United States

The relationship between the language a speaker conveys and the meaning a listener recovers is often probabilistic and sensitive to context. Numerous strategies attempt to estimate such a mapping, often leveraging recursive Bayesian models of communication. In parallel, large language models (LLMs) have been increasingly applied to semantic parsing applications, tasked with inferring logical representations from NL. While existing LLM explorations have been largely restricted to literal language use, in this work, we evaluate the capacity of LLMs to make inferences over pragmatic utterances. Specifically, we explore the case of threshold estimation on gradable adjectives, e.g., “strong”, contextually conditioned on a strength prior, then extending to composition with qualification, negation, polarity inversion, and class comparison. We find that LLMs can derive context-grounded, human-like inferences over interpretations for complex utterances such as “very strong for a beginner”, but struggle with composing such amortized inferences with negation or polarity inversion.

CogSci 2023

LLMs are pretty strong pragmatic reasoners: Investigating amortized inferences about gradable adjectives in large language models

### Welcome to ACL 2023, the 61st Annual Meeting of the Association for Computational Linguistics! 
 The conference will be held in Toronto, Canada, July 9-14, 2023. 
Following the succession of the recent conferences in our field, ACL 2023 will adopt a hybrid format.
While the impact of Covid has considerably diminished in terms of traveling, obtaining visas to Canada
entails a very long process. Moreover, the global economic conditions pose challenges for many individuals to travel to conferences. Recognizing these circumstances, we know many participants may not be
able to attend the conference in person. Therefore, we are committed to providing a great virtual platform
so everyone has the opportunity to interact with other participants and enjoy the conference. Based on the
current registered participants, approxiately 30% have chosen to attend the conference virtually. Whether
you join us in person or virtually, we sincerely hope everyone has a remarkable conference experience. 
This General Chair’s message is where I express my gratitude to the many individuals who have made
enormous contributions to the conference over the past year.

Read [**ACL 2023 General Chair's message**](https://docs.google.com/document/d/1WobYM7norbG4dI48s75HfJoD89qgX5a_F-6U8AteLSA/edit?usp=sharing/) in full.

##### **[Conference Handbook](https://2023.aclweb.org/downloads/acl2023-handbook.pdf)**

ACL 2023

The Association for Computational Linguistics (ACL) is the premier international scientific and professional society for people working on computational problems involving human language, a field often referred to as either computational linguistics or natural language processing.

poster

The 45th Annual Meeting of the Cognitive Science Society will take place in Sydney, Australia as a modified hybrid conference. Presentations from the main-session room will be live-streamed every day for all events. This will include all invited talks and award sessions. Recordings of all CogSci 2023 sessions will be made available for on-demand viewing after the conference. 
The in-person component will be held at the International Convention Centre Sydney at Darling Harbour. The Conference will take place Wednesday July 26th – Saturday July 29th, 2023 (Wednesday, July 26th is the Workshops Day).

Join us!

 [**The Program at a Glance can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2023/07/COgSci-PAAG-V2.pdf)
 [**The Detailed Program can be found here**](https://online.fliphtml5.com/ebtyf/mswz/)
 

#### **CogSci 2023 Program Book**
<iframe style="width:700px;height:400px" src="https://online.fliphtml5.com/ebtyf/tzgk/" seamless="seamless" scrolling="no" frameborder="0" allowtransparency="true" allowfullscreen="true" ></iframe>

The 45th Annual Meeting of the Cognitive Science Society presents the latest research across cognitive science and highlights the theme of Cognition in Context.

Recent research in experimental language evolution has considered how a pressure for efficient communication---i.e., the trade-off between production effort and communicative accuracy---impacts language structure. For instance, there is evidence that the use (or non-use) of case marking is influenced by this trade-off: both in natural languages, and in experimental studies of artificial language learning, case marking of grammatical arguments is more likely to be used in languages with flexible word order, and less likely in languages with fixed word order as a cue to argument structure. However, experimental evidence also suggests that social biases can influence efficiency. Language learners are themselves less efficient when there is a social bias in favour of a group whose productions are inefficient---e.g., who use case redundantly in a fixed word order language. Like the vast majority of studies in experimental psychology, studies on communicative efficiency and other cognitive factors that might shape language have been conducted on neurotypical populations. Here, we examine how the influence of social biases on efficient communication is impacted by autistic traits. We find that autistic people's use of case in the absence of a social bias is comparative to their neurotypical peers. However, we also find evidence that autistic people adhere more to social biases. In particular, they increase production effort in order to behave more like the group they are biased towards. We argue that some autistic people may be more likely to adhere to a social bias as a result of learnt social behaviours. More generally, these results underscore the importance of studying more diverse populations in language evolution research.

How communicative efficiency and social biases shape language in autistic and allistic learners

Patterns of language use change over time and may reflect and/or impact lexico-semantic representations across the lifespan.
In the current study, we quantified differences in lexico-semantic organization of different age groups and tested how they may be related to diachronic language change (measured using usage patterns in corpora from different decades).
Using diachronic word embeddings, we sampled words that have changed in meaning and words that have maintained the same meaning/usage patterns and tested how relatedness judgments differ when paired with word "neighbors'' from earlier vs. recent decades, for both younger and older adults.
Our findings suggest that individuals continuously update their lexico-semantic representations across the lifespan, and that the age-related differences may be due in part to the idiosyncratic language experience of older adults.
These results demonstrate the potential importance of considering language change when investigating lexico-semantic representations and aging.

Diachronic Language Change and Its Influence on Lexico-semantic Representations Across the Lifespan

As humans, we make sense of the world by organizing our daily experiences in terms of structured events. Events involve multiple participants, referred to as thematic roles. Here, we explore how thematic roles map onto language by investigating how speakers of two typologically different languages (English, Turkish) describe events that involve more peripheral event participants (Recipients, Instruments), whose conceptual status is highly debated. We manipulated the conceptual status of event participants by including events that either ‘require’ or ‘allow’ Recipients and Instruments.
Results showed that the require-allow distinction did not affect mention of recipients and instruments in any language. However, mention of highly optional event participants (e.g., allowed instruments) was affected by language-specific syntactic encoding options. We conclude that mention of participant roles in language is affected by factors beyond the cognitive salience of the roles.

A cross-linguistic investigation of possession-transfer and instrument events in language production

There is strong evidence about the preactivation of semantic information but controversial results about phonological preactivation. This research explored the individual differences in phono-semantic preactivation using the visual world paradigm. Participants looked at four competitors (semantic, phonological, and two unrelated) while hearing highly constraining sentences. Moreover, they were evaluated in verbal and nonverbal speed processing and working memory. Our results showed a strong semantic prediction but an inhibition of the phonological effect. These results were related to verbal and nonverbal working memory but not processing speed. The results were discussed in terms of lexical selection and inhibitory top-down influences.

Phono-semantic prediction during language comprehension: Effects of working memory

Recording the neural activity of multiple and large neural populations in the brain is becoming widely available with modern recording techniques. It still remains a challenge, however, to understand how distinct and anatomically different neural populations interact with each other to control behaviour. We propose a new method to discover causal interactions between neural populations based on recurrent switching dynamical systems. We introduce an extended dynamics model that incorporates the current time-step when calculating the latent state variables. We also introduce an acyclicity constrain in learning the parameters of the model. These mechanisms enable rich causal interactions between neuronal populations to be identified from the learned model. Our model outperforms previous work on discovering interactions between neural populations in simulated datasets, without sacrificing the prediction performance of firing rates. We also apply our method on real neural recordings from two macaque monkey brains performing a behavioral task, and show that the proposed method is able to detect causal interactions between brain regions related to the different time windows of the task.

Discovering Low-Dimensional Causal Pathways between Multiple Interacting Neuronal Populations

Non-CNS cancer and its treatment have been shown to affect memory in survivors, which is both objectively measurable and self-reported by patients. 

However, the traditional memory sum score measurement grants little insight into which memory processes are specifically impaired. 

The current research used a computerized visual-presentation version of the Auditory Verbal Learning Test, which is part of the Amsterdam Cognition Scan, to gather more detailed and precise data on test performance. 

Participants of this study were breast cancer survivors who received chemotherapy (n=68), and women controls (n=157). 

We analyzed these data through various hierarchical Bayesian cognitive models, with the goal to disentangle general performance metrics into more theoretically meaningful concepts. 

Results indicate that the memory processes in cancer patients are not generally impaired across all functions, but rather a selective impairment in 'memory retrieval'. 

There seemed to be no significant issues with patients in 'learning' or 'storage' processes.

Computational Modeling of Memory Processes in non-CNS Cancer Survivors

How does our environment impact what we will later remember? Early work in real-world environments suggested that a matching encoding / retrieval context improves memory. However, some laboratory-based studies have not replicated this advantageous context-dependent memory effect. Across three virtual reality studies, we find support for context-dependent memory effects: participants remembered more when placed in the same context as during encoding. This advantage has a tradeoff of falsely ‘recognizing’ similar lures, however. In addition, we do not find that the object-environment relationship is as impactful on context-dependent memory effects as prior studies suggest. Lastly, we find that differences in the presence (or absence) of dynamic background components differentially elicits the benefits of context-dependent memory. These findings not only add to our understanding of when and how context affects our memory, but they also present an exciting and more naturalistic approach to studying such effects.

Unpacking how Context Reinstatement aids Memory using Virtual Reality

A basic question in the study of human cognition concerns the ability to encode and retrieve associations between pairs of meaningful elements such as words. We employed a standard version of a paired-associate learning task using novel, arbitrary word pairs (restricted to nouns) presented one pair at a time during study without direct warning of a memory test to follow. After a distraction task, participants were provided with a cued-recall task which provided either the first or second word from each of the studied pairings. This paradigm was used to investigate the impact of two study tasks inspired by research in the domain of higher-order cognition. Specifically, we proposed a benefit in cued recall due to either: 1) articulating a conceptual combination of the two provided concepts, i.e., generating a novel integrative concept that might be referenced by the noun-noun pairing; or 2) articulating a specific proposition about how the two provided concepts could interact by fulfilling the roles of a semantic relationship. To illustrate, a pair of words like "cloth" and "donkey" might be integrated via conceptual combination to refer to a stuffed animal or via relational linking to indicate that a cloth could be draped over a donkey. The control condition in the experimental design was a simple imagery-based support task to allow assessment of the impact of the novel support tasks above and beyond the role of imagery. The results show that both study tasks dramatically improve memory performance.

Conceptual Integration and Semantic Relational Processing as Study Tasks to Promote Cued-Recall of Word Pairs

Memory retrieval is influenced by both prior and current experiences. The various factors (e.g., frequency, recency, or similarity) may interfere with retrieval due to prior experiences, while the context-dependent memory effect may enhance based on present experiences. Most memory research has been limited to controlled laboratory settings, but this study aims to examine memory retrieval in a more natural setting by using a GPS application (e.g., Traccar Client) to track participants’ daily GPS locations every 60 seconds for 5 weeks. Participants were then asked to recall their locations at a specific time, choosing from all locations visited in the previous 4 weeks. Results demonstrated the existence of context-dependent memory effect in real-world settings, with more frequent or recent visits leading to increased correct responses. This study is the first to use the current methodology to study the context-dependent memory effect and to measure an individual’s genuine memories in a more ecologically valid way.

Context Dependent Memory in the Wilds

Research has demonstrated an advantage for self-associated stimuli for a range of mental functions (Sui & Humphreys, 2017). However, relatively less is known about whether the prioritization also exists for internal representations (Yin et al., 2019). In the current study we first asked participants to associate social – labels (self, friend, stranger) with arbitrary geometrical shapes (triangle, quadrilateral, and pentagon) as in (Sui et al., 2012) and then tested them for the maintenance of one or more features (shape, location, or a combination) of the target stimuli during a delayed match – to – sample task. In line with our expectations, our participants indeed showed a distinct advantage for self- associated stimuli not only for the maintenance of single features (identity, location) but also for a combination (shape & location). Our findings are in line with the proposal that self-refence may aid in binding information in working memory (Sui & Humphreys, 2015).

Premium content

Next from CogSci 2023

How communicative efficiency and social biases shape language in autistic and allistic learners

Similar lecture

Evaluating statistical language models as pragmatic reasoners