initial and repeated references were equally successful in communicating information. Besides effects of repetition, we found systematic effects of visibility on gesture production, with more, longer, larger and more communicative gestures when participants could see each other. We discuss the implications of our findings for gesture research and for models of speech and gesture production.

Research paper thumbnail of Learning Preferences for Referring Expression Generation: Effects of Domain, Language and Algorithm

One important subtask of Referring Expression Generation (REG) algorithms is to select the attrib... more One important subtask of Referring Expression Generation (REG) algorithms is to select the attributes in a definite description for a given object. In this paper, we study how much training data is required for algorithms to do this properly. We compare two REG algorithms in terms of their performance: the classic Incremental Algorithm and the more recent Graph algorithm. Both rely on a notion of preferred attributes that can be learned from human descriptions. In our experiments, preferences are learned from training sets that vary in size, in two domains and languages. The results show that depending on the algorithm and the complexity of the domain, training on a handful of descriptions can already lead to a performance that is not significantly different from training on a much larger data set.

Research paper thumbnail of Need I say more? On factors causing referential overspecification

Proceedings of the CogSci workshop on the Production of Referring Expressions (PRE-CogSci 2009), Amsterdam, The Netherlands, Jul 29, 2009

We present the results of an elicitation experiment conducted to investigate which factors cause ... more We present the results of an elicitation experiment conducted to investigate which factors cause speakers to overspecify their referential expressions, where we hypothesized properties of the target and properties of the communicative setting to play a role. The results of this experiment show that speakers tend to provide more information when referring to a target in a more complex domain and when referring to plural targets. Moreover, written and spoken referring expressions do not differ in terms of redundancy, but do differ in terms of ...

Research paper thumbnail of The Effect of Scene Variation on the Redundant Use of Color in Definite Reference

Cognitive Science, 2013

This study investigates to what extent the amount of variation in a visual scene causes speakers ... more This study investigates to what extent the amount of variation in a visual scene causes speakers to mention the attribute color in their definite target descriptions, focusing on scenes in which this attribute is not needed for identification of the target. The results of our three experiments show that speakers are more likely to redundantly include a color attribute when the scene variation is high as compared with when this variation is low (even if this leads to overspecified descriptions). We argue that these findings are problematic for existing algorithms that aim to automatically generate psychologically realistic target descriptions, such as the Incremental Algorithm, as these algorithms make use of a fixed preference order per domain and do not take visual scene variation into account.

Research paper thumbnail of Reduction in gesture during the production of repeated references

In dialogue, repeated references contain fewer words (which are also acoustically reduced) and fe... more In dialogue, repeated references contain fewer words (which are also acoustically reduced) and fewer gestures than initial ones. In this paper, we describe three experiments studying to what extent gesture reduction is comparable to other forms of linguistic reduction. Since previous studies showed conflicting findings for gesture rate, we systematically compare two measures of gesture rate: gesture rate per word and per semantic attribute (Experiment I). In addition, we ask whether repetition impacts the form of gestures, by manual annotation of a number of features (Experiment I), by studying gradient differences using a judgment test (Experiment II), and by investigating how effective initial and repeated gestures are at communicating information (Experiment III). The results revealed no reduction in terms of gesture rate per word, but a U-shaped reduction pattern for gesture rate per attribute. Gesture annotation showed no reliable effects of repetition on gesture form, yet participants judged gestures from repeated references as less precise than those from initial ones. Despite this gradient reduction, gestures from initial and repeated references were equally successful in communicating information. Besides effects of repetition, we found systematic effects of visibility on gesture production, with more, longer, larger and more communicative gestures when participants could see each other. We discuss the implications of our findings for gesture research and for models of speech and gesture production.