Prior Language Knowledge Research Articles

Generative attention based framework for implicit language change detection

Spoken language change detection (LCD) refers to detecting language switching points in a multilingual speech signal. Most approaches in literature use the explicit framework that requires the modeling of intermediate phonemes and Senones to distinguish language. However, such techniques are limited when used with resource scare/ zero resource languages. Hence as an alternative, this study explores implicit frameworks to perform LCD. The focus of this work is to detect language change when a single speaker is speaking two languages. In this direction, a subjective study is performed to analyze the method humans adapt to discriminate languages. The outcome of the subjective study suggests humans require more neighborhood duration to detect language change. The initial observation suggests, that detecting language change is challenging using the baseline implicit unsupervised distance-based approach. Inspired by human cognition, prior language knowledge is integrated into the computational framework through the Gaussian mixture model and universal background model (GMM-UBM), temporal information via attention, and pattern storage using the Generative adversarial network (GAN) to enhance language discrimination. The experimental results on the Microsoft code-switched (MSCS) dataset show, compared to the unsupervised distance-based approach, the performance of the proposed LCD relatively improved by 19.3%, 47.3%, and 50.7% using the GMM-UBM, attention, and GAN-attention based framework, respectively.

Revealing sign language for individuals who are deaf and mute through the utilization of neural networks

Living without effective communication poses significant challenges for humans. Individuals employ various methods to convey and share their thoughts and ideas between the sender and receiver. Two of the most prevalent means of communication are verbal speech, which relies on auditory perception, and non-verbal communication through gestures involving bodily movements such as hand gestures and facial expressions. Sign language, specifically categorized as a gestural language, is a unique form of communication that relies on visual perception for understanding and expression. While many individuals incorporate gestures into their communication, for deaf individuals, sign language is often their primary and essential means of communication. Individuals who are deaf and dumb rely on communication to interact with others, gain knowledge, and engage in their surroundings. Sign language serves as a crucial link that reduces the distance between them and the broader society. In order to enhance this communication, we've created models with the ability to identify sign language gestures and translate them into conventional text. Through the training of these models on a dataset employing neural networks, remarkable outcomes have been attained. This technology enables individuals, without prior knowledge of sign language, to understand the intentions and messages of individuals with disabilities, fostering greater inclusivity and accessibility in our society. Three algorithms were used to achieve this work and the findings show very good outcomes, i.e. Random Forest at 98%, Logistic Regression at 99%, and Decision Tree at 91%. Keywords: Machine Learning, Sign Language, Random Forest, Neural Network, Logistic Regression Deaf and Dumb, Decision Tree

Hierarchical visual-semantic interaction for scene text recognition

Proper interaction between visual and semantic features is crucial to obtain a powerful feature representation for scene text recognition (STR). The existing interaction methods usually treat visual and semantic features as distinct tokens and use transformers to learn contextual information and prior language knowledge, and they achieve promising performance for STR task. However, there still remain several issues needed to be further addressed such as the imbalance number and mis-alignment between visual and semantic features, and the necessarily of stacking transformers to a progressive improvement in accuracy. To this aim, this paper proposes a novel interaction manner namely hierarchical visual-semantic interaction (HVSI) which contains three novel modules including a hierarchical visual-semantic interaction module, fusion module, and visual-semantic alignment module. The hierarchical visual-semantic interaction module employs multiple visual-semantic interaction blocks in various scales to enhance the representation power of visual features and semantic features. To better exploit multi-scale visual and semantic features, the fusion module is introduced to fuse multiple semantic features based on attention mechanisms. Furthermore, our HVSI presents a simple plug-in block named visual-semantic alignment module to alleviate mis-alignment of semantic features by mapping them into a unified semantic space, which helps improve the performance of HVSI. Extensive experiments on multiple benchmarks including English and Chinese text recognition datasets show that our method obtains state-of-the-art or competitive performances.

Utilizing prior language knowledge: A study of how Thai students translate unknown languages into English

This study aimed to explore how Thai university students applied their prior language knowledge when translating a text from Spanish, an unknown language, into English. The research involved a group of 34 students, whose second language was English, and their third language was French. The assessment of translated texts included quantitative analyses, such as measuring translation accuracy and conducting means comparisons, as well as qualitative investigations via post-task interviews. The results revealed that students with an intermediate level of proficiency in French achieved significantly higher translation accuracy compared to beginners. Insights gleaned from post-task interviews indicated that participants recognized similarities across lexical, phonological, and structural aspects among English, French, and Spanish. These findings highlight the cognitive advantages of exposing students to multiple languages and offer guidance to curriculum developers in higher education institutions seeking to cultivate multilingual graduates. Keywords: English; French; language acquisition; language transfers; multilingualism; translation task.

Dynamic brain responses to Russian word acquisition among Chinese adult learners: An event-related potential study.

Human learners are capable to acquire foreign language vocabulary at an impressive speed even in adulthood. Previous studies have examined the neural mechanisms underlying rapid acquisition of Latin-alphabet vocabulary and revealed dynamic changes in several event-related potential (ERP) components during novel word learning. However, scant attention has been paid to the acquisition of Russian words. The present study used ERP and examined dynamic brain responses to rapid Russian word acquisition in 53 native Chinese speakers with no prior knowledge of Russian language. Behavioral data showed robust individual differences in Russian word acquisition, with most participants being able to rapidly learn a subset of novel Russian words in a few exposures. ERP results revealed significant learning effects in the P200, N400, and P600 amplitudes. Moreover, P600 amplitude changes predicted participants' word acquisition after learning. These findings demonstrated dynamic brain responses to rapid Russian word learning and suggested that the P600 component may serve as a bio-marker for individual learning ability in Russian word acquisition.

ProtoCLIP: Prototypical Contrastive Language Image Pretraining.

Contrastive language image pretraining (CLIP) has received widespread attention since its learned representations can be transferred well to various downstream tasks. During the training process of the CLIP model, the InfoNCE objective aligns positive image-text pairs and separates negative ones. We show an underlying representation grouping effect during this process: the InfoNCE objective indirectly groups semantically similar representations together via randomly emerged within-modal anchors. Based on this understanding, in this article, prototypical contrastive language image pretraining (ProtoCLIP) is introduced to enhance such grouping by boosting its efficiency and increasing its robustness against the modality gap. Specifically, ProtoCLIP sets up prototype-level discrimination between image and text spaces, which efficiently transfers higher level structural knowledge. Furthermore, prototypical back translation (PBT) is proposed to decouple representation grouping from representation alignment, resulting in effective learning of meaningful representations under a large modality gap. The PBT also enables us to introduce additional external teachers with richer prior language knowledge. ProtoCLIP is trained with an online episodic training strategy, which means it can be scaled up to unlimited amounts of data. We trained our ProtoCLIP on conceptual captions (CCs) and achieved an + 5.81% ImageNet linear probing improvement and an + 2.01% ImageNet zero-shot classification improvement. On the larger YFCC-15M dataset, ProtoCLIP matches the performance of CLIP with 33% of training time.

English and German teachers’ language awareness in the Hungarian school context

The study aims to investigate the language use of English and German teachers in the school context. The teachers’ questionnaire has been conducted at minority primary schools and normal primary schools between 2019 and 2020. Quantitative data given by 28 teachers were analysed. German teachers working at minority schools used the target language more than teachers working at regular schools. English participants employed their mother tongue more than German respondents. German teachers used more translation than their English peers. Teachers working at minority schools used less translation than teachers working at normal schools. Most of the foreign language teachers made connections between the languages but only a few teachers exploited frequently the opportunity of prior language knowledge.

Dynamic assessment of multilingual children's word learning

Teachers and clinicians may struggle to provide early identification to support multilingual children's language development. Dynamic assessments are a promising approach to identify and support children's language development. We developed and studied a novel word learning task that is dynamic and language neutral. It makes use of multilingual children's abilities to apply language transfer, fast mapping and socially embedded language to the learning of new words. A total of 26 children attending kindergarten in French participated in this study. Within this group, 13 different home languages were spoken. Children took part in a dynamic assessment task of their word learning that consisted of a test-teach-retest task. Children's scores on this task were compared with their language abilities reported by their parents, amount of language exposure and scores on standardized tests of vocabulary. All tasks were delivered in French. Children had higher accuracy for known words as compared with new words in the task, which may suggest transfer of knowledge from their first language. They also showed increased accuracy in identifying and naming the new words across the three trials, suggesting fast mapping of these new vocabulary items. Finally, the scores on the dynamic task correlated to children's vocabulary scores on the standardized tests, but not parent report of language development, or the amount of exposure to the language of school. This novel dynamic assessment task taps into the process of vocabulary learning, but is less influenced by prior language knowledge. Together, these findings provide insight into early word learning by young multilingual children and proposes a conceptual model for identifying strategies to support second language acquisition. What is already known on the subject Many barriers exist with regards to assessing the language abilities of multilingual children when a clinician aims to assess their language abilities in both languages. An alternative approach is to measure children's language processing abilities. What this paper adds to existing knowledge A novel dynamic and multilingual task was developed and implemented in this study. This task builds on children's word learning abilities that include cross-language transfer, fast-mapping, and socially imbedded learning. This multilingual task was found to tap into vocabulary learning but was not influenced by prior language knowledge. What are the potential or actual clinical implications of this work? Applying a task that focuses on language processing abilities is a promising strategy to capture language abilities in multilingual children. In addition, the dynamic nature of this tasks allows a clinician to identify scaffolding strategies that best support children's word learning.

Intuitive Cognition-Based Method for Generating Speech Using Hand Gestures.

Muteness at its various levels is a common disability. Most of the technological solutions to the problem creates vocal speech through the transition from mute languages to vocal acoustic sounds. We present a new approach for creating speech: a technology that does not require prior knowledge of sign language. This technology is based on the most basic level of speech according to the phonetic division into vowels and consonants. The speech itself is expected to be expressed through sensing of the hand movements, as the movements are divided into three rotations: yaw, pitch, and roll. The proposed algorithm converts these rotations through programming to vowels and consonants. For the hand movement sensing, we used a depth camera and standard speakers in order to produce the sounds. The combination of the programmed depth camera and the speakers, together with the cognitive activity of the brain, is integrated into a unique speech interface. Using this interface, the user can develop speech through an intuitive cognitive process in accordance with the ongoing brain activity, similar to the natural use of the vocal cords. Based on the performance of the presented speech interface prototype, it is substantiated that the proposed device could be a solution for those suffering from speech disabilities.

On the status of transfer in adult third language acquisition of early bilinguals.

The study of linguistic transfer—understood here in terms of the copying of previous linguistic representations—seeks to reveal how domain-relevant prior language knowledge impacts the acquisition and development of new mental representations more generally. Studying sequential multilingualism offers a natural laboratory to observe cognitive-economical mechanisms that avoid redundancy in language learning. One of the key dividing questions between theories of transfer in sequential multilingualism is the extent of transfer, that is, whether a whole previous grammar is transferred (full transfer) or a potentially different source language is selected for each linguistic property (property-by-property transfer). We adopted a novel methodological approach to this question, examining four different linguistic properties from unrelated domains of grammar across the three languages of a heterogeneous population of highly proficient, early Catalan/Spanish bilinguals with different degrees of language dominance and order of acquisition, at the very beginning of (adult) L3 English. Results are variably complex across the different properties, but compatible with a scenario where one of the previous languages, Catalan, was selected as the basis for the initial L3 English grammar of these speakers. We discuss the theoretical implications of these findings.

Crosslinguistic influence and L2 grammar learning: Proposed replications of Ellis and Sagarra (2011) and Tolentino and Tokowicz (2014)

AbstractInvestigations of crosslinguistic effects in second language acquisition (SLA) have contributed rich understandings about the ways in which prior language knowledge and experience can influence additional language learning. Building on this work, one recent line of research has examined the extent to which SLA findings about crosslinguistic influence can be used to improve L2 learning, indicating that explicit instruction tailored to address the nature of the crosslinguistic learning problem can benefit L2 performance. At the same time, however, this line of inquiry is relatively novel and under-researched. The aim of this paper, therefore, is to suggest further research which might advance knowledge and understanding about the ways in which explicit instruction can facilitate second language (L2) learning of crosslinguistically difficult target features. To this end, I lay the ground for a series of close replications designed to understand the role of prior language knowledge and experience in L2 learning: Ellis and Sagarra (2011) and Tolentino and Tokowicz (2014). These studies were selected because they are grounded in empirical SLA research and SLA theory. In addition, the selected studies facilitate replication because they have either (i) made their data collection materials available in digital repositories or (ii) have robustly described their design and provided the stimuli used.

Acquiring Knowledge from Pre-Trained Model to Neural Machine Translation

Pre-training and fine-tuning have achieved great success in natural language process field. The standard paradigm of exploiting them includes two steps: first, pre-training a model, e.g. BERT, with a large scale unlabeled monolingual data. Then, fine-tuning the pre-trained model with labeled data from downstream tasks. However, in neural machine translation (NMT), we address the problem that the training objective of the bilingual task is far different from the monolingual pre-trained model. This gap leads that only using fine-tuning in NMT can not fully utilize prior language knowledge. In this paper, we propose an Apt framework for acquiring knowledge from pre-trained model to NMT. The proposed approach includes two modules: 1). a dynamic fusion mechanism to fuse task-specific features adapted from general knowledge into NMT network, 2). a knowledge distillation paradigm to learn language knowledge continuously during the NMT training process. The proposed approach could integrate suitable knowledge from pre-trained models to improve the NMT. Experimental results on WMT English to German, German to English and Chinese to English machine translation tasks show that our model outperforms strong baselines and the fine-tuning counterparts.

Reading the news on Twitter: Source and item memory for social media in younger and older adults

BackgroundSocial media content is well-remembered, possibly because of its personal relevance and gossipy nature. It is unclear whether the mnemonic advantage of social media extends to a population less familiar with these platforms and whether knowing the content is from social media sources influences memory. This study examined how the presentation of news-like content in social media affected both item and source memory across two age groups. Younger adults (n = 42) and older adults (n = 32) studied tweets and news headlines that appeared in the format of Twitter posts or CNN headlines - these items were designed to be either congruent (e.g., tweets formatted as Twitter posts) or incongruent (e.g., tweets formatted as CNN headlines).ResultsFor item memory, both age groups correctly recognized tweets more than headlines. Source identification was more accurate when format and content were congruent than incongruent. Signal detection analyses indicated that the source advantage for congruent items was largely driven by a bias to select the format that matched the content’s original source and that this tendency was stronger in older adults.ConclusionsThese results replicate previous literature on the mnemonic advantage of social media content. Although both younger and older adults remembered the content of social media better than the content of news sources, older adults were more sensitive than younger adults to congruency effects in source memory. These findings suggest that older adults rely more on their prior knowledge of conventional language and style in traditional and social media.

Ecological perspectives on implementing multilingual pedagogies in adult foreign language classrooms – a comparative case study

ABSTRACT Multilingual pedagogies draw on learners’ prior language knowledge and can offer considerable advantages in learning the target language. Yet language instruction, particularly in foreign language learning contexts, does not often capitalise on the multilingual assets of learners. The goal of this study is to understand what determines the implementation of multilingual pedagogies in two institutes of German as Foreign Language, one in the US and one in India. In this comparative case study, ecological perspectives are adopted to examine what factors impede or foster the use of multilingual pedagogies in adult foreign language instruction in both schools. Data collection methods involved open-ended questionnaires, semi-structured interviews with teachers and an analysis of context-specific documents. Findings revealed that there were encouraging, constraining and contingent factors across both schools. While some factors were common to both schools, and there were other context-specific factors that determined the adoption of multilingual teaching. The results of the study also reflect larger issues such as a gap between theory and practice and a lack of teacher collaboration. The paper concludes with some suggestions on how such issues might be addressed.

Question to Query: Converting Human Language to DBMS Query

In this paper a method has been proposed keeping in the mind the need for systems that could generate structured queries from normal language keeping in mind that the user has no prior knowledge of database query language. A novel method which aims at aiding analyst who aren’t well versed with codes, but need quantitative outputs to analyze, predict and alert the business or market. A python model is used, which aims at converting any sentence typed in English to a query provided that such tables and database is present for query processing. Tree tagging is used here to relate words typed in to SQL query syntax. Any sentence typed in by analyst, it further annotated by parts of speech and lemmas. A list of generic words and stop words is used while parsing the input the sentence and tagging it. Query is generated by simultaneously removing the stop words, mapping the keywords with the one’s used in structured query language. The generated query comes out in form of a JSON file.

The Bilingual Advantage and the Language Background Bias

The idea that bilingualism can give us an advantage in life is of great interest to the scientific community due to its significant positive implications for healthcare and education at large. In recent years, several scholars have provided evidence in favour of the so-called bilingual advantage or benefit, suggesting a positive association between bilingualism and cognitive development. In order to understand whether the claim is fully warranted, the present paper sets out to examine the evidence in support and against the existence of a bilingual benefit for individuals. Following a brief discussion on the use of the terms bilingualism and multilingualism in the literature, the paper proceeds to provide a summary of evidence of advantages and disadvantages currently associated with prior language knowledge in the mind, highlighting some of the possible reasons for the different results that are being reported and introducing the language background bias. The paper ends with some suggestions for future research that can help us move forward and increase our understanding of the bi-/multilingual advantage as a broader phenomenon.

Learning short-text semantic similarity with word embeddings and external knowledge sources

We present a novel method based on interdependent representations of short texts for determining their degree of semantic similarity. The method represents each short text as two dense vectors: the former is built using the word-to-word similarity based on pre-trained word vectors, the latter is built using the word-to-word similarity based on external sources of knowledge. We also developed a preprocessing algorithm that chains coreferential named entities together and performs word segmentation to preserve the meaning of phrasal verbs and idioms. We evaluated the proposed method on three popular datasets, namely Microsoft Research Paraphrase Corpus, STS2015 and P4PIN, and obtained state-of-the-art results on all three without using prior knowledge of natural language, e.g., part-of-speech tags or parse tree, which indicates the interdependent representations of short text pairs are effective and efficient for semantic textual similarity tasks.

Preservation of categorical perception for speech in autism with and without speech onset delay.

Recent accounts of autistic perception, including Bayesian accounts, hypothesize a reduced influence of prior knowledge on perception across different domains in the autism spectrum (AS). The purpose of this study was to investigate the influence of prior linguistic knowledge, in the form of phonemic categorical knowledge, on speech perception in adults with AS condition. As phonemic categorical knowledge is shaped by language experience and abilities, we furthermore distinguished AS participants with (AS-SOD) or without a history of speech onset delay (AS-noSOD); the control group comprises typical individuals matched for age, nonverbal intelligence, and reading abilities. We also controlled for the influence of auditory-verbal short-term retention capacities by administering word list and nonword list repetition tasks. We did not observe any reduced influence of prior phonemic knowledge on the perception of speech stimuli nor did we observed any increased perceptual abilities for atypical variants of speech stimuli or nonspeech auditory stimuli, either between the two autistic groups or relative to the control group. Short-term memory abilities appeared to be superior in the AS-noSOD group relative to the AS-SOD and control groups, but this strength could be accounted for by their higher vocabulary knowledge. The preservation of categorical perception in verbal autistic adults observed in this study challenges models claiming a reduced influence of prior knowledge on perception across domains in the AS. Autism Res 2019. © 2019 International Society for Autism Research, Wiley Periodicals, Inc. LAY SUMMARY: A reduced influence of prior knowledge has been considered to characterize perceptual abilities in people with autism. In this article, we examine this claim by assessing nonlinguistic and linguistic auditory perception abilities in adults with autism, and by further distinguishing between autism with or without a history of delayed language development. We did not observe any reduced influence of prior language knowledge on the perception of speech stimuli nor did we observe any increased perceptual abilities for atypical variants of speech stimuli or nonspeech auditory stimuli, and this relative to a control group matched on age, nonverbal intellectual efficiency, and reading abilities. Our results challenge models claiming a reduced influence of prior knowledge on perception across domains in the AS.

Teachers’ beliefs about multilingual pedagogies and the role of initial training

ABSTRACTTeacher training programmes have often ignored the new multilingual paradigm and, consequently, educators may hold some misconceptions about how additional languages are learnt and should be taught in multilingual contexts. A number of recent studies have investigated language teachers’ beliefs regarding multilingual education [Arocena-Egaña, E., Cenoz, J., & Gorter, D. (2015). Teachers’ beliefs in multilingual education in the Basque country and in Friesland. Journal of Immersion and Content-Based Language Education, 3(2), 169–193; De Angelis, G. (2011). Teachers’ beliefs about the role of prior language knowledge in learning and how these influence teaching practices. International Journal of Multilingualism, 8(3), 216–234; Griva, E., & Chostelidou, D. (2012). Multilingual competence development in the Greek educational system: FL teachers’ beliefs and attitudes. International Journal of Multilingualism, 9(3), 257–271; Haukås, Å. (2016). Teacher’s beliefs about multilingualism and a multilingual pedagogical approach. International Journal of Multilingualism, 13(1), 1–18; Mitits, L. (2018). Multilingual students in Greek school: teachers’ views and teaching practices. Journal of education and e-learning research, 5(1), 28–36. Otwinowska, A. (2014). Does multilingualism influence plurilingual awareness of Polish teachers of English? International Journal of Multilingualism, 11(1), 97–119]. Unlike this extant literature, the current paper focuses not on language teachers, but on a sample of 121 subject teachers-to-be. More specifically, it aims to explore preschool and primary prospective teachers’ beliefs about the learning and teaching of English as an L3 in the Valencian Community. The main data collection tool was a questionnaire, partially based on [Griva, E., & Chostelidou, D. (2012). Multilingual competence development in the Greek educational system: FL teachers’ beliefs and attitudes. International Journal of Multilingualism, 9(3), 257–271] semi-structured interviews. This questionnaire was implemented as a pretest and as a posttest in order to analyse the effect of instruction for multilingual education on subject teachers’ beliefs. Results point to the effectiveness of initial teacher training on tertiary/additional language teaching. However, they also confirm the persistence of some monolingual views concerning key issues, such as the younger-the better myth, as already highlighted in previous studies conducted in similar multilingual contexts (see Arocena-Egaña, E., Cenoz, J., & Gorter, D. (2015). Teachers’ beliefs in multilingual education in the Basque country and in Friesland. Journal of Immersion and Content-Based Language Education, 3(2), 169–193].

Foreign language learning in the third age: A pilot feasibility study on cognitive, socio-affective and linguistic drivers and benefits in relation to previous bilingualism of the learner

While there is a growing body of research on second language acquisition in children and prime-of-life learners, much remains to be explored about how older adults learn a new language and how good additional language learning is for them (see e.g. Mackey & Sachs, 2012). In this study we present the findings of a longitudinal pilot study in which 12 German-speaking subjects, half of them German-Slovenian bilinguals, between 63 and 90 years of age attended a four-week intensive English course for beginners, and were tested in a pre-/post-test design on a range of linguistic, cognitive, and socio-affective parameters. The results of non-parametric statistical tests and qualitative analyses suggest that: (1) the learning of an additional language in the third age can contribute to healthy and active aging, as it has a positive effect on executive function, (linguistic) self-confidence, autonomy, communicative skills and overall well-being, irrespective of age and prior language knowledge (bilingualism); and (2) age-related social, psychological and contextual factors seem to play as significant a role as strictly maturational factors. Such research promises in due course to inform adult educators about the establishment of relevant third-age learner profiles as well as the design of individualized third-age language training.