Refine
Document Type
- Other (17) (remove)
Language
- English (17) (remove)
Is part of the Bibliography
- yes (17)
Keywords
- Basic English (1)
- C. K. ogden (1)
- Danish (1)
- Dutch (1)
- Dänisch (1)
- English dialects (1)
- Exploratory interfaces (1)
- Generalized additive mixed-effects modeling (1)
- Icelandic (1)
- International language (1)
Institute
- Department Linguistik (17) (remove)
Zero-shot learning in Language & Vision is the task of correctly labelling (or naming) objects of novel categories. Another strand of work in L&V aims at pragmatically informative rather than "correct" object descriptions, e.g. in reference games. We combine these lines of research and model zero-shot reference games, where a speaker needs to successfully refer to a novel object in an image. Inspired by models of "rational speech acts", we extend a neural generator to become a pragmatic speaker reasoning about uncertain object categories. As a result of this reasoning, the generator produces fewer nouns and names of distractor categories as compared to a literal speaker. We show that this conversational strategy for dealing with novel objects often improves communicative success, in terms of resolution accuracy of an automatic listener.
Speech scientists have long noted that the qualities of naturally-produced vowels do not remain constant over their durations regardless of being nominally "monophthongs" or "diphthongs". Recent acoustic corpora show that there are consistent patterns of first (F1) and second (F2) formant frequency change across different vowel categories. The three Australian English (AusE) close front vowels /i:, 1, i/ provide a striking example: while their midpoint or mean F1 and F2 frequencies are virtually identical, their spectral change patterns distinctly differ. The results indicate that, despite the distinct patterns of spectral change of AusE /i:, i, la/ in production, its perceptual relevance is not uniform, but rather vowel-category dependent.
The keynote article (Mayberry & Kluender, 2017) makes an important contribution to questions concerning the existence and characteristics of sensitive periods in language acquisition. Specifically, by comparing groups of non-native L1 and L2 signers, the authors have been able to ingeniously disentangle the effects of maturation from those of early language exposure. Based on L1 versus L2 contrasts, the paper convincingly argues that L2 learning is a less clear test of sensitive periods. Nevertheless, we believe Mayberry and Kluender underestimate the evidence for maturational factors in L2 learning, especially that coming from recent research.
The Gradient Symbolic Computation (GSC) model presented in the keynote article (Goldrick, Putnam & Schwarz) constitutes a significant theoretical development, not only as a model of bilingual code-mixing, but also as a general framework that brings together symbolic grammars and graded representations. The authors are to be commended for successfully integrating a theory of grammatical knowledge with the voluminous research on lexical co-activation in bilinguals. It is, however, unfortunate that a certain conception of bilingualism was inherited from this latter research tradition, one in which the contrast between native and non-native language takes a back seat.
Predictive coding and its generalization to active inference offer a unified theory of brain function. The underlying predictive processing paradigmhas gained significant attention in artificial intelligence research for its representation learning and predictive capacity. Here, we suggest that it is possible to integrate human and artificial generative models with a predictive coding network that processes sensations simultaneously with the signature of predictive coding found in human neuroimaging data. We propose a recurrent hierarchical predictive coding model that predicts low-dimensional representations of stimuli, electroencephalogram and physiological signals with variational inference. We suggest that in a shared environment, such hybrid predictive coding networks learn to incorporate the human predictive model in order to reduce prediction error. We evaluate the model on a publicly available EEG dataset of subjects watching one-minute long video excerpts. Our initial results indicate that the model can be trained to predict visual properties such as the amount, distance and motion of human subjects in videos.
Words as social tools
(2019)
We compare Visual Berrypicking, an interactive approach allowing users to explore large and highly faceted information spaces using similarity-based two-dimensional maps, with traditional browsing techniques. For large datasets, current projection methods used to generate maplike overviews suffer from increased computational costs and a loss of accuracy resulting in inconsistent visualizations. We propose to interactively align inexpensive small maps, showing local neighborhoods only, which ideally creates the impression of panning a large map. For evaluation, we designed a web-based prototype for movie exploration and compared it to the web interface of The Movie Database (TMDb) in an online user study. Results suggest that users are able to effectively explore large movie collections by hopping from one neighborhood to the next. Additionally, due to the projection of movie similarities, interesting links between movies can be found more easily, and thus, compared to browsing serendipitous discoveries are more likely.
Voice onset time (VOT), a primary cue for voicing in many languages including English and German, is known to vary greatly between speakers, but also displays robust within-speaker consistencies, at least in English. The current analysis extends these findings to German. VOT measures were investigated from voiceless alveolar and velar stops in CV syllables cued by a visual prompt in a cue-distractor task. Comparably to English, a considerable portion of German VOT variability can be attributed to the syllable’s vowel length and the stop’s place of articulation. Individual differences in VOT still remain irrespective of speech rate. However, significant correlations across places of articulation and between speaker-specific mean VOTs and standard deviations indicate that talkers employ a relatively unified VOT profile across places of articulation. This could allow listeners to more efficiently adapt to speaker-specific realisations.
Since the Shallow Structure Hypothesis (SSH) was first put forward in 2006, it has inspired a growing body of research on grammatical processing in nonnative (L2) speakers. More than 10 years later, we think it is time for the SSH to be reconsidered in the light of new empirical findings and current theoretical assumptions about human language processing. The purpose of our critical commentary is twofold: to clarify some issues regarding the SSH and to sketch possible ways in which this hypothesis might be refined and improved to better account for L1 and L2 speakers’ performance patterns.
We investigated online electrophysiological components of distributional learning, specifically of tones by listeners of a non tonal language. German listeners were presented with a bimodal distribution of syllables with lexical tones from a synthesized continuum based on Cantonese level tones. Tones were presented in sets of four standards (within-category tokens) followed by a deviant (across-category token). Mismatch negativity (MMN) was measured. Earlier behavioral data showed that exposure to this bimodal distribution improved both categorical perception and perceptual acuity for level tones [I]. In the present study we present analyses of the electrophysiological response recorded during this exposure, i.e., the development of the MMN response during distributional learning. This development over time is analyzed using Generalized Additive Mixed Models and results showed that the MMN amplitude increased for both within and across-category tokens, reflecting higher perceptual acuity accompanying category formation. This is evidence that learners zooming in on phonological categories undergo neural changes associated with more accurate phonetic perception.
Clause typing in Germanic
(2018)
The questionnaire investigates the functional left periphery of various finite clauses in Germanic languages, with particular attention paid to clause-typing elements and the combinations thereof. The questionnaire is mostly concerned with clause typing in embedded clauses, but main clause counterparts are also considered for comparative purposes. The chief aim was to achieve comparable results across Germanic languages, though the standardised questionnaire may also be helpful in the study of other languages, too. Most questions examine the availability of various complementisers and clause-typing operators, and in some cases the movement of verbs to the left periphery is also taken into account. The questionnaire is split into seven major parts according to the types of clauses under scrutiny.
All instructions were given in English and the individual questions either concern translations of given sentences from English into the target language, and/or they ask for specific details about the constructions in the target language.
The present document contains the questionnaire itself (together with the instructions given at the beginning of the questionnaire and at the beginning of the individual sections, as well as the questions asking for personal data), the sociolinguistic data of the speakers, and the actual results for the individual languages. Five Germanic languages are included: Dutch, Danish, Icelandic, Norwegian and Swedish. For each language, two informants were recruited. Given the small number of informants, the present study serves as a qualitative investigation and as a basis for further, quantitative and experimental studies.