Refine
Year of publication
- 2022 (42) (remove)
Document Type
- Article (37)
- Doctoral Thesis (3)
- Bachelor Thesis (1)
- Postprint (1)
Language
- English (42) (remove)
Is part of the Bibliography
- yes (42)
Keywords
- German (3)
- alternatives (3)
- focus (3)
- word order (3)
- Individual differences (2)
- N400 (2)
- Prosody (2)
- information structure (2)
- interference (2)
- processing (2)
Institute
- Department Linguistik (42) (remove)
Neural conversation models aim to predict appropriate contributions to a (given) conversation by using neural networks trained on dialogue data. A specific strand focuses on non-goal driven dialogues, first proposed by Ritter et al. (2011): They investigated the task of transforming an utterance into an appropriate reply. Then, this strand evolved into dialogue system approaches using long dialogue histories and additional background context. Contributing meaningful and appropriate to a conversation is a complex task, and therefore research in this area has been very diverse: Serban et al. (2016), for example, looked into utilizing variable length dialogue histories, Zhang et al. (2018) added additional context to the dialogue history, Wolf et al. (2019) proposed a model based on pre-trained Self-Attention neural networks (Vasvani et al., 2017), and Dinan et al. (2021) investigated safety issues of these approaches. This trend can be seen as a transformation from trying to somehow carry on a conversation to generating appropriate replies in a controlled and reliable way.
In this thesis, we first elaborate the meaning of appropriateness in the context of neural conversation models by drawing inspiration from the Cooperative Principle (Grice, 1975). We first define what an appropriate contribution has to be by operationalizing these maxims as demands on conversation models: being fluent, informative, consistent towards given context, coherent and following a social norm. Then, we identify different targets (or intervention points) to achieve the conversational appropriateness by investigating recent research in that field.
In this thesis, we investigate the aspect of consistency towards context in greater detail, being one aspect of our interpretation of appropriateness.
During the research, we developed a new context-based dialogue dataset (KOMODIS) that combines factual and opinionated context to dialogues. The KOMODIS
dataset is publicly available and we use the data in this thesis to gather new insights in context-augmented dialogue generation.
We further introduced a new way of encoding context within Self-Attention based neural networks. For that, we elaborate the issue of space complexity from knowledge graphs,
and propose a concise encoding strategy for structured context inspired from graph neural networks (Gilmer et al., 2017) to reduce the space complexity of the additional context. We discuss limitations of context-augmentation for neural conversation models, explore the characteristics of knowledge graphs, and explain how we create and augment knowledge graphs for our experiments.
Lastly, we analyzed the potential of reinforcement and transfer learning to improve context-consistency for neural conversation models. We find that current reward functions need to be more precise to enable the potential of reinforcement learning, and that sequential transfer learning can improve the subjective quality of generated dialogues.
Dynamical models make specific assumptions about cognitive processes that generate human behavior. In data assimilation, these models are tested against timeordered data. Recent progress on Bayesian data assimilation demonstrates that this approach combines the strengths of statistical modeling of individual differences with the those of dynamical cognitive models.
Young infants can segment continuous speech with statistical as well as prosodic cues. Understanding how these cues interact can be informative about how infants solve the segmentation problem. Here we investigate how German-speaking adults and 9-month-old German-learning infants weigh statistical and prosodic cues when segmenting continuous speech. We measured participants' pupil size while they were familiarized with a continuous speech stream where prosodic cues were pitted off against transitional probabilities. Adult participants' changes in pupil size synchronized with the occurrence of prosodic words during the familiarization and the temporal alignment of these pupillary changes was predictive of adult participants' performance at test. Further, 9-month-olds as a group failed to consistently segment the familiarization stream with prosodic or statistical cues. However, the variability in temporal alignment of the pupillary changes at word frequency showed that prosodic and statistical cues compete for dominance when segmenting continuous speech. A followup language development questionnaire at 40 months of age suggested that infants who entrained to prosodic words performed better on a vocabulary task and those infants who relied more on statistical cues performed better on grammatical tasks. Together these results suggest that statistics and prosody may serve different roles in speech segmentation in infancy.
In 2019 the Journal of Memory and Language instituted an open data and code policy; this policy requires that, as a rule, code and data be released at the latest upon publication. How effective is this policy? We compared 59 papers published before, and 59 papers published after, the policy took effect. After the policy was in place, the rate of data sharing increased by more than 50%. We further looked at whether papers published under the open data policy were reproducible, in the sense that the published results should be possible to regenerate given the data, and given the code, when code was provided. For 8 out of the 59 papers, data sets were inaccessible. The reproducibility rate ranged from 34% to 56%, depending on the reproducibility criteria. The strongest predictor of whether an attempt to reproduce would be successful is the presence of the analysis code: it increases the probability of reproducing reported results by almost 40%. We propose two simple steps that can increase the reproducibility of published papers: share the analysis code, and attempt to reproduce one's own analysis using only the shared materials.
Pronouns can sometimes covary with a non c-commanding quantifier phrase (QP). To obtain such 'telescoping' readings, a semantic representation must be computed in which the QP's semantic scope extends beyond its surface scope. Non-native speakers have been claimed to have more difficulty than native speakers deriving such non-isomorphic syntax-semantics mappings, but evidence from processing studies is scarce. We report the results from an eye-movement monitoring experiment and an offline questionnaire investigating whether native and non-native speakers of German can link personal pronouns to non c-commanding QPs inside relative clauses. Our results show that both participant groups were able to obtain telescoping readings offline, but only the native speakers showed evidence of forming telescoping dependencies during incremental parsing. During processing the non-native speakers focused on a discourse-prominent, non-quantified alternative antecedent instead. The observed group differences indicate that non-native comprehenders have more difficulty than native comprehenders computing scope-shifted representations in real time.
The picture-word interference paradigm (participants name target pictures while ignoring distractor words) is often used to model the planning processes involved in word production. The participants' naming times are delayed in the presence of a distractor (general interference). The size of this effect depends on the relationship between the target and distractor words. Distractors of the same semantic category create more interference (semantic interference), and distractors overlapping in phonology create less interference (phonological facilitation). The present study examined the relationships between these experimental effects, processing times, and attention in order to better understand the cognitive processes underlying participants' behavior in this paradigm. Participants named pictures with a superimposed line of Xs, semantically related distractors, phonologically related distractors, or unrelated distractors. General interference, semantic interference, and phonological facilitation effects were replicated. Distributional analyses revealed that general and semantic interference effects increase with naming times, while phonological facilitation decreases. The phonological facilitation and semantic interference effects were found to depend on the synchronicity in processing times between the planning of the picture's name and the processing of the distractor word. Finally, electroencephalographic power in the alpha band before stimulus onset varied with the position of the trial in the experiment and with repetition but did not predict the size of interference/facilitation effects. Taken together, these results suggest that experimental effects in the picture-word interference paradigm depend on processing times to both the target word and distractor word and that distributional patterns could partly reflect this dependency.
In this paper we examine the effect of uncertainty on readers’ predictions about meaning. In particular, we were interested in how uncertainty might influence the likelihood of committing to a specific sentence meaning. We conducted two event-related potential (ERP) experiments using particle verbs such as turn down and manipulated uncertainty by constraining the context such that readers could be either highly certain about the identity of a distant verb particle, such as turn the bed […] down, or less certain due to competing particles, such as turn the music […] up/down. The study was conducted in German, where verb particles appear clause-finally and may be separated from the verb by a large amount of material. We hypothesised that this separation would encourage readers to predict the particle, and that high certainty would make prediction of a specific particle more likely than lower certainty. If a specific particle was predicted, this would reflect a strong commitment to sentence meaning that should incur a higher processing cost if the prediction is wrong. If a specific particle was less likely to be predicted, commitment should be weaker and the processing cost of a wrong prediction lower. If true, this could suggest that uncertainty discourages predictions via an unacceptable cost-benefit ratio. However, given the clear predictions made by the literature, it was surprisingly unclear whether the uncertainty manipulation affected the two ERP components studied, the N400 and the PNP. Bayes factor analyses showed that evidence for our a priori hypothesised effect sizes was inconclusive, although there was decisive evidence against a priori hypothesised effect sizes larger than 1μV for the N400 and larger than 3μV for the PNP. We attribute the inconclusive finding to the properties of verb-particle dependencies that differ from the verb-noun dependencies in which the N400 and PNP are often studied.
Studies of word production often make use of picture-naming tasks, including the picture-word-interference task. In this task, participants name pictures with superimposed distractor words. They typically need more time to name pictures when the distractor word is semantically related to the picture than when it is unrelated (the semantic interference effect). The present study examines the distributional properties of this effect in a series of Bayesian meta-analyses. Meta-analytic estimates of the semantic interference effect first show that the effect is present throughout the reaction time distribution and that it increases throughout the distribution. Second, we find a correlation between a participant's mean semantic interference effect and the change in the effect in the tail of the reaction time distribution, which has been argued to reflect the involvement of selective inhibition in the naming task. Finally, we show with simulated data that this correlation emerges even when no inhibition is used to generate the data, which suggests that inhibition is not needed to explain this relationship.
It was not until the 1960s and 70s of the 20th century that researchers turned their special interest to colloquial Russian (hereafter CR) and its interaction with codified (normative) Russian. Colloquial Russian uses its grammatical constructions in deviation from the norms of the written language. Since codified language is the basis of colloquial language on the grammatical level, among others, the question arises, how the standard forms are used in oral speech. Lapteva (1976) has looked in particular at the syntax of CR and made a classification of CR constructions that differ from their standard forms. The present study deals with two constructions from this classification: an embedded temporal subordinate clause and a temporal subordinate clause with the meaningless conjunction kogda (as/if), which leaves its normative position in the sentence. In addition to the special forms of temporal adverbial clauses, the frequency of their standard implementation as preceding and the following constructions will be examined. Two hypotheses were formulated:
• The frequency of certain constructions classified by Lapteva (1976) as transitional constructions decreases over decades.
• The ratio between prefixed and suffixed temporal subordinate clauses will be in favor of the latter due to the spontaneity of oral speech. The corpus study was conducted with the oral language sub-corpus of the National'nyj Korpus Russkogo Jazyka (National Corpus of the Russian Language). No evidence of a correlation between the number of CR constructions and the year of recording was found either in the whole oral sub-corpus or in its largest section - the collection of private conversations. The proportion of prefixed temporal constructions was greatest in both public and non-public corpora compared to postfixed ones. The study did not provide evidence for the hypotheses put forward, due to the limitations of the corpus study, such as missing or incomplete context of the conversations, lack of punctuation and/or marking of intonation.