Day 2

The paper discusses the process of information structure acquisition by Russian native children at the age from 3;7 to 7;6. The research is based on the material collected at a series of experiments with Russian monolingual children attending public kindergartens and schools in Saint-Petersburg and Cherepovets (Vologda region). The series comprised experiments of 4 different designs that allowed obtaining a corpus of more than 200 narratives both spoken and written.

Each spoken narrative was audio and video recorded, each subject had to retell a story without any previous preparation. Written narratives were collected at school rooms, and children had to write their narratives right after watching the cartoon or after looking through the picture book without any pervious discussion of their future story. Russian school curriculum does not include any focused training of storytelling in primary school, and at that age children are at their first stage of written skills mastering.

The problem of narrative organization has been widely studied on the material of different languages (cf. Berman, Slobin 1994). The previous studies of spoken narratives in Russian have shown that children at the age of 3;7-5;6 tend to retell only those episodes with the principal character, and this character serves as a kind of anchor (cf. Hickmann 2003) to organize the narrative coherence. At the same time they tend to build the narrative cohesion with the help of the so called “narrative topic” – a time or place markers that start each utterance but do not perform any syntactic or information role within it (Eismont et al. 2018). Only after the age of 5;6 Russian children start producing narratives with other means of cohesion – such as anaphora referencing, conjunctions, etc. (cf. Bamberg 1987; Narasimhan, Dimroth 2008; van Dam 2010 for other languages).

Studies by Berman and Ravid (2009), Jisa (2004) and others have shown that the structure of written narratives differs from spoken narratives in different stages of language acquisition. The presentation proves this difference between written and spoken narratives in Russian material and shows that children tend to use the same ways of coherence and cohesion organization in their first written narratives as they used to perform at the first stages of spoken narrative skills acquisition.

The study explores the syntactic-discursive environments of Hebrew pronominal subjects in children’s conversational interactions. We assume that the manifestation of grammatical subjects is usagebased, that is, context-bound by the type of language being acquired, language-specific verb inflectional morphology as representing pragmatic information, and discursive context relative to conversation. This was a corpus-based study of children’s peer talk between 2-8 years, based on the idea that peer conversation presents a unique picture of linguistic distributions, rendering children active participants without receiving adult scaffolding (Veneziano, 2018).

We focus on the indexical 1st and 2nd pronominal subjects, which are cognitively accessible (Ariel, 2004) with no option of lexical alternatives. A mixed effects logistic model shows that while the structural environment fits both zero and pronominal subjects, zero subjects generally prevail. Specifically, the chances for using a pronominal or zero subject are affected by: (i) participants’ age, as chances for pronominal subjects increase with age; (ii) verb tense: narrative and commentary in past tense vs. future tense expressing plans and intentions; (iii) subject number and person, indicating reference to the speaker, hearer, and the peer group as a unified agentive entity; and (iv) the conversational role of the utterance, underscoring the difference between initiating a topic and responding in conversation. Furthermore, we show that different usage patterns of pronominal subjects are associated with different age groups.

The study argues that the grammatical category of "subject" is not a uniform, all-or-nothing grammatical concept. Rather, the morphological realization of pronominal subjects is related to both form (inflectional morphology) and function (the deployment of pragmatic information in conversation) vis a vis developing communicational needs. In sum, usage patterns of pronominal and zero subjects by Hebrew-speaking children are shown to depend on cognitive and psycholinguistic development. These patterns reflect children’s growing command of the grammatical options in their language, side by side with the consolidation of conversational skills, wider reference abilities, and richer content.

Narrating personal stories plays an important role in human communication since it serves as a vehicle for sharing life experiences and construing one’s identity (e.g, Schiff, 2012). Yet, the devices for marking relations between events do not appear immediately in language, and some markers develop at a later stage than others (Dachkovsky et al. 2019). Here we investigate the grammaticalization of a specific time expression in two age groups of Israeli Sign Language (ISL) users, paying special attention to the central role played by prosody in this process.

Grammaticalization of a variety of function words in many languages follows a conceptual trajectory space > time > text (Traugott & Hopper 2003). For example, the English temporal conjunction while has Proto-Germanic *hwilo (originally “rest, bed”) as its source. Still later, the temporal while grammaticalized into a marker of adversative textual relations. However, one aspect which has received little attention in this area is the role that prosody plays in the development of temporal markers in language.

In spoken languages prosodic features are usually not documented historically. Therefore, we refer to a young language in the visual modality, ISL, to address this question. In sign languages, prosody is conveyed by the rhythmic manual cues as well as by facial expression and head movement (Nespor & Sandler 1999). Here we trace the grammaticalization of a particular time expression TIME-PASS by analyzing episode transitions in personal narratives produced by 15 ISL signers representing two different age groups.

The analysis shows that the function of the expression changes from a whole proposition signaling the unfolding of time to a temporal conjunction connecting discourse units. These two functions are characterized by distinct prosodic behavior. TIME-PASS produced by older signers typically comprises a whole intonational phrase by itself, accompanied by lowered brows and head down. For younger signers, the duration of TIME-PASS is reduced, and it is incorporated into a larger intonational unit. In this later form, the sign is accompanied by raised head.

These findings demonstrate that TIME-PASS is undergoing grammaticalization in ISL. Our results provide further support for the prominence of the metaphor ‘time is motion through space’ in human cognition while also showing how signers reanalyze and expand its meaning to convey ‘motion through discourse.’ On a more general level, however, we aim to contribute to the discussion on the relation between cognition, grammaticalization and prosody.

This study is part of an interactional linguistic (Couper-Kuhlen & Selting 2018) investigation of clicks ([ʘ], [|], [!], etc.) in audio- and video-recordings from the Haifa (Multimodal) Corpus of Spoken Hebrew (Maschler et al. 2020a,b). We find clicks serving a wide range of functions, including negation, repair initiation (Schegloff, Jefferson & Sacks 1977), and transformative responses (Stivers & Hayashi 2010). We focus here on an additional function – constructing diverse frame shifts (Goffman 1981) in interaction. Wright (2005, 2011) described “new sequence indexing” clicks in English, which “demarcate the onset of new and disjunctive sequences,” adopting Schegloff’s definition of sequences as “courses of action implemented through talk” (2007:3). E.g., she shows clicks separating the greeting sequence opening a phone call from the reason-for-call. A significantly more prominent frame shift occurs in the following Hebrew example:

245 Alon: ..ʔoyoyoyoy.

{singing}

246 ...(kisses baby)

247 (inhales, straightens back and looks up, then at Hillel)

248 ...[|]

249 ..ze ma šeʔani carix.

“that’s what I need.”

250 Hillel: (looks up at Alon)

251 Alon: ...xala?,

“challah?,”

252 ...šilav?,

“Shilav [baby goods store]?,”

253 ...sidur rexev.

“car errand.”

Alon and Hillel are in their living room with their baby. Alon is singing to the baby (245-246), and Hillel is looking down. Alon renews the conversation with an inbreath and marked change in posture and orientation (247), followed by a click (248) and a reversed pseudocleft (249, cf. Hopper & Thompson 2008). Precisely at this moment, Hillel displays attention by looking at Alon (248), who proceeds to list his chores for the weekend (251-253). The actions pursued before and after this click could hardly be more distinct, rendering this a major interactional frame shift.

Clicks in Hebrew are also found, however, in more subtle frame shifts, such as switches from one reported speaker to another in constructed dialogue (Tannen 1989), where the parts delimited clearly belong to the same “course of action.” In such contexts clicks typically coincide with less marked resources, in terms of both grammar and embodied conduct.

In our talk we characterize several types of frame shift which routinely feature clicks, in varying orders of magnitude (cf. Maschler 1997). Significant shifts, as in our example, are indicated by multiple marked resources, such as clicks, complex syntax and pronounced changes in gaze and posture. In subtle frame shifts, a click may be the only liminal linguistic element signaling the repositioning.

During every-day communication, many of the utterances we encounter are corrupted by production or perception errors. Yet, most of the time we manage to recover the intended meaning from noisy input. Recent studies suggest that readers this do so using rational (Bayesian) inference [1-2], such that a sentence’s interpretation can be pulled towards a "near neighbor" if that would render the sentence more probable. The current study investigates whether readers assume an error occurred in the input, rather than anticipate a grammatical but rare resolution of the sentence.

We consider the processing of Hebrew relative clauses in which the verb mismatches the filler in number and examine if readers are willing to construct a subject relative (SR), ignoring the agreement mismatch. Grammatical features of Hebrew enable us to consider two cases in which further input may allow a grammatical alternative analysis. One such possibility is the occurrence of a post-verbal subject, producing a grammatical, but rare, object relative (OR). Another possibility is interpreting the vacant subject position as a plural impersonal null subject, which is relatively more common.

Experiment 1 (N=45, later replicated in an additional experiment with 60 new participants) included sentences with plural fillers and a singular verb in the relative clause, thus ruling out an impersonal reading. We observed increased processing costs at the post-verbal subject, relative to an unambiguous baseline, suggesting that participants constructed a SR and did not predict a post-verbal subject. Turning to cases in which impersonal null subjects are licensed, Experiment 2 (N=60, later replicated in an additional experiment with 45 new participants) revealed a reanalysis costs at the object position, suggesting an OR (rather than an SR) analysis was adopted in these cases of agreement mismatch. Lastly, results of a sentence completion experiment (N=120) exhibited that participants tend to complete sentence preambles with fillerverb mismatch mostly as SRs when only post-verbal subject may restore grammaticality yet produce mostly OR completions impersonal null subjects are possible.

These findings provide support for the noisy channel model and suggest that during incremental parsing, comprehenders (i) maintain uncertainty as to the fidelity of the input (ii) apply elaborate knowledge of the distribution of structures in the language, such that they are willing to compromise agreement to avoid improbable grammatical structures.

Broadly, language comprehension requires comprehenders to take the surface form of an utterance and recover the speaker’s intended message. Comprehenders must decide, among other things, whether an utterance corresponds to actual literal truth or whether the intended meaning is conveyed figuratively (Grice 1979; for less dichotomous approaches see Gibbs 1994; Giora 1997).

Here we ask how a sentence-internal cue impacts comprehenders’ resolution of ambiguity between a meaning derivable from a straightforward composition of the sentence’s semantic elements and an interpretation requiring inference. The expressions we target are ones that, semantically at least, signal the speaker’s commitment to the truth of the current proposition (a commitment expected of speakers in general across contexts). We test comprehenders’ interpretation of metaphoric descriptions following expressions like real/literal/actual: Do these adjectives support literal or non-literal interpretations? The results highlight the —perhaps counter-intuitive— role that cues to message truth can play in helping listeners move beyond literal truth, possibly via speaker emphasis on the aptness of the metaphor.

Methods: Our study compares descriptions across three conditions. (1) He is a deceitful weasel… [metaphor-supporting]

(2) He is a real weasel… [truth-endorsing]

(3) He is a furry weasel… [non-metaphoric]

An adverbial signals the temporary nature of the description (e.g., … sometimes at work), rendering the metaphor-supporting condition more sensible than the non-metapahoric condition. Fillers prevent participants from learning that truth-endorsing adjectives consistently precede metaphoric descriptions (e.g., He is a real childminder sometimes after work; She is an actual volunteer sometimes for charity). We report self-paced reading data (Drummond 2013), with two additional experiments underway: one to establish metaphor conventionality (e.g., "Does He is a deceitful weasel describe a person or an animal?") and another to replicate the reading time study with new items and to test whether the observed effects vary with conventionality. Results below model the reading times at the disambiguating adverbial (sometimes) using a linear mixed regression (N=23 participants).

Results: The truth-endorsing adjectives pattern with the metaphor-supporting adjectives, yielding reading times that are statistically indistinguishable from metaphor-supporting adjectives (p=0.86) and reliably faster than non-metaphoric ones (p=0.03), contra an account in which truth-endorsing adjectives’ semantics provide the primary meaning during incremental processing. As such, a cue whose semantics signal a message’s literal truth is shown to ease non-literal processing. The results attest to comprehenders’ usage-driven awareness of the multiple roles of real/literal/actual, an understudied adjective class (cf. analysis of privative adjecives about falsehood, e.g., fake/counterfeit; Partee 2010).

Ample evidence indicates that while processing linguistic input, comprehenders pre-activate words based on their probability of occurrence in the upcoming input, which facilitates the processing of predictable words (e.g. Kutas & Hillyard, 1984). However, recent studies also show the consequences of prediction failure, i.e. that when a strong prediction is formed, the occurrence of an unexpected word entails measurable costs (e.g. Federmeier et al., 2007; Ness & MeltzerAsscher, 2018). Different situations may thus vary in how beneficial it is for a comprehender to engage in prediction. For example, in a situation where unexpected input is often encountered, the costs of prediction failure may outweigh the benefit derived from facilitation of processing successfully predicted words.

In the current study we hypothesized that comprehenders adapt to the likelihood of prediction failure in the experimental environment (i.e. the environment's predictive validity). In two experiments (N=120, in Hebrew; N = 150, in English), we tested whether repeated disconfirmation of predictions in high constraint contexts (i.e. contexts that have a highly probable continuation) results in lesser engagement in strong prediction, reducing prediction failure costs in subsequent high constraint contexts in the experiment. Trials consisted of two-word phrases, in which the first word was either highly constraining (e.g. global, creating a strong prediction for warming) or not (e.g. vegetable, not creating any strong prediction). In each experiment, the proportion of unexpected words in high constraint contexts (e.g. epidemic following global) was manipulated between participants (two experimental lists/groups in Experiment 1, and three in Experiment 2).

The results showed that higher proportion of unexpected words in high constraint contexts led to overall reduced prediction failure costs, as hypothesized. To account for the trial-by-trial data, we formulated a Bayesian adaptation model (e.g. Delaney-Busch et al., 2019), in which the comprehender iteratively updates their belief about predictive validity in the current situation. The estimate of predictive validity increases when the predictable word appears in a high constraint context, and decreases when an unexpected word appears in such contexts. This estimate of predictive validity is then used to weigh the strength of the subsequent prediction. The model outperformed different models which did not include predictive validity in accounting for the data.

We conclude that adaptation is triggered by the disconfirmation of strong predictions, and that comprehenders rationally adapt to the statistics of the environment by using their belief about predictive validity to determine how strongly they should engage in prediction.

Consistent with embodied theories of language processing, several studies have shown that language comprehenders automatically activate perceptual-visual information about verbally described objects during sentence reading in their first language (L1; e.g., Zwaan, Stanfield & Yaxley, 2002).

The current study aimed to investigate the extent to which second language (L2) comprehension involves perceptual-visual activations, and to examine the relative contribution of the two cerebral hemispheres to this process in the L1 and the L2.

Late Hebrew-English bilinguals performed a sentence-picture verification task in either their L1 (Hebrew) or L2 (English). In the task, they decided in each trial whether a pictured object was mentioned in a preceding sentence. In critical trials, the pictured object (e.g., balloon) was indeed mentioned in the sentence (e.g., The boy saw the balloon in the air), however, its shape could have either matched (e.g., inflated balloon) or mismatched (e.g., deflated balloon) the shape implied by the sentence. Faster responses in match relative to mismatch trials (i.e., the shape effect) suggest that readers activate perceptual visual information during sentence reading, and thus, the visual processing of pictures in match trials is facilitated.

In Experiment 1, pictures were presented centrally, to both hemispheres. In Experiment 2, pictures were presented laterally, either in the right visual field (RVF) to the left hemisphere (LH), or in the left visual field (LVF) to the right hemisphere (RH).

Under central presentations (Experiment 1), responses were significantly faster in the match than in the mismatch condition, but only in the L1-Hebrew task. Furthermore, under lateral presentations (Experiment 2), a significant shape effect was observed, irrespective of the target language, and the effect was stronger in the LVF/RH than in the RVF/LH, in both languages.

These findings suggest that (1) under normal reading conditions (i.e., the joint processing of both hemispheres) the activation of perceptual-visual features during sentence reading is significantly reduced in the L2, in comparison to the L1; (2) the RH may be more crucial than the LH for the activation of implied visual information during sentence reading; (3) The RH may be more involved in L1 than in L2 sentence reading. As a result, the contribution of the RH to visual simulations is less pronounced during normal L2 reading and embodiment effects are reduced, at least in the case of late bilinguals

In the world’s languages, in temporal and spatial contexts locative markers such as adpositions are often dropped: optionally, like in Latin, uenisse Athenis in Ephesum ‘that she came from Athens to Ephesus’ (Luraghi 2010: 33) or obligatorily, (*in) next year, cf. in 2021 (Haspelmath 1997; Hagège 2010; Mel’čuk 2018).

Our quantitative study is aimed at understanding the conditions on Preposition-drop (P-drop) in the Russian speech of people from Daghestanian villages. Daniel et al. 2010, based on data collected from three villages, cite it as a prominent characteristic of this variety of Russian and explain it by the interference with the Nakh-Daghestanian morphological system. Our data come from the DagRus corpus (http://www.parasolcorpus.org/dagrus/) consisting of interviews with speakers from 25 villages, with a Nakh-Daghestanian or Turkic language as L1 (228 thousand tokens, 31 hour). So far, we have closely analyzed the interviews coming from native speakers of Kumyk and Azerbaijani (Turkic) that did not figure in previous research. P-drop is widely attested in their speech as well, predominantly, in spatial and temporal contexts -- a pattern similar to the one observed by Daniel et al.

(1) kogda my byvajem Azerbajdžane...

when we COP.PRS.PL Azerbaijan.LOC

‘when we are in Azerbaijan...’

(2) Pervyj istočnik о Darvage ja našol

first source about Darwag.LOC 1SG found

prošlom godu.

last.LOC year.LOC

‘I found the first source about Darwag last year.’

Since Turkic languages are typologically distinct from Nakh-Daghestanian languages, we may conclude that the propensity to drop prepositions does not directly depend on the language family -- rather, it is a feature of the specific variety of Russian used as a lingua franca in rural Daghestan. Notably, P-drop is also observed in other contact varieties of Russian (Shagal 2016; Khomchenkova et al. 2017). We might speculate that the typological tendency of optional locative marking of spatial and temporal locations may be more pronounced in languages in the situation of contact. A possible motivation for the observed pattern is that adpositions in such contexts have quite abstract, ‘empty’ semantics: time- and place-referring expressions have the necessary semantics built in; relatedly, it is not predictable which adposition(s) a language chooses to employ in such contexts: there is much idiosyncrasy involved (Haspelmath 1997). Given that certain temporal expressions in Standard Russian require or allow P-drop, non-native speakers might extend this strategy and choose a formally simpler, preposition-less variant whenever functionally possible.

Many usage-based models of language processing predict difficulty during comprehension when the input requires an infrequent or surprising interpretation. Here we tested this prediction by contrasting the processing of null pronouns and overt pronouns in Brazilian Portuguese (BP). Over the past thirty years, a large body of linguistic studies has demonstrated that speakers of BP, unlike speakers of other romance null-subject languages such as European Portuguese, Spanish and Italian, have been shifting their preference from using null pronouns towards using full pronouns in the subject position in finite sentences.

Our first study employed an acceptability-rating task to verify that BP speakers rate two-sentence discourses with full pronouns less acceptable than sentences with null pronouns when the antecedent is salient (the subject of the previous sentence) but not when the antecedent is not salient (the object of the previous sentence). Our second study tested participants from the same population in a self-paced reading experiment using the same items. This study found slower reading of discourses with overt pronoun references than discourses with null pronoun references to salient antecedents, but not to non-salient antecedents. In our third study, we conducted a corpus analysis of usage patterns, which revealed that while the use of null pronouns in the constructions and with the verbs we used is still prevalent in written language, it is less frequent in spoken language and recent printed news.

Together, these results show that despite BP speakers’ meta-linguistics preference for overt pronouns, their self-paced reading times still show a strong preference for null pronouns during actual comprehension. Since this difference was observed only when antecedents were salient, these findings cannot be attributed to a simple difference between items with null and full pronouns. The discrepancy between speakers’ judgments and their actual processing presents a challenge to simple usage based account, which would attribute both measures to usage frequency.

We propose instead that, at present, pronoun processing in Brazilian Portuguese (and indeed any other language) is best explained by a pragmatic principle stating that overall reading times reflect the balance between initial processing cost and discourse function, and that, due to the transitory state of overt pronouns in Brazilian Portuguese, overt pronouns are still more costly to process. We note that our results are not necessary inconsistent with usage-based accounts, but nevertheless show that the effects of changing usage patterns on meta-linguistic judgments is more immediate than on actual processing.

The goal of my research program is to understand the computations and representations that enable us to share complex thoughts with one another via language, and their neural implementation. A decade ago, I developed a robust new approach to the study of language in the brain based on identifying language-responsive cortex functionally in individual participants. Originally developed for fMRI, we have since extended this approach to other modalities, like electrocorticography. Using this functional-localization approach, I identified and characterized a set of frontal and temporal brain areas that i) support language comprehension and production (spoken and written); ii) are robustly separable from the lower-level perceptual (e.g., speech processing) and motor (e.g., articulation) brain areas; iii) are spatially and functionally similar across diverse languages (>40 languages from 11 language families); and iv) form a functionally integrated system with substantial redundancy across different components.

I will highlight three key findings from our work. First, I will show that the language brain regions are highly selective for language over diverse non-linguistic processes—from math and music, to executive processes, to non-verbal semantic cognition, and even processing computer code—while also showing a deep and intriguing link with a system that supports social cognition. Second, I will show that, contra many leading accounts, the language regions support both understanding of word meanings and sentence-structure building, with no part of the language network being selective for syntactic processing. Further, the 'temporal integration window' of the language system is only a few words long—in line with the fact that most dependencies among words are local across the world’s languages—and appears to be relatively insensitive to word order.

Finally, I will present recent evidence of predictive coding in the language network during naturalistic comprehension, and show that state-of-the-art artificial neural network language models—optimized for predictive processing—accurately capture neural responses during language comprehension. The latter line of work is a critical first step to developing mechanistic accounts of language comprehension.