Publications

Abstract

Ideophones, also termed “mimetics” or “expressives,” are marked words that depict sensory imagery. They are found in many of the world’s languages, and sizable lexical classes of ideophones are particularly well-documented in languages of Asia, Africa, and the Americas. Ideophones are not limited to onomatopoeia like meow and smack, but cover a wide range of sensory domains, such as manner of motion (e.g., plisti plasta ‘splish-splash’ in Basque), texture (e.g., tsaklii ‘rough’ in Ewe), and psychological states (e.g., wakuwaku ‘excited’ in Japanese). Across languages, ideophones stand out as marked words due to special phonotactics, expressive morphology including certain types of reduplication, and relative syntactic independence, in addition to production features like prosodic foregrounding and common co-occurrence with iconic gestures.

Three intertwined issues have been repeatedly debated in the century-long literature on ideophones. (a) Definition: Isolated descriptive traditions and cross-linguistic variation have sometimes obscured a typologically unified view of ideophones, but recent advances show the promise of a prototype definition of ideophones as conventionalised depictions in speech, with room for language-specific nuances. (b) Integration: The variable integration of ideophones across linguistic levels reveals an interaction between expressiveness and grammatical integration, and has important implications for how to conceive of dependencies between linguistic systems. (c) Iconicity: Ideophones form a natural laboratory for the study of iconic form-meaning associations in natural languages, and converging evidence from corpus and experimental studies suggests important developmental, evolutionary, and communicative advantages of ideophones.

Abstract

Proficient readers typically fixate near the center of a word, with a slight bias towards word onset. We explore a novel account of this phenomenon based on combining information-theory with visual perceptual constraints in a connectionist model of visual word recognition. This account posits that the amount of information-content available for word identification varies across fixation locations and across languages, thereby explaining the overall fixation location bias in different languages, making the novel prediction that certain words are more readily identified when fixating at an atypical fixation location, and predicting specific cross-linguistic differences. We tested these predictions across several simulations in English and Hebrew, and in a pilot behavioral experiment. Results confirmed that the bias to fixate closer to word onset aligns with maximizing information in the visual signal, that some words are more readily identified at atypical fixation locations, and that these effects vary to some degree across languages.

Abstract

Infants learn the meaning of words from accumulated experiences of real-time interactions with their caregivers. To study the effects of visual sensory input on word learning, we recorded infant's view of the world using head-mounted eye trackers during free-flowing play with a caregiver. While playing, infants were exposed to novel label-object mappings and later learning outcomes for these items were tested after the play session. In this study we use a classification based approach to link properties of infants' visual scenes during naturalistic labeling moments to their word learning outcomes. We find that a model which integrates both highly informative and ambiguous sensory evidence is a better fit to infants' individual learning outcomes than models where either type of evidence is taken alone, and that raw labeling frequency is unable to account for the word learning differences we observe. Here we demonstrate how a computational model, using only raw pixels taken from the egocentric scene image, can derive insights on human language learning.

Abstract

Most work on part-of-speech (POS) tagging is focused on high resource languages, or examines low-resource and active learning settings through simulated studies. We evaluate POS tagging techniques on an actual endangered language, Griko. We present a resource that contains 114 narratives in Griko, along with sentence-level translations in Italian, and provides gold annotations for the test set. Based on a previously collected small corpus, we investigate several traditional methods, as well as methods that take advantage of monolingual data or project cross-lingual POS tags. We show that the combination of a semi-supervised method with cross-lingual transfer is more appropriate for this extremely challenging setting, with the best tagger achieving an accuracy of 72.9%. With an applied active learning scheme, which we use to collect sentence-level annotations over the test set, we achieve improvements of more than 21 percentage points

Abstract

Data augmentation techniques have been widely used in visual recognition tasks as it is easy to generate new
data by simple and straight forward image transformations. However, when it comes to text data augmen-
tations, it is difficult to find appropriate transformation techniques which also preserve the contextual and
grammatical structure of language texts. In this paper, we explore various text data augmentation techniques
in text space and word embedding space. We study the effect of various augmented datasets on the efficiency
of different deep learning models for relation classification in text.

Abstract

The Romance languages have a rich numeral system that includes cardinals—providing the bases on which the other types of numeral series are built—ordinals, fractions, collectives, approximatives, distributives, and multiplicatives. Latin plays a decisive and continued role in their formation, both as the language to which many numerals go back directly and as an ongoing source for lexemes and formatives. While the Latin numeral system was synthetic, with a distinct ending for each type of numeral, the Romance numerals often feature more than one (unevenly distributed) marker or structure per series, which feature varying degrees of inherited, borrowed, or innovative elements. Formal consistency is strongest in cardinals, followed by ordinals and then the other types of numeral, which also tend to be more analytic or periphrastic. From a morphological perspective, Romance numerals overall have moved away from the inherited syntheticity, but several series continue to be synthetic formations—at least in part—with morphological markers drawn from Latin that may have undergone functional change (e.g. distributive > ordinal > collective). The underlying syntax of Romance numerals is in line with the overall grammatical patterns of Romance languages, as reflected in the prevalence of word order (with arithmetical correlates), connectors, (partial) loss of agreement, and analyticity. Innovation is prominent in the formation of higher numerals with bases beyond ‘thousand’, of teens and decads in Romanian, and of vigesimals in numerous Romance varieties.

Abstract

The event-related potential (ERP) component named phonological mismatch negativity (PMN) arises when listeners hear an unexpected word form in a spoken sentence [1]. The PMN is thought to reflect the mismatch between expected and perceived auditory speech input. In this paper, we use the PMN to test a central premise in the predictive coding framework [2], namely that the mismatch between prior expectations and sensory input is an important mechanism of perception. We test this with natural speech materials containing approximately 50,000 word tokens. The corresponding EEG-signal was recorded while participants (n = 48) listened to these materials. Following [3], we quantify the mismatch with two word probability distributions (WPD): a WPD based on preceding context, and a WPD that is additionally updated based on the incoming audio of the current word. We use the between-WPD cross entropy for each word in the utterances and show that a higher cross entropy correlates with a more negative PMN. Our results show that listeners anticipate auditory input while processing each word in naturalistic speech. Moreover, complementing previous research, we show that predictive language processing occurs across the whole probability spectrum.

Abstract

The mismatch between top-down predicted and bottom-up perceptual input is an important mechanism of perception according to the predictive coding framework (Friston, [1]). In this paper we develop and validate a new information-theoretic measure that quantifies the mismatch between expected and observed auditory input during speech processing. We argue that such a mismatch measure is useful for the study of speech processing. To compute the mismatch measure, we use naturalistic speech materials containing approximately 50,000 word tokens. For each word token we first estimate the prior word probability distribution with the aid of statistical language modelling, and next use automatic speech recognition to update this word probability distribution based on the unfolding speech signal. We validate the mismatch measure with multiple analyses, and show that the auditory-based update improves the probability of the correct word and lowers the uncertainty of the word probability distribution. Based on these results, we argue that it is possible to explicitly estimate the mismatch between predicted and perceived speech input with the cross entropy between word expectations computed before and after an auditory update.

Abstract

Studies have shown that cross-linguistically, consonants at the left edge of higher-level prosodic boundaries tend to be more forcefully articulated than those at lower-level boundaries, a phenomenon known as domain-initial strengthening. This study tests whether similar effects occur in Turkish, using the Autosegmental-Metrical model proposed by Ipek & Jun [1, 2] as the basis for assessing boundary strength. Productions of /t/ and /d/ were elicited in four domain-initial prosodic positions corresponding to progressively higher-level boundaries: syllable, word, intermediate phrase, and Intonational Phrase. A fifth position, nuclear word, was included in order to better situate it within the prosodic hierarchy. Acoustic correlates of articulatory strength were measured, including closure duration for /d/ and /t/, as well as voice onset time and burst energy for /t/. Our results show that closure duration increases cumulatively from syllable to intermediate phrase, while voice onset time and burst energy are not influenced by boundary strength. These findings provide corroborating evidence for Ipek & Jun’s model, particularly for the distinction between word and intermediate phrase boundaries. Additionally, articulatory strength at the left edge of the nuclear word patterned closely with word-initial position, supporting the view that the nuclear word is not associated with a distinct phrasing domain

Abstract

Speech contains pronounced amplitude modulations in the 1–9 Hz range, correlating with the syllabic rate of speech. Recent models of speech perception propose that this rhythmic nature of speech is central to speech recognition and has beneficial effects on language processing. Here, we investigated the contribution of amplitude modulations to the subjective impression listeners have of public speakers. The speech from US presidential candidates Hillary Clinton and Donald Trump in the three TV debates of 2016 was acoustically analyzed by means of modulation spectra. These indicated that Clinton’s speech had more pronounced amplitude modulations than Trump’s speech, particularly in the 1–9 Hz range. A subsequent perception experiment, with listeners rating the perceived charisma of (low-pass filtered versions of) Clinton’s and Trump’s speech, showed that more pronounced amplitude modulations (i.e., more ‘rhythmic’ speech) increased perceived charisma ratings. These outcomes highlight the important contribution of speech rhythm to charisma perception.

Abstract

Learning a language involves learning how to map specific forms onto their associated meanings. Such mappings can utilise arbitrariness and non-arbitrariness, yet, our understanding of how these two systems operate at different stages of vocabulary development is still not fully understood. The Sound-Symbolism Bootstrapping Hypothesis (SSBH) proposes that sound-symbolism is essential for word learning to commence, but empirical evidence of exactly how sound-symbolism influences language learning is still sparse. It may be the case that sound-symbolism supports acquisition of categories of meaning, or that it enables acquisition of individualized word meanings. In two Experiments where participants learned form-meaning mappings from either sound-symbolic or arbitrary languages, we demonstrate the changing roles of sound-symbolism and arbitrariness for different vocabulary sizes, showing that sound-symbolism provides an advantage for learning of broad categories, which may then transfer to support learning individual words, whereas an arbitrary language impedes acquisition of categories of sound to meaning.

Abstract

This chapter focuses on connectionist modeling in language production, highlighting how
core principles of connectionism provide coverage for empirical observations about
representation and selection at the phonological, lexical, and sentence levels. The first
section focuses on the connectionist principles of localist representations and spreading
activation. It discusses how these two principles have motivated classic models of speech
production and shows how they cover results of the picture-word interference paradigm,
the mixed error effect, and aphasic naming errors. The second section focuses on how
newer connectionist models incorporate the principles of learning and distributed
representations through discussion of syntactic priming, cumulative semantic
interference, sequencing errors, phonological blends, and code-switching

Abstract

Speech input causes listeners to activate multiple
candidate words which then compete with one
another. These include onset competitors, that share a
beginning (bumper, butter), but also, counterintuitively,
rhyme competitors, sharing an ending
(bumper, jumper). In L1, competition is typically
stronger for onset than for rhyme. In L2, onset
competition has been attested but rhyme competition
has heretofore remained largely unexamined. We
assessed L1 (Dutch) and L2 (English) word
recognition by the same late-bilingual individuals. In
each language, eye gaze was recorded as listeners
heard sentences and viewed sets of drawings: three
unrelated, one depicting an onset or rhyme competitor
of a word in the input. Activation patterns revealed
substantial onset competition but no significant
rhyme competition in either L1 or L2. Rhyme
competition may thus be a “luxury” feature of
maximally efficient listening, to be abandoned when
resources are scarcer, as in listening by late
bilinguals, in either language.

Abstract

In this study, we compared the performance of a long short-term memory (LSTM) neural network to the behavior of human participants on a language task that requires hierarchically structured knowledge. We show that humans interpret ambiguous noun phrases, such as second blue ball, in line with their hierarchical constituent structure. LSTMs, instead, only do
so after unambiguous training, and they do not systematically generalize to novel items. Overall, the results of our simulations indicate that a model can behave hierarchically without relying on hierarchical constituent structure.

Abstract

Speaker diarization (answering 'who spoke when') is a widely researched subject within speech technology. Numerous experiments have been run on datasets built from broadcast news, meeting data, and call centers—the task sometimes appears close to being solved. Much less work has begun to tackle the hardest diarization task of all: spontaneous conversations in real-world settings. Such diarization would be particularly useful for studies of language acquisition, where researchers investigate the speech children produce and hear in their daily lives. In this paper, we study audio gathered with a recorder worn by small children as they went about their normal days. As a result, each child was exposed to different acoustic environments with a multitude of background noises and a varying number of adults and peers. The inconsistency of speech and noise within and across samples poses a challenging task for speaker diarization systems, which we tackled via retraining and data augmentation techniques. We further studied sources of structured variation across raw audio files, including the impact of speaker type distribution, proportion of speech from children, and child age on diarization performance. We discuss the extent to which these findings might generalize to other samples of speech in the wild.

Abstract

Part of the remarkable efficiency of listening is
accommodation to unfamiliar talkers’ specific
pronunciations by retuning of phonemic intercategory
boundaries. Such retuning occurs in second
(L2) as well as first language (L1); however, recent
research with emigrés revealed successful adaptation
in the environmental L2 but, unprecedentedly, not in
L1 despite continuing L1 use. A possible explanation
involving relative exposure to novel talkers is here
tested in heritage language users with Mandarin as
family L1 and English as environmental language. In
English, exposure to an ambiguous sound in
disambiguating word contexts prompted the expected
adjustment of phonemic boundaries in subsequent
categorisation. However, no adjustment occurred in
Mandarin, again despite regular use. Participants
reported highly asymmetric interlocutor counts in the
two languages. We conclude that successful retuning
ability requires regular exposure to novel talkers in
the language in question, a criterion not met for the
emigrés’ or for these heritage users’ L1.

Abstract

In two experiments, Mandarin listeners resolved potential syntactic ambiguities in spoken utterances in (a) their native language (L1) and (b) English which they had learned as a second language (L2). A new disambiguation task was used, requiring speeded responses to select the correct meaning for structurally ambiguous sentences. Importantly, the ambiguities used in the study are identical in Mandarin and in English, and production data show that prosodic disambiguation of this type of ambiguity is also realised very similarly in the two languages. The perceptual results here showed however that listeners’ response patterns differed for L1 and L2, although there was a significant increase in similarity between the two response patterns with increasing exposure to the L2. Thus identical ambiguity and comparable disambiguation patterns in L1 and L2 do not lead to immediate application of the appropriate L1 listening strategy to L2; instead, it appears that such a strategy may have to be learned anew for the L2.

Abstract

Using a phoneme detection task, the present series of
experiments examines whether listeners can entrain to
different combinations of prosodic cues to predict where focus
will fall in an utterance. The stimuli were recorded by four
female native speakers of Australian English who happened to
have used different prosodic cues to produce sentences with
prosodic focus: a combination of duration cues, mean and
maximum F0, F0 range, and longer pre-target interval before
the focused word onset, only mean F0 cues, only pre-target
interval, and only duration cues. Results revealed that listeners
can entrain in almost every condition except for where
duration was the only reliable cue. Our findings suggest that
listeners are flexible in the cues they use for focus processing.

Abstract

Listeners adapt rapidly to previously unheard talkers by
adjusting phoneme categories using lexical knowledge, in a
process termed lexically-guided perceptual learning. Although
this is firmly established for listening in the native language
(L1), perceptual flexibility in second languages (L2) is as yet
less well understood. We report two experiments examining L1
and L2 perceptual learning, the first in Mandarin-English late
bilinguals, the second in Australian learners of Mandarin. Both
studies showed stronger learning in L1; in L2, however,
learning appeared for the English-L1 group but not for the
Mandarin-L1 group. Phonological mapping differences from
the L1 to the L2 are suggested as the reason for this result.

Abstract

Listeners' recognition of spoken language involves complex decoding processes: The continuous speech stream must be segmented into its component words, and words must be recognized despite great variability in their pronunciation (due to talker differences, or to influence of phonetic context, or to speech register) and despite competition from many spuriously present forms supported by the speech signal. L1 listeners deal more readily with all levels of this complexity than L2 listeners. Fortunately, the decoding processes necessary for competent L2 listening can be taught in the classroom. Evidence-based methodologies targeted at the development of efficient speech decoding include teaching of minimal pairs, of phonotactic constraints, and of reduction processes, as well as the use of dictation and L2 video captions.

Abstract

This chapter discusses the genetic foundations of the human capacity for language. It reviews the molecular structure of the genome and the complex molecular mechanisms that allow genetic information to influence multiple levels of biology. It goes on to describe the active regulation of genes and their formation of complex genetic pathways that in turn control the cellular environment and function. At each of these levels, examples of genes and genetic variants that may influence the human capacity for language are given. Finally, it discusses the value of using animal models to understand the genetic underpinnings of speech and language. From this chapter will emerge the complexity of the genome in action and the multidisciplinary efforts that are currently made to bridge the gap between genetics and language.

Abstract

Do interlocutors adjust their conversational strategies to the specific contextual demands of a given situation? Prior studies have yielded conflicting results, making it unclear how strategies vary with demands. We combine insights from qualitative and quantitative approaches in a within-participant experimental design involving two different contexts: spontaneously occurring conversations (SOC) and task-oriented conversations (TOC). We systematically assess backchanneling, other-repair and linguistic alignment. We find that SOC exhibit a higher number of backchannels, a reduced and more generic repair format and higher rates of lexical and syntactic alignment. TOC are characterized by a high number of specific repairs and a lower rate of lexical and syntactic alignment. However, when alignment occurs, more linguistic forms are aligned. The findings show that conversational strategies adapt to specific contextual demands.

Abstract

This chapter makes the case for ‘ideophone’ as a comparative concept: a notion that captures a recurrent typological pattern and provides a template for understanding language-specific phenomena that prove similar. It revises an earlier definition to account for the observation that ideophones typically form an open lexical class, and uses insights from canonical typology to explore the larger typological space. According to the resulting definition, a canonical ideophone is a member of an open lexical class of marked words that depict sensory imagery. The five elements of this definition can be seen as dimensions that together generate a possibility space to characterise cross-linguistic diversity in depictive means of expression. This approach allows for the systematic comparative treatment of ideophones and ideophone-like phenomena. Some phenomena in the larger typological space are discussed to demonstrate the utility of the approach: phonaesthemes in European languages, specialised semantic classes in West-Chadic, diachronic diversions in Aslian, and depicting constructions in signed languages.

Abstract

In conversation, people regularly deal with problems of speaking, hearing, and understanding. We report on a cross-linguistic investigation of the conversational structure of other-initiated repair (also known as collaborative repair, feedback, requests for clarification, or grounding sequences). We take stock of formats for initiating repair across languages (comparable to English huh?, who?, y’mean X?, etc.) and find that different languages make available a wide but remarkably similar range of linguistic resources for this function. We exploit the patterned variation as evidence for several underlying concerns addressed by repair initiation: characterising trouble, managing responsibility, and handling knowledge. The concerns do not always point in the same direction and thus provide participants in interaction with alternative principles for selecting one format over possible others. By comparing conversational structures across languages, this paper contributes to pragmatic typology: the typology of systems of language use and the principles that shape them.

Abstract

Biologically inspired spiking networks are an important tool to study the nature of computation and cognition in neural systems. In this work, we investigate the representational capacity of spiking networks engaged in an identity mapping task. We compare two schemes for encoding symbolic input, one in which input is injected as a direct current and one where input is delivered as a spatio-temporal spike pattern. We test the ability of networks to discriminate their input as a function of the number of distinct input symbols. We also compare performance using either membrane potentials or filtered spike trains as state variable. Furthermore, we investigate how the circuit behavior depends on the balance between excitation and inhibition, and the degree of synchrony and regularity in its internal dynamics. Finally, we compare different linear methods of decoding population activity onto desired target labels. Overall, our results suggest that even this simple mapping task is strongly influenced by design choices on input encoding, state-variables, circuit characteristics and decoding methods, and these factors can interact in complex ways. This work highlights the importance of constraining computational network models of behavior by available neurobiological evidence.

Abstract

Previous studies have shown that speakers align their speech to each other at multiple linguistic levels. This study investigates whether alignment is mostly the result of priming from the immediately preceding
speech materials, focussing on pitch and articulation rate (AR). Native Dutch speakers completed sentences, first by themselves (pre-test), then in alternation with Confederate 1 (Round 1), with Confederate 2 (Round 2), with Confederate 1 again
(Round 3), and lastly by themselves again (post-test). Results indicate that participants aligned to the confederates and that this alignment lasted during the post-test. The confederates’ directly preceding sentences were not good predictors for the participants’ pitch and AR. Overall, the results indicate that alignment is more of a global effect than a local priming effect.

Abstract

This chapter reviews the computational processes that are responsible for recognizing word forms in the speech stream. We outline the different stages in a processing hierarchy from the extraction of general acoustic features, through speech‐specific prelexical processes, to the retrieval and selection of lexical representations. We argue that two recurring properties of the system as a whole are abstraction and adaptability. We also present evidence for parallel processing of information on different timescales, more specifically that segmental material in the speech stream (its consonants and vowels) is processed in parallel with suprasegmental material (the prosodic structures of spoken words). We consider evidence from both psycholinguistics and neurobiology wherever possible, and discuss how the two fields are beginning to address common computational problems. The challenge for future research in speech perception will be to build an account that links these computational problems, through functional mechanisms that address them, to neurobiological implementation.

Abstract

Over the decades, high-intensity language learners scattered over the globe referred to as “hyperpolyglots” have undertaken a natural experiment into the limits of learning and acquiring proficiencies in multiple languages. This chapter details several ways in which hyperpolyglots are relevant to research on aptitude. First, historical hyperpolyglots Cardinal Giuseppe Mezzofanti, Emil Krebs, Elihu Burritt, and Lomb Kató are described in terms of how they viewed their own exceptional outcomes. Next, I draw on results from an online survey with 390 individuals to explore how contemporary hyperpolyglots consider the explanatory value of aptitude. Third, the challenges involved in studying the genetic basis of hyperpolyglottism (and by extension of language aptitude) are discussed. This mosaic of data is meant to inform the direction of future aptitude research that takes hyperpolyglots, one type of exceptional language learner and user, into account.

Abstract

Anaphora, a ubiquitous feature of natural language, poses a particular challenge to young children as they first learn language due to its referential ambiguity. In spite of this, parents and caregivers use anaphora frequently in child-directed speech, potentially presenting a risk to effective communication if children do not yet have the linguistic capabilities of resolving anaphora successfully. Through an eye-tracking study in a naturalistic free-play context, we examine the strategies that parents employ to calibrate their use of anaphora to their child's linguistic development level. We show that, in this way, parents are able to intuitively scaffold the complexity of their speech such that greater referential ambiguity does not hurt overall communication success.

Abstract

This paper shows that the dictation task, a well-
known testing instrument in language education, has
untapped potential as a research tool for studying
speech perception. We describe how transcriptions
can be scored on measures of lexical, orthographic,
phonological, and semantic similarity to target
phrases to provide comprehensive information about
accuracy at different processing levels. The former
three measures are automatically extractable,
increasing objectivity, and the middle two are
gradient, providing finer-grained information than
traditionally used. We evaluate the measures in an
English dictation task featuring phonetically reduced
continuous speech. Whereas the lexical and
orthographic measures emphasize listeners’ word
identification difficulties, the phonological measure
demonstrates that listeners can often still recover
phonological features, and the semantic measure
captures their ability to get the gist of the utterances.
Correlational analyses and a discussion of practical
and theoretical considerations show that combining
multiple measures improves the dictation task’s
utility as a research tool.

Abstract

Lexically guided perceptual learning has traditionally been studied with ambiguous consonant sounds to which native listeners are exposed in a purely receptive listening context. To extend previous research, we investigate whether lexically guided learning applies to a vowel shift encountered by non-native listeners in an interactive dialogue. Dutch participants played a two-player game in English in either a control condition, which contained no evidence for a vowel shift, or a lexically constraining condition, in which onscreen lexical information required them to re-interpret their interlocutor’s /ɪ/ pronunciations as representing /ε/. A phonetic categorization pre-test and post-test were used to assess whether the game shifted listeners’ phonemic boundaries such that more of the /ε/-/ɪ/ continuum came to be perceived as /ε/. Both listener groups showed an overall post-test shift toward /ɪ/, suggesting that vowel perception may be sensitive to directional biases related to properties of the speaker’s vowel space. Importantly, listeners in the lexically constraining condition made relatively more post-test /ε/ responses than the control group, thereby exhibiting an effect of lexically guided adaptation. The results thus demonstrate that non-native listeners can adjust their phonemic boundaries on the basis of lexical information to accommodate a vowel shift learned in interactive conversation.

Abstract

The Cha’palaa language of Ecuador (Barbacoan) features verbal morphology for marking knowledge-based categories that, in usage, show a variant of the cross-linguistically recurrent pattern of ‘egophoric distribution': specific forms associate with speakers in contrast to others in statements and with addressees in contrast to others in questions. These are not person markers, but rather are used by speakers to portray their involvement in states of affairs as active, agentive participants (ego) versus other types of involvement (non-ego). They interact with person and argument structure, but through pragmatic ‘person sensitivities’ rather than through grammatical agreement. Not only does this pattern appear in verbal morphology, it also can be observed in alternations of predicate construction types and case alignment, helping to show how egophoric marking is a pervasive element of Cha'palaa's linguistic system. This chapter gives a first account of egophoricity in Cha’palaa, beginning with a discussion of person sensitivity, egophoric distribution, and issues of flexibility of marking with respect to degree of volition or control. It then focuses on a set of intransitive experiencer (or ‘endopathic') predicates that refer to internal states which mark egophoric values for the undergoer role, not the actor role, showing ‘quirky’ accusative marking instead of nominative case. It concludes with a summary of how egophoricity in Cha'palaa interacts with issues of argument structure in comparison to a language with person agreement, here represented by examples from Cha’palaa’s neighbor Ecuadorian Highland Quechua.

Abstract

Structural imaging based on computerized tomography (CT) and magnetic resonance imaging (MRI) has progressively replaced traditional post‐mortem studies in the process of identifying the neuroanatomical basis of language. In the clinical setting, the information provided by structural imaging has been used to confirm the exact diagnosis and formulate an individualized treatment plan. In the research arena, neuroimaging has permitted to understand neuroanatomy at the individual and group level. The possibility to obtain quantitative measures of lesions has improved correlation analyses between severity of symptoms, lesion load, and lesion location. More recently, the development of structural imaging based on diffusion MRI has provided valid solutions to two major limitations of more conventional imaging. In stroke patients, diffusion can visualize early changes due to a stroke that are otherwise not detectable with more conventional structural imaging, with important implications for the clinical management of acute stroke patients. Beyond the sensitivity to early changes, diffusion imaging tractography presents the possibility of visualizing the trajectories of individual white matter pathways connecting distant regions. A pathway analysis based on tractography is offering a new perspective in neurolinguistics. First, it permits to formulate new anatomical models of language function in the healthy brain and allows to directly test these models in the human population without any reliance on animal models. Second, by defining the exact location of the damage to specific white matter connections we can understand the contribution of different mechanisms to the emergence of language deficits (e.g., cortical versus disconnection mechanisms). Finally, a better understanding of the anatomical variability of different language networks is helping to identify new anatomical predictors of language recovery. In this chapter we will focus on the principles of structural MRI and, in particular, diffusion imaging and tractography and present examples of how these methods have informed our understanding of variance in language performances in the healthy brain and language deficits in patient populations.

Abstract

Achieving linguistic proficiency requires identifying words from speech, and discovering the constraints that govern the way those words are used. In a recent study of non-adjacent dependency learning, Frost and Monaghan (2016) demonstrated that learners may perform these tasks together, using similar statistical processes - contrary to prior suggestions. However, in their study, non-adjacent dependencies were marked by phonological cues (plosive-continuant-plosive structure), which may have influenced learning. Here, we test the necessity of these cues by comparing learning across three conditions; fixed phonology, which contains these cues, varied phonology, which omits them, and shapes, which uses visual shape sequences to assess the generality of statistical processing for these tasks. Participants segmented the sequences and generalized the structure in both auditory conditions, but learning was best when phonological cues were present. Learning was around chance on both tasks for the visual shapes group, indicating statistical processing may critically differ across domains.

Abstract

Language acquisition is complex. However, one thing that has been suggested to help learning is the way that information is distributed throughout language; co-occurrences among particular items (e.g., syllables and words) have been shown to help learners discover the words that a language contains and figure out how those words are used. Humans’ ability to draw on this information—“statistical learning”—has been demonstrated across a broad range of studies. However, evidence from non-WEIRD (Western, Educated, Industrialized, Rich, and Democratic) societies is critically lacking, which limits theorizing on the universality of this skill. We extended work on statistical language learning to a new, non-WEIRD linguistic population: speakers of Yélî Dnye, who live on a remote island off mainland Papua New Guinea (Rossel Island). We performed a replication of an existing statistical learning study, training adults on an artificial language with statistically defined words, then examining what they had learnt using a two-alternative forced-choice test. Crucially, we implemented several key amendments to the original study to ensure the replication was suitable for remote field-site testing with speakers of Yélî Dnye. We made critical changes to the stimuli and materials (to test speakers of Yélî Dnye, rather than English), the instructions (we re-worked these significantly, and added practice tasks to optimize participants’ understanding), and the study format (shifting from a lab-based to a portable tablet-based setup). We discuss the requirement for acute sensitivity to linguistic, cultural, and environmental factors when adapting studies to test new populations.

Abstract

Graph neural networks (GNNs) have emerged as the standard method for numerous tasks on graph-structured data such as node classification. However, real-world graphs are often evolving over time and even new classes may arise. We model these challenges as an instance of lifelong learning, in which a learner faces a sequence of tasks and may take over knowledge acquired in past tasks. Such knowledge may be stored explicitly as historic data or implicitly within model parameters. In this work, we systematically analyze the influence of implicit and explicit knowledge. Therefore, we present an incremental training method for lifelong learning on graphs and introduce a new measure based on k-neighborhood time differences to address variances in the historic data. We apply our training method to five representative GNN architectures and evaluate them on three new lifelong node classification datasets. Our results show that no more than 50% of the GNN's receptive field is necessary to retain at least 95% accuracy compared to training over the complete history of the graph data. Furthermore, our experiments confirm that implicit knowledge becomes more important when fewer explicit knowledge is available.

Abstract

We analyze the problem of response suggestion in a closed domain along a real-world scenario of a digital library. We present a text-processing pipeline to generate question-answer pairs from chat transcripts. On this limited amount of training data, we compare retrieval-based, conditioned-generation, and dedicated representation learning approaches for response suggestion. Our results show that retrieval-based methods that strive to find similar, known contexts are preferable over parametric approaches from the conditioned-generation family, when the training data is limited. We, however, identify a specific representation learning approach that is competitive to the retrieval-based approaches despite the training data limitation.

Abstract

Large-scale graph data in real-world applications is often not static but dynamic,
i. e., new nodes and edges appear over time. Current graph convolution approaches
are promising, especially, when all the graph’s nodes and edges are available dur-
ing training. When unseen nodes and edges are inserted after training, it is not
yet evaluated whether up-training or re-training from scratch is preferable. We
construct an experimental setup, in which we insert previously unseen nodes and
edges after training and conduct a limited amount of inference epochs. In this
setup, we compare adapting pretrained graph neural networks against retraining
from scratch. Our results show that pretrained models yield high accuracy scores
on the unseen nodes and that pretraining is preferable over retraining from scratch.
Our experiments represent a first step to evaluate and develop truly online variants
of graph neural networks.

Abstract

COVID-19 research datasets are crucial for analyzing research dynamics. Most collections of COVID-19 research items do not to include cited works and do not have annotations
from a controlled vocabulary. Starting with ZB MED KE data on COVID-19, which comprises CORD-19, we assemble a new dataset that includes cited work and MeSH annotations for all records. Furthermore, we conduct experiments on the analysis of research dynamics, in which we investigate predicting links in a co-annotation graph created on the basis of the new dataset. Surprisingly, we find that simple heuristic methods are better at
predicting future links than more sophisticated approaches such as graph neural networks.

Abstract

We present multi-modal adversarial autoencoders for recommendation and evaluate them on two different tasks: citation recommendation and subject label recommendation. We analyze the effects of adversarial regularization, sparsity, and different input modalities. By conducting 408 experiments, we show that adversarial regularization consistently improves the performance of autoencoders for recommendation. We demonstrate, however, that the two tasks differ in the semantics of item co-occurrence in the sense that item co-occurrence resembles relatedness in case of citations, yet implies diversity in case of subject labels. Our results reveal that supplying the partial item set as input is only helpful, when item co-occurrence resembles relatedness. When facing a new recommendation task it is therefore crucial to consider the semantics of item co-occurrence for the choice of an appropriate model.

Abstract

Child songs are a great source for linguistic learning. Here we explore whether children can acquire novel words in a second language by playing a game featuring child songs in a playhouse. We present data from three studies that serve as scientific proof for the functionality of one game of the playhouse: the Energy Center. For this game, three hand-bikes were mounted on a panel. When children start moving the hand-bikes, child songs start playing simultaneously. Once the children produce enough energy with the hand-bikes, the songs are additionally accompanied with the sounds of musical instruments. In our studies, children executed a picture-selection task to evaluate whether they acquired new vocabulary from the songs presented during the game. Two of our studies were run in the field, one at a Dutch and one at an Indian pre-school. The third study features data from a more controlled laboratory setting. Our results partly confirm that the Energy Center is a successful means to support vocabulary acquisition in a second language. More research with larger sample sizes and longer access to the Energy Center is needed to evaluate the overall functionality of the game. Based on informal observations at our test sites, however, we are certain that children do pick up linguistic content from the songs during play, as many of the children repeat words and phrases from songs they heard. We will pick up upon these promising observations during future studies

Abstract

This chapter aims to provide an updated list of all Bantu languages known at present and to provide individual pointers to further information on the inventory. The area division has some correlation with what are perceived genealogical relations between Bantu languages, but they are not defined as such and do not change whenever there is an update in our understanding of genealogical relations. Given the popularity of Guthrie codes in Bantu linguistics, our listing also features a complete mapping to Guthrie codes. The language inventory listed excludes sign languages used in the Bantu area, speech registers, pidgins, drummed/whistled languages and urban youth languages. Pointers to such languages in the Bantu area are included in the continent-wide overview in Hammarstrom. The most important alternative names, subvarieties and spelling variants are given for each language, though such lists are necessarily incomplete and reflect some degree of arbitrary selection.

Abstract

Children with developmental language disorder (DLD) regularly use the base form of verbs (e.g., dance) instead of inflected forms (e.g., danced). We propose an account of this behavior in which children with DLD have difficulty processing novel inflected verbs in their input. This leads the inflected form to face stronger competition from alternatives. Competition is resolved by the production of a more accessible alternative with high semantic overlap with the inflected form: in English, the bare form. We test our account computationally by training a nonparametric Bayesian model that infers the productivity of the inflectional suffix (-ed). We systematically vary the number of novel types of inflected verbs in the input to simulate the input as processed by children with and without DLD. Modeling results are consistent with our hypothesis, suggesting that children’s inconsistent use of inflectional morphemes could stem from inferences they make on the basis of impoverished data.

Abstract

Prediction in language has traditionally been studied using
simple designs in which neural responses to expected
and unexpected words are compared in a categorical
fashion. However, these designs have been contested
as being ‘prediction encouraging’, potentially exaggerating
the importance of prediction in language understanding.
A few recent studies have begun to address
these worries by using model-based approaches to probe
the effects of linguistic predictability in naturalistic stimuli
(e.g. continuous narrative). However, these studies
so far only looked at very local forms of prediction, using
models that take no more than the prior two words into
account when computing a word’s predictability. Here,
we extend this approach using a state-of-the-art neural
language model that can take roughly 500 times longer
linguistic contexts into account. Predictability estimates
fromthe neural network offer amuch better fit to EEG data
from subjects listening to naturalistic narrative than simpler
models, and reveal strong surprise responses akin to
the P200 and N400. These results show that predictability
effects in language are not a side-effect of simple designs,
and demonstrate the practical use of recent advances
in AI for the cognitive neuroscience of language.

Abstract

This paper reports on an on-going project designed to collect comparable corpus data on child language and child-directed language in under-researched languages. Despite a long history of cross-linguistic research, there is a severe empirical bias within language acquisition research: Data is available for less than 2% of the world's languages, heavily skewed towards the larger and better-described languages. As a result, theories of language development tend to be grounded in a non-representative sample, and we know little about the acquisition of typologically-diverse languages from different families, regions, or sociocultural contexts. It is very likely that the reasons are to be found in the forbidding methodological challenges of constructing child language corpora under fieldwork conditions with their strict requirements on participant selection, sampling intervals, and amounts of data. There is thus an urgent need for proposals that facilitate and encourage language acquisition research across a wide variety of languages. Adopting a language documentation perspective, we illustrate an approach that combines the construction of manageable corpora of natural interaction with and between children with a sketch description of the corpus data – resulting in a set of comparable corpora and comparable sketches that form the basis for cross-linguistic comparisons.

Abstract

Using the visual-word paradigm, the present study investigated the effects of word onset and offset masking on the time course of non-native spoken-word recognition in the presence of background noise. In two experiments, Dutch non-native listeners heard English target words, preceded by carrier sentences that were noise-free (Experiment 1) or contained intermittent noise (Experiment 2). Target words were either onset- or offset-masked or not masked at all. Results showed that onset masking delayed target word recognition more than offset masking did, suggesting that – similar to natives – non-native listeners strongly rely on word onset information during word recognition in noise.

Abstract

Conversation Analysis (CA) is an inductive, micro-analytic, and predominantly qualitative
method for studying human social interactions. This chapter describes and illustrates the basic
methods of CA. We first situate the method by describing its sociological foundations, key areas
of analysis, and particular approach in using naturally occurring data. The bulk of the chapter is
devoted to practical explanations of the typical conversation analytic process for collecting data
and producing an analysis. We analyze a candidate interactional practice – the assessmentimplicative
interrogative – using real data extracts as a demonstration of the method, explicitly
laying out the relevant questions and considerations for every stage of an analysis. The chapter
concludes with some discussion of quantitative approaches to conversational interaction, and
links between CA and psycholinguistic concerns

Abstract

What makes some words harder to learn than others in a second language? Although some robust factors have been identified based on small scale experimental studies, many relevant factors are difficult to study in such experiments due to the amount of data necessary to test them. Here, we investigate what factors affect the ease of learning of a word in a second language using a large data set of users learning English as a second language through the Duolingo mobile app. In a regression analysis, we test and confirm the well-studied effect of cognate status on word learning accuracy. Furthermore, we find significant effects for both cross-linguistic semantic alignment and English semantic density, two novel predictors derived from large scale distributional models of lexical semantics. Finally, we provide data on several other psycholinguistically plausible word level predictors. We conclude with a discussion of the limits, benefits and future research potential of using big data for investigating second language learning.

Abstract

This chapter deals with the question of whether there is one syntactic system that is shared by language production and comprehension or whether there are two separate systems. It first discusses arguments in favor of one or the other option and then presents the current evidence on the brain structures involved in sentence processing. The results of meta-analyses of numerous neuroimaging studies suggest that there is one system consisting of functionally distinct cortical regions: the dorsal part of Broca’s area subserving compositional syntactic processing; the ventral part of Broca’s area subserving compositional semantic processing; and the left posterior temporal cortex (Wernicke’s area) subserving the retrieval of lexical syntactic and semantic information. Sentence production, the comprehension of simple and complex sentences, and the parsing of sentences containing grammatical violations differ with respect to the recruitment of these functional components.

Abstract

A common assumption in the cognitive sciences is that artificial and natural language learning rely on shared mechanisms. However, attempts to bridge the two have yielded ambiguous results. We suggest that an empirical disconnect between the computations employed during learning and the methods employed at test may explain these mixed results. Further, we propose statistically-based chunking as a potential computational link between artificial and natural language learning. We compare the acquisition of non-adjacent dependencies to that of natural language structure using two types of tasks: reflection-based 2AFC measures, and processing-based recall measures, the latter being more computationally analogous to the processes used during language acquisition. Our results demonstrate that task-type significantly influences the correlations observed between artificial and natural language acquisition, with reflection-based and processing-based measures correlating within – but not across – task-type. These findings have fundamental implications for artificial-to-natural language comparisons, both methodologically and theoretically.

Abstract

Computer models of cultural evolution have shown language properties emerging on interacting agents with a brain that lacks dedicated, nativist language modules. Notably, models using Bayesian agents provide a precise specification of (extra-)liguististic factors (e.g., genetic) that shape language through iterated learning (biases on language), and demonstrate that weak biases get expressed more strongly over time (bias amplification). Other models attempt to lessen assumption on agents’ innate predispositions even more, and emphasize self-organization within agents, highlighting glossogenesis (the development of language from a nonlinguistic state). Ultimately however, one also has to recognize that biology and culture are strongly interacting, forming a coevolving system. As such, computer models show that agents might (biologically) evolve to a state predisposed to language adaptability, where (culturally) stable language features might get assimilated into the genome via Baldwinian niche construction. In summary, while many questions about language evolution remain unanswered, it is clear that it is not to be completely understood from a purely biological, cognitivist perspective. Language should be regarded as (partially) emerging on the social interactions between large populations of speakers. In this context, agent models provide a sound approach to investigate the complex dynamics of genetic biasing on language and speech

Abstract

This study investigates effects of prosodic factors (prominence, boundary) on coarticulatory Vnasalization in Australian English (AusE) in CVN and NVC in comparison to those in American English
(AmE). As in AmE, prominence was found to
lengthen N, but to reduce V-nasalization, enhancing N’s nasality and V’s orality, respectively (paradigmatic contrast enhancement). But the prominence effect in CVN was more robust than that in AmE. Again similar to findings in AmE, boundary
induced a reduction of N-duration and V-nasalization phrase-initially (syntagmatic contrast enhancement), and increased the nasality of both C and V phrasefinally.
But AusE showed some differences in terms
of the magnitude of V nasalization and N duration. The results suggest that the linguistic contrast enhancements underlie prosodic-structure modulation of coarticulatory V-nasalization in
comparable ways across dialects, while the fine phonetic detail indicates that the phonetics-prosody interplay is internalized in the individual dialect’s phonetic grammar.

Abstract

It has been suggested that some individuals may benefit more from social robots than do others. Using second
language (L2) as an example, the present study examined how individual differences in attitudes toward robots and personality
traits may be related to learning outcomes. Preliminary results with 24 Turkish-speaking adults suggest that negative attitudes
toward robots, more specifically thoughts and anxiety about the negative social impact that robots may have on the society,
predicted how well adults learned L2 words from a social robot. The possible implications of the findings as well as future directions are also discussed

Abstract

A wealth of studies has shown that more proficient monolingual speakers are better at predicting upcoming information during language comprehension. Similarly, prediction skills of adult second language (L2) speakers in their L2 have also been argued to be modulated by their L2 proficiency. How exactly language proficiency and prediction are linked, however, is yet to be systematically investigated. One group of language users which has the potential to provide invaluable insights into this link is bilingual children. In this paper, we compare bilingual children’s prediction skills with those of monolingual children and adult L2 speakers, and show how investigating bilingual children’s prediction skills may contribute to our understanding of how predictive processing works.

Abstract

There is a strong relation between children’s exposure to
spatial terms and their later memory accuracy. In the current
study, we tested whether the production of spatial terms by
children themselves predicts memory accuracy and whether
and how language modality of these encodings modulates
memory accuracy differently. Hearing child speakers of
Turkish and deaf child signers of Turkish Sign Language
described pictures of objects in various spatial relations to each
other and later tested for their memory accuracy of these
pictures in a surprise memory task. We found that having
described the spatial relation between the objects predicted
better memory accuracy. However, the modality of these
descriptions in sign, speech, or speech-plus-gesture did not
reveal differences in memory accuracy. We discuss the
implications of these findings for the relation between spatial
language, memory, and the modality of encoding.

Abstract

Like Einstein's general theory of relativity is concerned with explaining the basics of an observable experience – i.e., gravity – most people take for granted that Chomsky's theory of generative grammar (GG) is concerned with the basic nature of language. This chapter highlights a mere subset of central constructs in GG, showing how they have featured prominently and thus shaped formal linguistic studies in multilingualism. Because multilingualism includes a wide range of nonmonolingual populations, the constructs are divided across child bilingualism and adult third language for greater coverage. In the case of the former, the chapter examines how poverty of the stimulus has been investigated. Using the nascent field of L3/Ln acquisition as the backdrop, it discusses how the GG constructs of I-language versus E-language sit at the core of debates regarding the very notion of what linguistic transfer and mental representations should be taken to be.

Abstract

Citations play a crucial role in the scientific discourse, in information retrieval, and in bibliometrics. Many initiatives are currently promoting the idea of having free and open citation data. Creation of citation data, however, is not part of the cataloging workflow in libraries nowadays.
In this paper, we present our project Linked Open Citation Database, in which we design distributed processes and a system infrastructure based on linked data technology. The goal is to show that efficiently cataloging citations in libraries using a semi-automatic approach is possible. We specifically describe the current state of the workflow and its implementation. We show that we could significantly improve the automatic reference extraction that is crucial for the subsequent data curation. We further give insights on the curation and linking process and provide evaluation results that not only direct the further development of the project, but also allow us to discuss its overall feasibility.

Abstract

It is widely held that smells and flavors are impossible to put into words. In this paper we test this claim by seeking predictive patterns in wine reviews, which ostensibly aim to provide guides to perceptual content. Wine reviews have previously been critiqued as random and meaningless. We collected an English corpus of wine reviews with their structured metadata, and applied machine learning techniques to automatically predict the wine's color, grape variety, and country of origin. To train the three supervised classifiers, three different information sources were incorporated: lexical bag-of-words features, domain-specific terminology features, and semantic word embedding features. In addition, using regression analysis we investigated basic review properties, i.e., review length, average word length, and their relationship to the scalar values of price and review score. Our results show that wine experts do share a common vocabulary to describe wines and they use this in a consistent way, which makes it possible to automatically predict wine characteristics based on the review text alone. This means that odors and flavors may be more expressible in language than typically acknowledged.

Abstract

Language is a social phenomenon. The author learns, processes, and uses it in social contexts. In other words, the social environment shapes the linguistic knowledge and use of the knowledge. To a degree, this is trivial. A child exposed to Japanese will become fluent in Japanese, whereas a child exposed to only Spanish will not understand Japanese but will master the sounds, vocabulary, and grammar of Spanish. Language is a structured system. Sounds and words do not occur randomly but are characterized by regularities. Learners are sensitive to these regularities and exploit them when learning language. People differ in the sizes of their social networks. Some people tend to interact with only a few people, whereas others might interact with a wide range of people. This is reflected in people’s holiday greeting habits: some people might send cards to only a few people, whereas other would send greeting cards to more than 350 people.

Abstract

Verb learning is important for young children. While most
previous research has focused on linguistic and conceptual
challenges in early verb learning (e.g. Gentner, 1982, 2006),
the present paper examined early verb learning at the
attentional level and quantified the input for early verb learning
by measuring verb-action co-occurrence statistics in parent-
child interaction from the learner’s perspective. To do so, we
used head-mounted eye tracking to record fine-grained
multimodal behaviors during parent-infant joint play, and
analyzed parent speech, parent and infant action, and infant
attention at the moments when parents produced verb labels.
Our results show great variability across different action verbs,
in terms of frequency of verb utterances, frequency of
corresponding actions related to verb meanings, and infants’
attention to verbs and actions, which provide new insights on
why some verbs are harder to learn than others.

Abstract

We present the Narrative Brain Dataset, an fMRI dataset that was collected during spoken presentation of short excerpts of three
stories in Dutch. Together with the brain imaging data, the dataset contains the written versions of the stimulation texts. The texts are
accompanied with stochastic (perplexity and entropy) and semantic computational linguistic measures. The richness and unconstrained
nature of the data allows the study of language processing in the brain in a more naturalistic setting than is common for fMRI studies.
We hope that by making NBD available we serve the double purpose of providing useful neural data to researchers interested in natural
language processing in the brain and to further stimulate data sharing in the field of neuroscience of language.

Abstract

Continuous Bag of Words (CBOW) is a powerful text embedding method. Due to its strong capabilities to encode word content, CBOW embeddings perform well on a wide range of downstream tasks while being efficient to compute. However, CBOW is not capable of capturing the word order. The reason is that the computation of CBOW's word embeddings is commutative, i.e., embeddings of XYZ and ZYX are the same. In order to address this shortcoming, we propose a
learning algorithm for the Continuous Matrix Space Model, which we call Continual Multiplication of Words (CMOW). Our algorithm is an adaptation of word2vec, so that it can be trained on large quantities of unlabeled text. We empirically show that CMOW better captures linguistic properties, but it is inferior to CBOW in memorizing word content. Motivated by these findings, we propose a hybrid model that combines the strengths of CBOW and CMOW. Our results show that the hybrid CBOW-CMOW-model retains CBOW's strong ability to memorize word content while at the same time substantially improving its ability to encode other linguistic information by 8%. As a result, the hybrid also performs better on 8 out of 11 supervised downstream tasks with an average improvement of 1.2%.

Abstract

For (semi-)automated subject indexing systems in digital libraries, it is often more practical to use metadata such as the title of a publication instead of the full-text or the abstract. Therefore, it is desirable to have good text mining and text classification algorithms that operate well already on the title of a publication. So far, the classification performance on titles is not competitive with the performance on the full-texts if the same number of training samples is used for training. However, it is much easier to obtain title data in large quantities and to use it for training than full-text data. In this paper, we investigate the question how models obtained from training on increasing amounts of title training data compare to models from training on a constant number of full-texts. We evaluate this question on a large-scale dataset from the medical domain (PubMed) and from economics (EconBiz). In these datasets, the titles and annotations of millions of publications are available, and they outnumber the available full-texts by a factor of 20 and 15, respectively. To exploit these large amounts of data to their full potential, we develop three strong deep learning classifiers and evaluate their performance on the two datasets. The results are promising. On the EconBiz dataset, all three classifiers outperform their full-text counterparts by a large margin. The best title-based classifier outperforms the best full-text method by 9.4%. On the PubMed dataset, the best title-based method almost reaches the performance of the best full-text classifier, with a difference of only 2.9%.