Search

Recently, deep learning models have increasingly been used in cognitive modelling of language. This study asks whether deep learning can help us to better understand the learning problem that needs to be solved by speakers, above and beyond linear methods. We utilise the Discriminative Lexicon Model (DLM, Baayen et al., 2019), which models comprehension and production with mappings between numeric form and meaning vectors. While so far, these mappings have been linear (Linear Discriminative Learning, LDL), in the present study we replace them with deep dense neural networks (Deep Discriminative Learning, DDL). We find that DDL affords more accurate mappings for large and diverse datasets from English and Dutch, but not necessarily for Estonian and Taiwan Mandarin. DDL outperforms LDL in particular for words with pseudo-morphological structure such as slend+er. Applied to average reaction times, we find that DDL is outperformed by frequency-informed linear mappings (FIL). However, DDL trained in a frequency-informed way ('frequency-informed' deep learning, FIDDL) substantially outperforms FIL. Finally, while linear mappings can very effectively be updated from trial-to-trial to model incremental lexical learning (Heitmeier et al., 2023), deep mappings cannot do so as effectively. At present, both linear and deep mappings are informative for understanding language., Comment: 17 pages, 6 figures

In Mandarin, the tonal contours of monosyllabic words produced in isolation or in careful speech are characterized by four lexical tones: a high-level tone (T1), a rising tone (T2), a dipping tone (T3) and a falling tone (T4). However, in spontaneous speech, the actual tonal realization of monosyllabic words can deviate significantly from these canonical tones due to intra-syllabic co-articulation and inter-syllabic co-articulation with adjacent tones. In addition, Chuang et al. (2024) recently reported that the tonal contours of disyllabic Mandarin words with T2-T4 tone pattern are co-determined by their meanings. Following up on their research, we present a corpus-based investigation of how the pitch contours of monosyllabic words are realized in spontaneous conversational Mandarin, focusing on the effects of contextual predictors on the one hand, and the way in words' meanings co-determine pitch contours on the other hand. We analyze the F0 contours of 3824 tokens of 63 different word types in a spontaneous Taiwan Mandarin corpus, using the generalized additive (mixed) model to decompose a given observed pitch contour into a set of component pitch contours. We show that the tonal context substantially modify a word's canonical tone. Once the effect of tonal context is controlled for, T2 and T3 emerge as low flat tones, contrasting with T1 as a high tone, and with T4 as a high-to-mid falling tone. The neutral tone (T0), which in standard descriptions, is realized based on the preceding tone, emerges as a low tone in its own right, modified by the other predictors in the same way as the standard tones T1, T2, T3, and T4. We also show that word, and even more so, word sense, co-determine words' F0 contours. Analyses of variable importance using random forests further supported the substantial effect of tonal context and an effect of word sense.

In Standard Chinese, Tone 3 (the dipping tone) becomes Tone 2 (rising tone) when followed by another Tone 3. Previous studies have noted that this sandhi process may be incomplete, in the sense that the assimilated Tone 3 is still distinct from a true Tone 2. While Mandarin Tone 3 sandhi is widely studied using carefully controlled laboratory speech (Xu, 1997) and more formal registers of Beijing Mandarin (Yuan and Chen, 2014), less is known about its realization in spontaneous speech, and about the effect of contextual factors on tonal realization. The present study investigates the pitch contours of two-character words with T2-T3 and T3-T3 tone patterns in spontaneous Taiwan Mandarin conversations. Our analysis makes use of the Generative Additive Mixed Model (GAMM, Wood, 2017) to examine fundamental frequency (f0) contours as a function of normalized time. We consider various factors known to influence pitch contours, including gender, speaking rate, speaker, neighboring tones, word position, bigram probability, and also novel predictors, word and word sense (Chuang et al., 2024). Our analyses revealed that in spontaneous Taiwan Mandarin, T3-T3 words become indistinguishable from T2-T3 words, indicating complete sandhi, once the strong effect of word (or word sense) is taken into account. For our data, the shape of f0 contours is not co-determined by word frequency. In contrast, the effect of word meaning on f0 contours is robust, as strong as the effect of adjacent tones, and is present for both T2-T3 and T3-T3 words.

The pitch contours of Mandarin two-character words are generally understood as being shaped by the underlying tones of the constituent single-character words, in interaction with articulatory constraints imposed by factors such as speech rate, co-articulation with adjacent tones, segmental make-up, and predictability. This study shows that tonal realization is also partially determined by words' meanings. We first show, on the basis of a Taiwan corpus of spontaneous conversations, using the generalized additive regression model, and focusing on the rise-fall tone pattern, that after controlling for effects of speaker and context, word type is a stronger predictor of pitch realization than all the previously established word-form related predictors combined. Importantly, the addition of information about meaning in context improves prediction accuracy even further. We then proceed to show, using computational modeling with context-specific word embeddings, that token-specific pitch contours predict word type with 50% accuracy on held-out data, and that context-sensitive, token-specific embeddings can predict the shape of pitch contours with 30% accuracy. These accuracies, which are an order of magnitude above chance level, suggest that the relation between words' pitch contours and their meanings are sufficiently strong to be functional for language users. The theoretical implications of these empirical findings are discussed.

Word frequency is a strong predictor in most lexical processing tasks. Thus, any model of word recognition needs to account for how word frequency effects arise. The Discriminative Lexicon Model (DLM; Baayen et al., 2018a, 2019) models lexical processing with linear mappings between words' forms and their meanings. So far, the mappings can either be obtained incrementally via error-driven learning, a computationally expensive process able to capture frequency effects, or in an efficient, but frequency-agnostic solution modelling the theoretical endstate of learning (EL) where all words are learned optimally. In this study we show how an efficient, yet frequency-informed mapping between form and meaning can be obtained (Frequency-informed learning; FIL). We find that FIL well approximates an incremental solution while being computationally much cheaper. FIL shows a relatively low type- and high token-accuracy, demonstrating that the model is able to process most word tokens encountered by speakers in daily life correctly. We use FIL to model reaction times in the Dutch Lexicon Project (Keuleers et al., 2010) and find that FIL predicts well the S-shaped relationship between frequency and the mean of reaction times but underestimates the variance of reaction times for low frequency words. FIL is also better able to account for priming effects in an auditory lexical decision task in Mandarin Chinese (Lee, 2007), compared to EL. Finally, we used ordered data from CHILDES (Brown, 1973; Demuth et al., 2006) to compare mappings obtained with FIL and incremental learning. The mappings are highly correlated, but with FIL some nuances based on word ordering effects are lost. Our results show how frequency effects in a learning model can be simulated efficiently, and raise questions about how to best account for low-frequency words in cognitive models., Comment: 32 pages, 12 figures, 3 tables; revised version

Visual grounding of Language aims at enriching textual representations of language with multiple sources of visual knowledge such as images and videos. Although visual grounding is an area of intense research, inter-lingual aspects of visual grounding have not received much attention. The present study investigates the inter-lingual visual grounding of word embeddings. We propose an implicit alignment technique between the two spaces of vision and language in which inter-lingual textual information interacts in order to enrich pre-trained textual word embeddings. We focus on three languages in our experiments, namely, English, Arabic, and German. We obtained visually grounded vector representations for these languages and studied whether visual grounding on one or multiple languages improved the performance of embeddings on word similarity and categorization benchmarks. Our experiments suggest that inter-lingual knowledge improves the performance of grounded embeddings in similar languages such as German and English. However, inter-lingual grounding of German or English with Arabic led to a slight degradation in performance on word similarity benchmarks. On the other hand, we observed an opposite trend on categorization benchmarks where Arabic had the most improvement on English. In the discussion section, several reasons for those findings are laid out. We hope that our experiments provide a baseline for further research on inter-lingual visual grounding., Comment: - added more results - paper accepted to appear at UM-IoS workshop, EMNLP 2022

Distributional semantics offers new ways to study the semantics of morphology. This study focuses on the semantics of noun singulars and their plural inflectional variants in English. Our goal is to compare two models for the conceptualization of plurality. One model (FRACSS) proposes that all singular-plural pairs should be taken into account when predicting plural semantics from singular semantics. The other model (CCA) argues that conceptualization for plurality depends primarily on the semantic class of the base word. We compare the two models on the basis of how well the speech signal of plural tokens in a large corpus of spoken American English aligns with the semantic vectors predicted by the two models. Two measures are employed: the performance of a form-to-meaning mapping and the correlations between form distances and meaning distances. Results converge on a superior alignment for CCA. Our results suggest that usage-based approaches to pluralization in which a given word's own semantic neighborhood is given priority outperform theories according to which pluralization is conceptualized as a process building on high-level abstraction. We see that what has often been conceived of as a highly abstract concept, [+plural], is better captured via a family of mid-level partial generalizations., Comment: 29 pages including references, 24 pages excluding references, 11 Figures, 3 Tables. This article is under review in "The Mental Lexicon" journal

Trial-to-trial effects have been found in a number of studies, indicating that processing a stimulus influences responses in subsequent trials. A special case are priming effects which have been modelled successfully with error-driven learning (Marsolek, 2008), implying that participants are continuously learning during experiments. This study investigates whether trial-to-trial learning can be detected in an unprimed lexical decision experiment. We used the Discriminative Lexicon Model (DLM; Baayen et al., 2019), a model of the mental lexicon with meaning representations from distributional semantics, which models error-driven incremental learning with the Widrow-Hoff rule. We used data from the British Lexicon Project (BLP; Keuleers et al., 2012) and simulated the lexical decision experiment with the DLM on a trial-by-trial basis for each subject individually. Then, reaction times were predicted with Generalised Additive Models (GAMs), using measures derived from the DLM simulations as predictors. We extracted measures from two simulations per subject (one with learning updates between trials and one without), and used them as input to two GAMs. Learning-based models showed better model fit than the non-learning ones for the majority of subjects. Our measures also provide insights into lexical processing and individual differences. This demonstrates the potential of the DLM to model behavioural data and leads to the conclusion that trial-to-trial learning can indeed be detected in unprimed lexical decision. Our results support the possibility that our lexical knowledge is subject to continuous changes., Comment: 48 pages, 13 figures; revised version

Semantic differentiation of nominal pluralization is grammaticalized in many languages. For example, plural markers may only be relevant for human nouns. English does not appear to make such distinctions. Using distributional semantics, we show that English nominal pluralization exhibits semantic clusters. For instance, pluralization of fruit words is more similar to one another and less similar to pluralization of other semantic classes. Therefore, reduction of the meaning shift in plural formation to the addition of an abstract plural meaning is too simplistic. A semantically informed method, called CosClassAvg, is introduced that outperforms pluralization methods in distributional semantics which assume plural formation amounts to the addition of a fixed plural vector. In comparison with our approach, a method from compositional distributional semantics, called FRACSS, predicted plural vectors that were more similar to the corpus-extracted plural vectors in terms of direction but not vector length. A modeling study reveals that the observed difference between the two predicted semantic spaces by CosClassAvg and FRACSS carries over to how well a computational model of the listener can understand previously unencountered plural forms. Mappings from word forms, represented with triphone vectors, to predicted semantic vectors are more productive when CosClassAvg-generated semantic vectors are employed as gold standard vectors instead of FRACSS-generated vectors., Comment: 45 pages (including references), 14 figures. This article is under review at `Morphology'

This paper presents three case studies of modeling aspects of lexical processing with Linear Discriminative Learning (LDL), the computational engine of the Discriminative Lexicon model (Baayen et al., 2019). With numeric representations of word forms and meanings, LDL learns to map one vector space onto the other, without being informed about any morphological structure or inflectional classes. The modeling results demonstrated that LDL not only performs well for understanding and producing morphologically complex words, but also generates quantitative measures that are predictive for human behavioral data. LDL models are straightforward to implement with the JudiLing package (Luo et al., 2021). Worked examples are provided for three modeling challenges: producing and understanding Korean verb inflection, predicting primed Dutch lexical decision latencies, and predicting the acoustic duration of Mandarin words.

This study addresses a series of methodological questions that arise when modeling inflectional morphology with Linear Discriminative Learning. Taking the semi-productive German noun system as example, we illustrate how decisions made about the representation of form and meaning influence model performance. We clarify that for modeling frequency effects in learning, it is essential to make use of incremental learning rather than the endstate of learning. We also discuss how the model can be set up to approximate the learning of inflected words in context. In addition, we illustrate how in this approach the wug task can be modeled in considerable detail. In general, the model provides an excellent memory for known words, but appropriately shows more limited performance for unseen data, in line with the semi-productivity of German noun inflection and generalization performance of native German speakers., Comment: 38 pages, 5 figures, 10 tables; acknowledgements added

Thul et al. (2020) called attention to problems that arise when chronometric experiments implementing specific factorial designs are analysed with the generalized additive mixed model (GAMM), using factor smooths to capture trial-to-trial dependencies. From a series of simulations incorporating such dependencies, they conclude that GAMMs are inappropriate for between-subject designs. They argue that in addition GAMMs come with too many modeling possibilities, and advise using the linear mixed model (LMM) instead. We address the questions raised by Thul et al. (2020), who clearly demonstrated that problems can indeed arise when using factor smooths in combination with factorial designs. We show that the problem does not arise when using by-smooths. Furthermore, we have traced a bug in the implementation of factor smooths in the mgcv package, which will have been removed from version 1.8-36 onwards. To illustrate that GAMMs now produce correct estimates, we report simulation studies implementing different by-subject longitudinal effects. The maximal LMM emerges as slightly conservative compared to GAMMs, and GAMMs provide estimated coefficients that can be less variable across simulation runs. We also discuss two datasets where time-varying effects interact with numerical predictors in a theoretically informative way. Furthermore, we argue that the wide range of tools that GAMMs make available to researcher across all domains of scientific inquiry do not come with uncontrolled researcher degrees of freedom once confronted with a specific psycholinguistic datasets. We also introduce a distinction between replicable and non-replicable non-linear effects. We conclude that GAMMs are an excellent and reliable tool for understanding experimental data, including chronometric data with time-varying effects., Comment: 29 pages, 6 figures, 14 tables

Language grounding aims at linking the symbolic representation of language (e.g., words) into the rich perceptual knowledge of the outside world. The general approach is to embed both textual and visual information into a common space -the grounded space-confined by an explicit relationship between both modalities. We argue that this approach sacrifices the abstract knowledge obtained from linguistic co-occurrence statistics in the process of acquiring perceptual information. The focus of this paper is to solve this issue by implicitly grounding the word embeddings. Rather than learning two mappings into a joint space, our approach integrates modalities by determining a reversible grounded mapping between the textual and the grounded space by means of multi-task learning. Evaluations on intrinsic and extrinsic tasks show that our embeddings are highly beneficial for both abstract and concrete words. They are strongly correlated with human judgments and outperform previous works on a wide range of benchmarks. Our grounded embeddings are publicly available here., Comment: To be published in the 25th Conference on Computational Natural Language Learning (CoNLL 2021)

Recent research in the field of spiking neural networks (SNNs) has shown that recurrent variants of SNNs, namely long short-term SNNs (LSNNs), can be trained via error gradients just as effective as LSTMs. The underlying learning method (e-prop) is based on a formalization of eligibility traces applied to leaky integrate and fire (LIF) neurons. Here, we show that the proposed approach cannot fully unfold spike timing dependent plasticity (STDP). As a consequence, this limits in principle the inherent advantage of SNNs, that is, the potential to develop codes that rely on precise relative spike timings. We show that STDP-aware synaptic gradients naturally emerge within the eligibility equations of e-prop when derived for a slightly more complex spiking neuron model, here at the example of the Izhikevich model. We also present a simple extension of the LIF model that provides similar gradients. In a simple experiment we demonstrate that the STDP-aware LIF neurons can learn precise spike timings from an e-prop-based gradient signal.

A problem that tends to be ignored in the statistical analysis of experimental data in the language sciences is that responses often constitute time series, which raises the problem of autocorrelated errors. If the errors indeed show autocorrelational structure, evaluation of the significance of predictors in the model becomes problematic due to potential anti-conservatism of p-values. This paper illustrates two tools offered by Generalized Additive Mixed Models (GAMMs) (Lin and Zhang, 1999; Wood, 2006, 2011, 2013) for dealing with autocorrelated errors, as implemented in the current version of the fourth author's mgcv package (1.8.9): the possibility to specify an ar(1) error model for Gaussian models, and the possibility of using factor smooths for random-effect factors such as subject and item. These factor smooths are set up to have the same smoothing parameters, and are penalized to yield the non-linear equivalent of random intercepts and random slopes in the classical linear framework. Three case studies illustrate these issues., Comment: 10 figures