Modeling the predictive potential of extralinguistic context with script knowledge: The case of fragments

Abstract

We describe a novel approach to estimating the predictability of utterances given extralinguistic context in psycholinguistic research. Predictability effects on language production and comprehension are widely attested, but so far predictability has mostly been manipulated through local linguistic context, which is captured with n-gram language models. However, this method does not allow to investigate predictability effects driven by extralinguistic context. Modeling effects of extralinguistic context is particularly relevant to discourse-initial expressions, which can be predictable even if they lack linguistic context at all. We propose to use script knowledge as an approximation to extralinguistic context. Since the application of script knowledge involves the generation of prediction about upcoming events, we expect that scrips can be used to manipulate the likelihood of linguistic expressions referring to these events. Previous research has shown that script-based discourse expectations modulate the likelihood of linguistic expressions, but script knowledge has often been operationalized with stimuli which were based on researchers’ intuitions and/or expensive production and norming studies. We propose to quantify the likelihood of an utterance based on the probability of the event to which it refers. This probability is calculated with event language models trained on a script knowledge corpus and modulated with probabilistic event chains extracted from the corpus. We use the DeScript corpus of script knowledge to obtain empirically founded estimates of the likelihood of an event to occur in context without having to resort to expensive pre-tests of the stimuli. We exemplify our method at a case study on the usage of nonsentential expressions (fragments), which shows that utterances that are predictable given script-based extralinguistic context are more likely to be reduced.

Cross-Linguistic Trade-Offs and Causal Relationships Between Cues to Grammatical Subject and Object, and the Problem of Efficiency-Related Explanations

“…As for morphosyntactic coding asymmetries and splits, it is well known that more predictable grammatical meanings are expressed by shorter forms (including zero) than less predictable ones (e.g., Jäger, 2007;Kurumada and Jaeger, 2015;Kurumada and Grimm, 2019;Haspelmath, 2021). Lemke et al (2021) demonstrate that fragments (i.e., incomplete sentential structures) encoding events known from everyday scripts and scenarios are perceived as more natural than fragments encoding unpredictable events. See more examples in Hawkins (2004), Jaeger and Tily (2011), and Gibson et al (2019).…”

Predictable Words Are More Likely to Be Omitted in Fragments–Evidence From Production Data

“…Bergen and Goodman (2015) do not apply it to more realistic communication situations which involve more diverse utterances, potentially communicated messages and predictability effects drive by extralinguistic context. Lemke et al (2021) in turn explain the choice as adaptation to the processing 1 In the theoretical literature there is no agreement on a definition of the notions nonsentential utterance or fragment. Researchers diverge in particular with respect to the question of which elliptical utterances are categorized as fragments and with respect to the presence or absence of an explicit antecedent.…”

“…For instance, in the taxi example (1), it seems more natural for the passenger to say to the university than me the university, even though both of these fragments reduce the utterance to a similar extent. While Lemke et al (2021) just show that the reduction of predictable utterances is more acceptable, Bergen and Goodman (2015) include a cost term in their model that penalizes utterances that are effortful to produce. Since Bergen and Goodman (2015) derive a preference for fragments from this cost term, it is most likely intended to be affected by the length of an utterance, but they do not make this explicit or discuss other sources of production effort, like a cost for retrieving unpredictable words (Ferreira and Dell, 2000).…”

“…An information-theoretic approach is particularly promising for two reasons: First, Lemke et al (2021) found that fragments are overall more strongly preferred in predictive contexts. This finding is in line with our account, because in predictive contexts the words within an utterance are overall more likely and thus also more likely to be omitted.…”

“…This finding is in line with our account, because in predictive contexts the words within an utterance are overall more likely and thus also more likely to be omitted. However, Lemke et al (2021) did not look into the more fine-grained predictions of the information-theoretic account on the word level. Second, information-theoretic constraints have been shown to explain the distribution of omissions particularly on the word level, such as those of complementizers, pronouns, articles and case markers (Levy and Jaeger, 2007;Frank and Jaeger, 2008;Jaeger, 2010;Asr and Demberg, 2015;Kurumada and Jaeger, 2015;Norcliffe and Jaeger, 2016;Lemke et al, 2017).…”