Search

People repeat each other words in conversation, for example by using the word "brolly" instead of the word "umbrella" after their interlocutor used "umbrella". In other words, people entrain at a lexical level (Branigan et al., 2011). Lexical entrainment has been described as a consequence of alignment at the level of lexical representations, based on priming (Pickering and Garrod, 2004) but also as an audience design mechanism (e.g. speakers may repeat their interlocutors' words to facilitate their comprehension, in particular when their linguistic competence is assumed not to be optimal) (Brennan and Clark, 1996) or a combination of the two (Branigan et al.,2011). Speakers entrain with their interlocutors regardless of who they are. However, some elements of interlocutors' identity can modulate entrainment: Not only do people repeat their interlocutors' words when they interact with other people, but also when they interact with computer systems (such as virtual agents or social robots) (Branigan et al.,2011; Cirillo et al., 2021). Branigan et al. (2011) showed that the tendency to entrain is modulated by the perceived competence of the interlocutors (i.e. speakers entrain more often with systems that are perceived as less competent), which supports the audience design explanation or the combination between audience design and priming. However, it is still not clear how automatic mechanisms (based on priming) and strategic mechanisms (based on audience design) interact. Ivanova et al. (2020) proposed that attention might have a role: People might pay more attention to interlocutors when they assume they would need more help for comprehension. Attention would increase the depth of processing and encoding of linguistic information which would lead to greater alignment, and entrainment consequentially. In the current experiment, we replicate Branigan et al. (2011)'s experiment, by manipulating participants' beliefs about their interlocutor. One group of participants will play a picture matching/picture naming task with a virtual agent which is presented as highly competent (high competence condition), while another group will play the same task with the same virtual agent, presented, this time, as a beta version (low competence condition). Moreover, half of the participants will be asked to perform a secondary memory task (load condition). Additionally, all participants will be asked to perform a surprise follow-up memory task (word recognition task) that will allow us to investigate whether they attended to interlocutors' linguistic behaviour to a different extent in different conditions and whether attention boosts entrainment.

In conversations, speakers align with each other at several linguistic levels. Speakers mimic the way their conversational partners choose, combine and utter words. Moreover, speakers eventually converge to similar conceptualisations, reaching a common understanding of the situation under discussion (Garrod & Pickering, 2004). Why and how do speakers align with each other? Different explanations have been proposed. Whereas unmediated accounts explain alignment in terms of priming, mediated accounts explain alignment in terms of strategic choices employed to enhance mutual comprehension (Branigan et al., 2011) or to regulate social distance (Van Baaren et al., 2003). The two types of mechanisms are not mutually exclusive, indeed it is reasonable to think that priming effects might be modulated by strategic choices, when necessary. Lexical alignment, which, at a behavioural level, manifests itself as the repetition of each other’s words, is a suitable field to study how strategic and automatic mechanisms may interact (e.g. Branigan et al., 2011; Suffill et al., 2021). Sociolinguistic studies suggest that social status, community membership and power dynamics might influence linguistic alignment (Dragojevic et al., 2016). More specifically, alignment may be seen as an attempt to reduce social distance, by increasing social affiliation. Moreover, observational studies suggest that speakers tend to align more to higher status partners (Aronsson et al., 1987; Gregory & Webster, 1996). However, it is not clear whether such tendency can be attested in experimental settings and whether it also concerns lexical alignment. Admittedly, experimental settings - simplified and controlled approximations of real-life dialogues - might indeed reduce the impact of social information on lexical production. Social information might modulate speakers’ tendency to align only when speakers draw their attention to it, as when the context requires to take partner-specific information into account (Pickering & Garrod, 2004). In our present study, we adopt a widely used experimental paradigm that allows measuring lexical alignment in different conditions (Tobar-Henríquez et al., 2019). In an online experiment, we ask participants to match and name some objects following their partner's directions, which consist of a referring expression for each object. We operationalise lexical alignment as the participants' tendency to repeat low frequency referring expressions previously used by their conversational partner. In fact, participants don't play with human partners, but with a bot. Participants play two sessions of the same task with two different partners (in fact, the same bot) and stimuli. Before each session, we ask participants to provide a written description and a picture of themselves, and we provide them with a written description and a picture of their partner. We carefully created the descriptions by combining information about education level, job and hobbies to produce high and low-status fictitious partners. We investigate whether partners' social status modulate participants' tendency to align at a lexical level. Moreover, by asking the participants to rate the social status of their partner before or at the end of the task, we investigate to what extent the attention paid to such social information influences lexical alignment.

This study investigates the effect of two types of cognitive load manipulations (dual task paradigm) on the amount and types of listener responses in a collaborative referential game.

Dialogue requires coordination: speakers need to plan what to say, when to say it and how to adjust to the contributions of their interlocutors. Nonetheless, dialogue is the easiest way speakers possess to come to similar conceptualisations of the world (Pickering & Garrod, 2004). We aim to study how dialogue supports mutual understanding, in particular when it comes to contestable concepts (e.g. “beauty”, “democracy” or “fairness”) or concepts associated with terms for which “there is no one clearly definable general use […] which can be set up as the correct or standard use.” (Gallie, 1956, p. 168). Indeed, whereas speakers can share similar representations of non-contestable concepts (e.g. the number of freely accessible parks Edinburgh) by referring to an epistemic authority (e.g. the City of Edinburgh Council), sharing similar representations of contestable concepts (e.g. how exciting living in Edinburgh is) may require more linguistic negotiation. We investigate how speakers negotiate contestable and non-contestable concepts, and in what measure the contestability of concepts influences speakers' alignment (here meant as the process of becoming more similar with respect to their mental representations of concepts). More precisely, we test whether believing that a concept (here operationalised as a ranked list of items) is contestable or not affects the process of alignment in conversation. Studying contestability will eventually help us understand the role of language use and social cognition in the representation of abstract concepts. Contestability, in fact, is also a characterising feature of abstract concepts, as they are characterised by individual and cultural level variability and they cannot be easily represented by attending directly to entities in the world (Borghi et al., 2017). In the current study, participants are asked to individually rank 15 items in a list from the most to the least useful in the desert. After that, participants are paired in dyads to discuss the items and then asked to rank the list a second time. In one condition (non-contestable condition), they are told that there is an objective ranking of the list as established by a recognised epistemic authority. In the other condition (contestable condition), speakers are not given any information that there is a factual ranking of the list and instead are told that the experimenters are aware of the ranking being debatable. We will compute the similarity between the ranked lists of speakers before and after the interaction. We predict that the lists provided by the pairs of participants in both conditions will become more similar after the interaction so that the two members of the pairs will align with each other. However, we also predict that the effect of the interaction in achieving greater similarity would be greater in the contestable condition, thus highlighting the role of the dialogue in coming to a similar understanding of contestable concepts. Participants, in fact, would not be able to appeal to any assumed standard that justifies being conservative with their first set of choices. Moreover, we plan to run an exploratory analysis of the data to identify different styles of negotiation across conditions, specifically focusing on feedback (Dingemanse et al., 2015).

Ostracism is a community-level phenomenon, shared by most social animals, including humans. Its detection plays a crucial role for the individual, with possible evolutionary consequences for the species. Considering (1) its bound with communication and (2) its social nature, we hypothesise the combination of (a) linguistic and (b) community-level features to have a positive impact on the automatic recognition of ostracism in human online communities. We model an English linguistic community through Reddit data and we analyse the performance of simple classification algorithms. We show how models based on the combination of (a) and (b) generally outperform the same architectures when fed by (a) or (b) in isolation.

On behalf of the Program Committee, a very warm welcome to the Seventh Italian Conference on Computational Linguistics (CLiC-it 2020). This edition of the conference is held in Bologna and organised by the University of Bologna. The CLiC-it conference series is an initiative of the Italian Association for Computational Linguistics (AILC) which, after six years of activity, has clearly established itself as the premier national forum for research and development in the fields of Computational Linguistics and Natural Language Processing, where leading researchers and practitioners from academia and industry meet to share their research results, experiences, and challenges.

Knowledge on porcine bitter and fat taste receptors and on their expression in gastrointestinal tract of pigs is scarce. We searched for the presence of porcine homologous sequences for 13 human transcripts of bitter and fat taste receptors in ENSEMBL and National Center for Biotechnology Information databases. For taste 2 receptor (TAS2R) 8, alignment was not observed; for TAS2R13 and TAS2R46 the porcine predicted sequence aligned with several other human bitter genes. For 7 genes for bitter taste (TAS2R1, TAS2R3, TAS2R7, TAS2R9, TAS2R10, TAS2R16, and TAS2R38) and for 3 genes for fat taste (GPR40, GPR43, and GPR120), a full homology for exon sequences was found and primers were designed by Primer3. These 7 genes were amplified with real-time PCR and verified on agarose gel in 5 gastrointestinal segments of weaned pigs: oxyntic (ST1), pyloric (ST2), and cardiac to oxyntic transition mucosa (ST3), jejunum (JEJ), and colon (COL). Suitability of mRNA was verified by amplifying RPL4 and HMBS2 genes. Each bitter taste gene was detectable on agarose gel in at least 1 subject of all the gastrointestinal segments except for TAS2R3 and TAS2R38 that were never detected in ST1 and COL, respectively. The inspection of bitter taste genes amplification curve indicated that the expression was in general very low. GPR43 and GPR120 were present in all segments from all pigs. Expression was not detected for GPR40. Data also indicate that colon is the preeminent tract where fat detection by GPR120 takes place (P0.001). The presence of gene expression for several chemosensing receptors for bitter and fat taste in different compartments of the stomach confirms that this organ should be considered a player for the early detection of bolus composition.

Pig meat quality is determined by several parameters, such as lipid content, tenderness, water-holding capacity, pH, color and flavor, that affect consumers’ acceptance and technological properties of meat. Carcass quality parameters are important for the production of fresh and dry-cure high-quality products, in particular the fat deposition and the lean cut yield. The identification of genes and markers associated with meat and carcass quality traits is of prime interest, for the possibility of improving the traits by marker-assisted selection (MAS) schemes. Therefore, the aim of this thesis was to investigate seven candidate genes for meat and carcass quality traits in pigs. In particular, we focused on genes belonging to the family of the lipid droplet coat proteins perilipins (PLIN1 and PLIN2) and to the calpain/calpastatin system (CAST, CAPN1, CAPN3, CAPNS1) and on the gene encoding for PPARg-coactivator 1A (PPARGC1A). In general, the candidate genes investigation included the protein localization, the detection of polymorphisms, the association analysis with meat and carcass traits and the analysis of the expression level, in order to assess the involvement of the gene in pork quality. Some of the analyzed genes showed effects on various pork traits that are subject to selection in genetic improvement programs, suggesting a possible involvement of the genes in controlling the traits variability. In particular, significant association results have been obtained for PLIN2, CAST and PPARGC1A genes, that are worthwhile of further validation. The obtained results contribute to a better understanding of biological mechanisms important for pig production as well as for a possible use of pig as animal model for studies regarding obesity in humans.

Fat deposition is a crucial aspect of pig meat quality: a suitable marbling and the right balance between lean meat and fat in carcass are both essential features to produce high quality dry cured hams. Moreover fat content influences both organoleptic and nutritive characteristics of meat products and consumer acceptance. Differences among breeds in intramuscular fat content and backfat thickness have been reported but the molecular events underlying the different aptitude to fat deposition are poorly known. The analysis of the expression profile of lipogenic genes in muscle and fat tissues in different pig breeds can contribute to a better knowledge of fat deposition control in pig meat and can provide new insights into carcass adiposity. The objective of this study is to investigate the expression level of the key adipogenic genes ATP citrate lyase and Fatty acid synthase in Italian Duroc and Italian Large White pigs with different fat deposition capacities. The obtained results suggest a tissue-specific mechanisms regulating gene transcription of FASN and ACL in the analysed samples. In adipose tissue, differential expression between the analysed breeds was observed and ILW pigs showed the highest FASN transcriptional level. In skeletal muscle tissue the greatest mRNA abundance of both lipogenic genes in ID pigs could be consistent with the high intermuscular fat deposition aptitude of this breed.

In pigs, several studies have been focused on structural and regulatory muscle genes and their possible relationship with meat production trait as they may play key role in muscle mass accretion. In the present work, we investigated four skeletal muscle genes, calpain 1 (CAPN1), calpastatin (CAST), myopalladin (MYPN) and titin (TTN). These genes were mapped on porcine chromosomes 2p14-p17, 2q21-q24, 14q25-q29 and 15q23-q26, respectively, and located in regions where quantitative trait loci for carcass traits were found. The calpain protein (coded by CAPN1 gene and expressed in skeletal muscle and other tissues) is involved in the early post mortem protein degradation promoting the initial disruption and destabilization of the myofibrillar structures. Calpastatin is a specific inhibitor of ubiquitous calpains and its post mortem activity influences meat tenderization process. Myopalladin is a sarcomeric protein which has central roles in the organization and assembly of the Z-line and seems to affect regulatory mechanisms of muscle gene expression. TTN encodes a large abundant protein of striated muscle that is involved in muscle assembly and ultrastructure and it is also implicated in muscle elasticity. To evaluate the putative effects of these four genes on carcass quality traits, we analysed single nucleotide polymorphisms (SNP) identified in the 3’ untranslated regions (MYPN and TTN) and in introns (CAST and CAPN1) in 272 sib tested Italian Large White (ILW) pigs and 114 sib tested Italian Duroc (ID). General linear model procedure of the SAS package was used to search for association between the SNPs and the productive traits average daily gain (ADG), backfat thickness (BFT), lean cuts (LC), ham weight (HW), feed conversion ratio (FCR) for both breeds and visible intramuscular fat (VIF) only for the ID breed. Statistically significant results for the association analyses were found for CAST in the ILW breed, with ADG, LC and FCR (P < 0.05) and in the ID breed with ADG, LC, BFT, HW (P < 0.05). Moreover, in the ID breed, the MYPN gene was associated with all considered traits (P < 0.01). The results confirm that CAST and MYPN could be considered positional and biological candidate genes affecting carcass traits, as previously reported in literature.