No description available.

The notion that prediction plays a role in language processing is getting less controversial, however research and discussion is ongoing as to the nature and extent of its involvement. Previous studies have mainly focused on prediction during comprehension in restricted paradigms. The current fMRI study makes use of the information theoretic measure of surprisal to investigate the neural basis of predictability during conversation. A computational model determined surprisal for each word of 288 minutes (4.8 hours) of existing conversational data summed up over twenty-four healthy participants and a confederate. Areas sensitive to surprisal for comprehension were the bilateral superior temporal and middle temporal gyri. In production, areas sensitive to surprisal were the bilateral supplementary motor area and sensorimotor areas, importantly extending to the inferior frontal cortex in the left hemisphere. We conclude that predictability during language processing plays a role at the level of word form also during conversation. A novel finding is an asymmetry for the surprisal effects on production and comprehension, respectively. This thesis also shows the viability of combining computational and neuroscientific methods to study linguistic data from naturalistic language use. / Uppfattningen om att förutsägelser är i bruk vid språkprocessning blir allt minde kontroversiell, men forskning och diskussion pågår fortfarande om dess natur och omfattning. Tidigare studier har främst fokuserat på förutsägelse vid förståelse i begränsade paradigm. Den aktuella fMRI-studien använder det informationsteroretiska måttet surprisal för att undersöka de neurala korrelaten för förutsägbarhet under konversation. En beräkningsmodell tog fram surprisalvärden för varje enskilt ord från 288 minuter (4,8 timmar) av befintliga konversationsdata summerat över tjugofyra friska deltagare och en forskningsledare. Områden som var känsliga för surprisal vid förståelse var bilaterala superiora temporala och middle temporala gyrus. Vid produktion var områden känsliga för surprisal de bilaterala supplementära motorområdena och sensorimotor områdena, av vikt också med utbredning till inferiora frontala cortex i vänstra hemisfären. Vi drar slutsatsen att förutsägbarhet under språkprocessning är i bruk på ord-nivå, också under konversation. En ny upptäckt är en asymmetri för surprisaleffekter på produktion respektive förståelse. Denna uppsats visar också på användbarheten av att kombinera beräknings- och neurovetenskapliga metoder för att studera lingvistisk data från naturligt språkbruk.

Information theoretical approaches to language state that the most efficient communication oc­curs when the amount of information transmitted is distributed as uniformly as possible over time. Previous research has shown that speakers tend to adhere to strategies for distributing information efficiently, using mechanisms at multiple linguistic levels. This study aims to in­vestigate whether articulation rate (AR) is used in continuous speech to achieve a more uniform distribution of information within sentences, quantified as surprisal estimated by the state­-of-­the-­art language model GPT-­2, and if this has an effect on the amplitude of the N400 brain response in listeners. In neurolinguistics, surprisal has been observed to be a good predictor of the N400, which is related to processing of semantics and meaning in general. The results showed a significant, though small, effect of surprisal on AR, indicating that AR may have some role in achieving more uniform distribution of information on the word level. In line with previous research, surprisal showed an effect on the N400 where higher surprisal led to larger amplitudes. Results regarding AR and distributional effects on the N400 were inconclusive, although some independent effects of AR were found that could be further explored in more controlled experimental settings. / Informationsteoretiska perspektiv på språk säger att den mest effektiva kommunikationen sker när information sänds ut så jämnt fördelat som möjligt över tid. Tidigare studier har visat att talare tenderar att följa vissa strategier för att distribuera information jämnt, vilket de gör på flera språkliga nivåer. Denna studie ämnar att undersöka om artikulationshastighet (eng. articulation rate (AR)) används i kontinuerligt tal för att uppnå en mer jämn distribution av information inom meningar, kvantifierat som informationsteoretisk surprisal med hjälp av språkmodellen GPT-­2, samt om detta ger effekt på hjärnresponsen N400:s amplitud hos lyssnare. Inom neurolingvistik har surprisal visats kunna predicera N400, som är kopplad till bearbetning av semantik och meningsfullhet generellt. Resultaten visade en signifikant, om än liten, effekt av surprisal på AR, en indikator på att AR kan ha en roll i att uppnå mer jämn distribution av information på ordnivå. I linje med tidigare forskning så hade surprisal en inverkan på N400, där högre surprisal gav större amplituder. Resultaten utifrån AR och distribution av information var inte entydiga, däremot observerades vissa självständiga effekter av AR på amplituden av N400 och dessa skulle kunna vidare undersökas i mer kontrollerade experiment.

Child-directed speech (CDS) differs from adult-directed speech (ADS) in several respects whose possible facilitating effects for language acquisition are still being studied. One such difference concerns articulation rate --- the number of linguistic units by the number of time units, excluding pauses --- which has been shown to be generally lower than in ADS. However, while it is well-established that ADS exhibits an inverse relation between articulation rate and information-theoretic surprisal --- the amount of information encoded in a linguistic unit --- this measure has been conspicuously absent in the study of articulation rate in CDS. Another issue is if the lower articulation rate in CDS is stable across utterances or an effect of local variation, such as final lengthening. The aim of this work is to arrive at a more comprehensive model of articulation rate in CDS by including surprisal and final lengthening. In particular, one-word utterances were studied, also in relation to word-length effects (the phenomenon that longer words generally have a higher articulation rate). To this end, a methodology for large-scale automatic phoneme-alignment was developed, which was applied to two longitudinal corpora of Swedish CDS. It was investigated i) how articulation rate in CDS varied with respect to child age, ii) whether there was a negative relation between articulation rate and surprisal in CDS, and iii) to what extent articulation rate was lower in CDS than in ADS. The results showed i) a weak positive effectof child age on articulation rate, ii) a negative relation between articulation rate and surprisal, and iii) that there was a lower articulation rate in CDS but that the difference could almost exclusively be attributed to one-word utterances and final lengthening. In other words, adults seem to adapt how fast they speak to their children's age, speaking faster to children is correlated with a reduced amount of information, and the difference in articulation rate between CDS and ADS is most prominent in isolated words and final lengthening. More generally, the results suggest that CDS is well-suited for word segmentation, since lower articulation rate in one-word utterances provides an additional cue.

This dissertation investigates how grammatical functions in transitive sentences (i.e., `subject' and `direct object') are distributed in written Swedish discourse with respect to morphosyntactic as well as semantic and referential (i.e., prominence-based) information. It also investigates how assignment of grammatical functions during on-line comprehension of transitive sentences in Swedish is influenced by interactions between morphosyntactic and prominence-based information. In the dissertation, grammatical functions are assumed to express role-semantic (e.g., Actor and Undergoer) and discourse-pragmatic (e.g., Topic and Focus) functions of NP arguments. Grammatical functions correlate with prominence-based information that is associated with these functions (e.g., animacy and definiteness). Because of these correlations, both prominence-based and morphosyntactic information are assumed to serve as argument interpretation cues during on-line comprehension. These cues are utilized in a probabilistic fashion. The weightings, interplay and availability of them are reflected in their distribution in language use, as shown in corpus data. The dissertation investigates these assumptions by using various methods in a triangulating fashion. The first contribution of the dissertation is an ERP (event-related brain potentials) experiment that investigates the ERP response to grammatical function reanalysis, i.e., a revision of a tentative grammatical function assignment, during on-line comprehension of transitive sentences. Grammatical function reanalysis engenders a response that correlates with the (re-)assignment of thematic roles to the NP arguments. This suggests that the comprehension of grammatical functions involves assigning role-semantic functions to the NPs. The second contribution is a corpus study that investigates the distribution of prominence-based, verb-semantic and morphosyntactic features in transitive sentences in written discourse. The study finds that overt morphosyntactic information about grammatical functions is used more frequently when the grammatical functions cannot be determined on the basis of word order or animacy. This suggests that writers are inclined to accommodate the understanding of their recipients by more often providing formal markers of grammatical functions in potentially ambiguous sentences. The study also finds that prominence features and their interactions with verb-semantic features are systematically distributed across grammatical functions and therefore can predict these functions with a high degree of confidence. The third contribution consists of three computational models of incremental grammatical function assignment. These models are based upon the distribution of argument interpretation cues in written discourse. They predict processing difficulties during grammatical function assignment in terms of on-line change in the expectation of different grammatical function assignments over the presentation of sentence constituents. The most prominent model predictions are qualitatively consistent with reading times in a self-paced reading experiment of Swedish transitive sentences. These findings indicate that grammatical function assignment draws upon statistical regularities in the distribution of morphosyntactic and prominence-based information in language use. Processing difficulties in the comprehension of Swedish transitive sentences can therefore be predicted on the basis of corpus distributions.

No description available.