Semantic Analysis of Winograd Schema No. 1

The Winograd Schema Challenge is a general test for Artificial Intelligence, based on problems of pronoun reference resolution. I investigate the semantics and interpretation of Winograd Schemas, concentrating on the original and most famous example. This study suggests that a rich ontology, detailed commonsense knowledge as well as special purpose inference mechanisms are all required to resolve just this one example. The analysis supports the view that a key factor in the interpretation and disambiguation of natural language is the preference for coherence. This preference guides the resolution of co-reference in relation to both explicitly mentioned entities and also implicit entities that are required to form an interpretation of what is being described. I suggest that assumed identity of implicit entities arises from the expectation of coherence and provides a key mechanism that underpins natural language understanding. I also argue that conceptual ontologies can play a decisive role not only in directly determining pronoun references but also in identifying implicit entities and implied relationships that bind together components of a sentence.

Impact of the Geographic Resolution on Population Synthesis Quality

Microsimulation-based models, increasingly used in the transportation domain, require richer datasets than traditional models. Precisely enumerated population data being usually unavailable, transportation researchers generate their statistical equivalent through population synthesis. While various synthesizers are proposed to optimize the accuracy of synthetic populations, no insight is given regarding the impact of the geographic resolution on population synthesis quality. In this paper, we synthesize populations for the Census Metropolitan Areas of Montreal, Toronto, and Vancouver at various geographic resolutions using the enhanced iterative proportional updating algorithm. We define accuracy (representativeness of the sociodemographic characteristics of the entire population) and precision (representativeness of the real population’s spatial heterogeneity) as metrics of synthetic populations’ quality and measure the impact of the reference resolution on them. Moreover, we assess census targets’ harmonization and double geographic resolution control as means of quality improvement. We find that with a less aggregate reference resolution, the gain in precision is higher than the loss in accuracy. The most disaggregate resolution is thus found to be the best choice. Harmonization proves to further optimize synthetic populations while double control harms their quality. Hence, synthesizing at the Dissemination Area resolution using harmonized census targets is found to yield optimal synthetic populations.

Are Dutch posture verbs lexical or functional elements?

In Dutch, posture verbs like liggen ‘lie’ and staan ‘stand’ are obligatorily used in locative constructions with inanimate subjects, classifying the spatial Figure-Ground relation. Prima facie, in this use, posture verbs seem more like functional elements than like lexical verbs. This paper investigates processing of Dutch posture verbs in a reference resolution task in the visual world paradigm, to get more clarity on the nature of these verbs. We know that lexical verbs like rinkelen ‘ring’ cause anticipatory looks towards a matching target referent like telefoon ‘telephone’; and that they suppress looks to a phonological competitor like telescoop ‘telescope’. The functional property of grammatical gender on determiners (de vs. het) is less robust in directing looks. When it comes to anticipating the target referent, and suppressing looks to a phonological competitor, do posture verbs pattern with lexical verbs, or with functional elements like grammatical gender?

Reading between the lines: Listener’s vmPFC simulates speaker cooperative choices in communication games

Humans have a remarkable ability to understand what is and is not being said by conversational partners. It has been hypothesized that listeners decode the intended meaning of a communicative signal by assuming speakers speak cooperatively, rationally simulating the speaker’s choice process and inverting it to recover the speaker’s most probable meaning. We investigated whether and how rational simulations of speakers are represented in the listener’s brain, by combining referential communication games with functional neuroimaging. We show that listeners’ ventromedial prefrontal cortex encodes the probabilistic inference of what a cooperative speaker should say given a communicative goal and context, even when such inferences are irrelevant for reference resolution. The listener’s striatum encodes the amount of update on intended meaning, consistent with inverting a simulated mental model. These findings suggest a neural generative mechanism, subserved by the frontal-striatal circuits, that underlies our ability to understand communicative and, more generally, social actions.

Gesture Influences Resolution of Ambiguous Statements of Neutral and Moral Preferences

When faced with an ambiguous pronoun, comprehenders use both multimodal cues (e.g., gestures) and linguistic cues to identify the antecedent. While research has shown that gestures facilitate language comprehension, improve reference tracking, and influence the interpretation of ambiguous pronouns, literature on reference resolution suggests that a wide set of linguistic constraints influences the successful resolution of ambiguous pronouns and that linguistic cues are more powerful than some multimodal cues. To address the outstanding question of the importance of gesture as a cue in reference resolution relative to cues in the speech signal, we have previously investigated the comprehension of contrastive gestures that indexed abstract referents – in this case expressions of personal preference – and found that such gestures did facilitate the resolution of ambiguous statements of preference. In this study, we extend this work to investigate whether the effect of gesture on resolution is diminished when the gesture indexes a statement that is less likely to be interpreted as the correct referent. Participants watched videos in which a speaker contrasted two ideas that were either neutral (e.g., whether to take the train to a ballgame or drive) or moral (e.g., human cloning is (un)acceptable). A gesture to the left or right side co-occurred with speech expressing each position. In gesture-disambiguating trials, an ambiguous phrase (e.g., I agree with that, where that is ambiguous) was accompanied by a gesture to one side or the other. In gesture non-disambiguating trials, no third gesture occurred with the ambiguous phrase. Participants were more likely to choose the idea accompanied by gesture as the stimulus speaker’s preference. We found no effect of scenario type. Regardless of whether the linguistic cue expressed a view that was morally charged or neutral, observers used gesture to understand the speaker’s opinion. This finding contributes to our understanding of the strength and range of cues, both linguistic and multimodal, that listeners use to resolve ambiguous references.

Common Ground Information Affects Reference Resolution: Evidence From Behavioral Data, ERPs, and Eye-Tracking

One of the most important social cognitive skills in humans is the ability to “put oneself in someone else’s shoes,” that is, to take another person’s perspective. In socially situated communication, perspective taking enables the listener to arrive at a meaningful interpretation of what is said (sentence meaning) and what is meant (speaker’s meaning) by the speaker. To successfully decode the speaker’s meaning, the listener has to take into account which information he/she and the speaker share in their common ground (CG). We here further investigated competing accounts about when and how CG information affects language comprehension by means of reaction time (RT) measures, accuracy data, event-related potentials (ERPs), and eye-tracking. Early integration accounts would predict that CG information is considered immediately and would hence not expect to find costs of CG integration. Late integration accounts would predict a rather late and effortful integration of CG information during the parsing process that might be reflected in integration or updating costs. Other accounts predict the simultaneous integration of privileged ground (PG) and CG perspectives. We used a computerized version of the referential communication game with object triplets of different sizes presented visually in CG or PG. In critical trials (i.e., conflict trials), CG information had to be integrated while privileged information had to be suppressed. Listeners mastered the integration of CG (response accuracy 99.8%). Yet, slower RTs, and enhanced late positivities in the ERPs showed that CG integration had its costs. Moreover, eye-tracking data indicated an early anticipation of referents in CG but an inability to suppress looks to the privileged competitor, resulting in later and longer looks to targets in those trials, in which CG information had to be considered. Our data therefore support accounts that foresee an early anticipation of referents to be in CG but a rather late and effortful integration if conflicting information has to be processed. We show that both perspectives, PG and CG, contribute to socially situated language processing and discuss the data with reference to theoretical accounts and recent findings on the use of CG information for reference resolution.