Canada

Polarity-sensitivity is a typologically general linguistic phenomenon. We focus on negative polarity items (NPIs, e.g. English &#39;any&#39;) -- expressions that are licensed only in negative contexts. The relevant notion of &#39;negative context&#39; could be defined lexically, syntactically or semantically. There is psycholinguistic evidence in favour of semantics as a driving factor for some NPIs in a couple of languages (Chemla, Homer, &amp; Rothschild, 2011; Denić, Homer, Rothschild, &amp; Chemla, 2021). Testing the scale of this analysis as a potential cross-linguistic universal experimentally is extremely hard. We turn to recent multilingual pre-trained language models -- multilingual BERT (Devlin, Chang, Lee, &amp; Toutanova, 2018) and XLM-RoBERTa (Conneau et al., 2019) -- and evaluate the models&#39; recognition of polarity-sensitivity and its cross-lingual generality. Further, using the artificial language learning paradigm, we look for the connection in neural language models between semantic profiles of expressions and their ability to license NPIs. We find evidence for such connection for negation but not for other items we study.

CogSci 2022

The driving forces of polarity-sensitivity: Experiments with multilingual pre-trained neural language models

**Welcome to EMNLP 2023!**

On behalf of the EMNLP 2023 Organizing Committee, I extend a warm and heartfelt welcome to all of you to the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP). It is with immense pleasure and excitement that we gather here in Singapore, a vibrant hub of innovation and technological advancement.

The conference program is packed with insightful presentations, thought-provoking workshops, and engaging networking opportunities. In addition to the technical sessions, we have also planned several social events that will provide you with the opportunity to connect with your colleagues.

The conference is held in person at the Resorts World Convention Centre in Singapore, and available on-line with the help of Underline.

For those in Singapore, I hope that you will find time to explore the exciting Sentosa Island, Gardens by the Bay, Singapore Botanic Gardens and the many other attractions unique to Singapore.

Of course, we are grateful to our sponsors and partners for their generous support of EMNLP 2023, whose contributions make it possible for us to host this world-class event.

Yuji Matsumoto (RIKEN AIP) 
EMNLP 2023 General Chair

To access the **EMNLP 2023** event page on Underline, you need to register for the Conference. 
Please follow **[this link](https://2023.emnlp.org/registration/)** for more details.

EMNLP 2023

EMNLP 2023 took place in Singapore from Dec 6th to Dec 10th, 2023.

Polarity-sensitivity is a typologically general linguistic phenomenon. We focus on negative polarity items (NPIs, e.g. English 'any') -- expressions that are licensed only in negative contexts. The relevant notion of 'negative context' could be defined lexically, syntactically or semantically. There is psycholinguistic evidence in favour of semantics as a driving factor for some NPIs in a couple of languages (Chemla, Homer, & Rothschild, 2011; Denić, Homer, Rothschild, & Chemla, 2021). Testing the scale of this analysis as a potential cross-linguistic universal experimentally is extremely hard. We turn to recent multilingual pre-trained language models -- multilingual BERT (Devlin, Chang, Lee, & Toutanova, 2018) and XLM-RoBERTa (Conneau et al., 2019) -- and evaluate the models' recognition of polarity-sensitivity and its cross-lingual generality. Further, using the artificial language learning paradigm, we look for the connection in neural language models between semantic profiles of expressions and their ability to license NPIs. We find evidence for such connection for negation but not for other items we study.

poster

Join us for CogSci2022! Our 44th Annual Conference will be centered around the theme of diversity, with an accompanying range of keynotes and invited speakers who themselves highlight the diversity of our field. Our three symposia are Cognition across Diverse Populations, Cognition within Diverse Societies, and Global Challenges for Cognitive Science. The conference is being offered in a fully hybrid environment for onsite and online attendees.
 
 [**The Program at a Glance can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2022/07/Cogsci-2022-PAAG-Detailed-July-12final-1.pdf)
 [**The Detailed Program can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2022/08/DeArmond-CogSci2022-program-v113-final.pdf)

This conference requires a registration to attend. Please use the link below to secure your ticket. The website will be available to all registered attendees on July 25th. 

[Register to Attend](https://cognitivesciencesociety.org/registration/)

Registration Is Required

The 44th Annual Meeting of the Cognitive Science Society highlighted research on the theme Cognitive Diversity in addition to the full breadth and diversity of research topics offered by the society’s membership. 





This paper proposes new empirical classifiers for evaluative concepts, including thin concepts like 'good' or 'bad' and thick concepts such as 'honest' or 'disgusting', based on quantitative corpus linguistics. Prior work in experimental philosophy has shown that sentiment analysis can be used to track differences between concept classes. Building on this, Task 1 investigates whether the relationship between sentiment and evaluativeness is parabolic rather than linear. Task 2 extends this question to the differences between evaluative and non-evaluative concept classes. The results of both Tasks show that the linear and the parabolic logistic regression classifiers perform equally well. Interestingly, this study also finds that adjectives attributed to animate entities (e.g. "generous customer") generally have a higher probability to be evaluative concepts than those attributed to inanimate entities (e.g."dry soil").

Why are reckless socks not (more of) a thing? Towards an empirical classification of evaluative concepts.

Models of word meaning, like the Topics model (Griffiths et al., 2007) and word2vec (Mikolov et al., 2013), condense word-by-context co-occurrence statistics to induce representations that organize words along semantically relevant dimensions (e.g., synonymy, antonymy, hyponymy etc.). However, their reliance on latent representations leaves them vulnerable to interference and makes them slow learners. We show how it is possible to construct the meaning of words online during retrieval to avoid these limitations. 
We implement our spreading activation account of word meaning in an associative net, a one-layer highly recurrent network of associations, called a Dynamic-Eigen-Net, that we developed to address the limitations of earlier variants of associative nets when scaling up to deal with unstructured input domains such as natural language text. After fixing the corpus across models, we show that spreading activation using a Dynamic-Eigen-Net outperforms the Topics model and word2vec in several cases when predicting human free associations and word similarity ratings. We argue in favour of the Dynamic-Eigen-Net as a fast learner that is not subject to catastrophic interference, and present it as an example of delegating the induction of latent relationships to process assumptions instead of assumptions about representation.

Constructing Word Meaning without Latent Representations using Spreading Activation

Skilled readers are typically most accurate at identifying words when fixating them slightly left of the central character, the so-called optimal viewing position. There are two main explanations for this effect, which are not mutually exclusive. The first claims that the optimal viewing position lies left-of-center due to the particular constraints of the human perceptual system. The second explains the effect in terms of the beginnings of words generally being more informative about word identity. The complexities of natural languages make it difficult to tease apart the relative contribution of each explanation. We explore this issue through the lens of a Bayesian cognitive model and two experiments using artificial lexicons in which we can carefully control how information is distributed across wordforms. Our results replicate previous findings and further suggest that readers may use implicit knowledge about information distribution to minimize uncertainty when targeting words.

Minimizing Expected Uncertainty in Visual Word Recognition: Are Readers Sensitive to the Distribution of Information across Word Forms?

Is the visual word recognition system sensitive to compound (e.g., SNOWBALL) and pseudocompound (e.g., CARPET) constituents? To answer this question, we employed red-blue anaglyphs glasses to parse red-blue colored (pseudo)compounds at the fovea—by hypothesis split along the vertical meridian—and allowing for the independent processing of word segments in the visual word form area. Seventy-one participants performed a visual masked lexical decision task with 133 millisecond stimuli presentations. Compounds and pseudocompounds were either split at the (pseudo)morpheme boundary (legally-split) or within a (pseudo)morpheme (illegally-split). We found a statistically significant interaction effect between word type and legality. Planned comparisons revealed an advantage in accuracy for legally-split compounds and pseudocompounds over their respective illegal splits. Furthermore, legally-split compounds elicited more accurate responses than legally-split pseudocompounds. In line with previous findings supporting a purely morpho-orthographic prelexical process with faster presentation times, these findings suggest a temporally distinct mechanism during visual word recognition.

Parsing Compounds and Pseudocompounds at the Fovea

Learning to read is a critical skill; yet only a small portion of children in the United States are reading at or above grade level. Attention is one crucial process that affects the
acquisition of reading skills. The process involves selectively choosing task relevant information and requires monitoring competing demands. Many books for beginning readers include illustrations, but this design choice may require learners to split their attention between multiple sources of information. This study employed eye tracking to examine whether embedding text within illustrations in children’s e-books inadvertently induces attentional competition. The results showed that spatially separating illustrations from the text in beginning reader books reduces attentional competition and improves children’s reading comprehension. This study shows that changes to the design of books for beginning readers can help promote literacy development in children.

Low Spatial Proximity Between Text and Illustrations Improves Children’s Comprehension and Attention: An Eye Tracking Study

Readers rely on sentence context to generate predictions about the upcoming words so that processing of their visual forms is less necessary. Consequently, processing of an orthographic neighbor of a strongly predicted word is facilitated by that context (as indicated by a reduced N400 ERP amplitude), regardless of the perceived item’s lexicality (i.e., whether it is a real word or a pseudoword). The current study investigated whether lexicality becomes important when the sentence context is less helpful in generating predictions. Our findings indicate that in weakly constraining sentences, the lexical status of a word impacts word recognition processes as indicated by a left anterior negativity, suggesting that readers rely on sublexical properties of words in the absence of strong expectations.

The effect of orthographic relationships, lexical status and contextual constraint on visual word recognition: Evidence from event-related potentials

It was recently found that letter-shapes have a non-arbitrary relation with their canonical pronunciations, in multiple orthographies and quantified across the whole of each orthography: letters that look similar tend to have similar pronunciations. Similarly, there is phonosemantic systematicity at the word level: words that sound similar tend to have similar meanings. We investigated for the first time whether a similar systematicity exists in Chinese characters. We measured all the pairwise phonological distances and all the pairwise orthographic distances of the 58 Chinese pictographic characters that are taught to year 1 and 2 in Chinese primary schools. The correlation was tested between the two lists of distances and verified by a Mantel test. We found a significant negative correlation between characters and their segmental pronunciations: characters that look similar tend to have dissimilar segmental pronunciations. This contrasts with the positive correlations found in previous similar research with alphabetic writing. We conclude, first, that questions of systematicity in the Chinese writing system are tractable in the same terms and by the same methodology as that applied to alphabetic writing systems. Second, segment-based processing requires to be augmented by tones for there to be systematicity that is comparable to that found in alphabetic writing systems. Any non-arbitrary relation between letter shapes and sounds may help bootstrap the acquisition of literacy.

Grapho-Syllabic Systematicity in Chinese: Chinese Pictographs Have a Non-Arbitrary relation with their Pronunciations

Some sentences are hard to read, and we don’t fully understand why. Memory-based and expectation-based constraints both attempt to explain sentence processing difficulties, and decades of sentence processing literature have found evidence in support of both theories. We further investigate theories of sentence processing by exploring subject- and object-extracted relative clause processing in Standard Arabic. We conducted a self-paced reading task and found that SRCs are easier to process than ORCs in Arabic, in line with expectation-based theories. A follow-up analysis of comprehension question answers revealed that when suggested with the possibility of a noisy interpretation, readers preferentially accept an SRC interpretation over an ORC interpretation. Our future research will explore these findings and test the threshold for acceptance of noisy interpretations.

Expectations and Noisy-Channel Processing of Relative Clauses in Arabic

The English use of "they" to refer to a singular referent that is non-specific or of unknown
gender dates back to the 1300s. Recently, "they" has emerged as the dominant pronoun
for individuals who identify as gender nonbinary, and a coherent subset of English
speakers (innovators) explicitly accept "they" when referring to a specific individual of
known gender ((i) “Sarah slept because they were tired”) in off-line judgments. The
present work examines whether on-line comprehension difficulty patterns with off-line
judgments. Using a maze-task, participants' read sentences like (i) containing “they” with
plural or singular subjects, or “he/she” with named subjects that matched or mismatched
in their typical gender. Gender mismatched evoked reading difficulty. Plural “they” also
caused difficulty, but significantly less. Innovators experienced less difficulty with singular
“they” than non-innovators, but still processed plural “they” faster than singular “they”.
This work suggests that the grammar of “they” may have yet to stabilize even for
innovative speakers.

Processing singular "they" is harder than plural "they", but does not cause referential failure

This visual world eye-tracking pilot study investigates the comprehension of register variants (Stelzen_colloquial vs. Beine_standard transl:.‘’legs’) in a German target sentence when this target sentence (mis)matches the formality of a preceding context sentence, given the object argument either matches or mismatches verb meaning constraints (e.g. Ich rasiere bald meine Beine/Stelzen/#Autos/#Karren, transt.: ‘I shave my legs_standard/legs_coll/#cars_standard/#cars_coll’). The aim of this study is to examine whether register congruence rapidly interacts with verb-argument semantic relations. LME results (n=9) show a main effect of verb-argument congruence but no main effect of formality-register congruence at the region between the verb onset and object-argument onset, indicating that verb-argument relations are computed and used rapidly in online language comprehension. These pilot results suggest that situation formality may indeed modulate verb-argument congruency processing, possibly indicating that standard language processing mechanisms interact closely with register representations.

Premium content

Downloads

Next from CogSci 2022

Why are reckless socks not (more of) a thing? Towards an empirical classification of evaluative concepts.

Similar lecture

NPIs Aren't Exactly Easy: Variation in Licensing across Large Language Models