5Dec 2020
00:00 UTC
#linguistweets
#abralin

Program

5Dec 2020 03:00 UTC
5Dec 2020 00:00 Local Time *

Prosódia e entoação: as questões disjuntivas do PB

Leonardo Machado (UFRJ - Brasil)

Co-author(s): Cláudia Cunha (Brasil)

As interrogativas disjuntivas são aquelas que oferecem as opções de resposta em sua composição: "quer uva ou maçã?". Buscando responder quais são e como se manifestam as marcas dialetais do PB, investigamos a entoação desse subtipo frásico em gravações de 25 capitais brasileiras.
A interrogativa disjuntiva é aquela cuja formulação oferece uma opção entre dois, ou mais elementos, dos quais, um deles constituirá a resposta. Em sua construção, sempre haverá a conjunção alternativa “ou” operando a alternância entre os termos (Moraes 1982, 1984), por exemplo, “Você quer leite ou café?”. A presente pesquisa faz parte de um trabalho maior o qual descreveu as características prosódicas da interrogativa disjuntiva e respondeu como elas recaem nas variedades do português brasileiro (PB). A justificativa para esta investigação surgiu à medida que poucos estudos contemplaram a caracterização do tipo frásico em análise, destacando trabalhos recentes como Lira (2009) e Rosignoli (2017). Para a recolha dos nossos dados, recorremos ao questionário de prosódia do Atlas Linguístico do Brasil (ALiB), projeto ao qual esta pesquisa integra-se. Foram analisados dados de fala das 25 capitais brasileiras compreendidas pelo Atlas. Em cada cidade, foram inspecionadas as produções de oito informantes (quatro homens e quatro mulheres) estratificados, equitativamente, segundo a faixa etária e escolaridade. A descrição e a análise dos nossos dados levou em conta a apuração acústica e instrumental das incursões do parâmetro acústico da frequência fundamental (doravante F0) no interior das sílabas que compõem a pergunta: pré-núcleo, primeiro e segundo núcleo disjuntivo. Para propormos a representação fonológica das sentenças, adotamos a abordagem Autossegmental e Métrica da Fonologia Entoacional (Pierrehumbert 1980; Pierrehumbert & Beckman 1988; Ladd 1996, 2008.) e, para delimitação dos contornos melódicos disjuntivos, nos apoiamos na Fonologia Prosódica (Selkirk 1984, 1986, 2000; Nespor & Vogel 1986, 2007), uma vez que, segundo autores (Bartels 1999; Pruitt e Roelofsen 2013), a questão disjuntiva difere-se dos demais tipos frásicos interrogativos por ser mapeada em dois constituintes prosódicos. Os resultados obtidos a partir de nossa investigação podem ser situados em pontos específicos da composição do tipo frásico: 1) o pré-núcleo mostrou-se descendente em todas as cidades estudadas, porém há possibilidade de realização de um pico inicial cuja altura é menor do que o pico nuclear, sendo encontrado, principalmente, nas regiões norte e nordeste; 2) o movimento ascendente característico do primeiro núcleo disjuntivo difere as localidades de acordo com a sílaba que atinge o pico do enunciado, sendo a tônica nuclear nas localidades norte e centro-oeste e as postônicas nas demais regiões; 3) em relação ao movimento associado às sílabas finais da pergunta, a disjunção com final descendente e fronteira baixa realizou-se em todos os dados, porém foi possível encontrar sílabas finais com ascendência e associação de fronteira alta nos falares do norte e do nordeste. Tal fato nos permite dividir as marcas regionais encontradas na disjunção em dois eixos distintos: norte-nordeste e centro-sul.
5Dec 2020 03:15 UTC
5Dec 2020 00:15 Local Time *

Determinatives are far from pronouns in English

Brett Reynolds (Humber College - Canada)

Using 232 morphological, syntactic, semantic & phonological features and the DISCO test, I show for the first time that determinatives & pronouns are statistically significantly different categories (p<0.001). This holds for the subsets of features (morph, syn, & sem).
Linguists often propose diagnostic criteria for lexical categories in various languages, but statistical estimates of our confidence in the resulting categories are neither common nor expected. In fact, there is no accepted method for arriving at such estimates. Contrast this with other fields, such as medical research, where such statistics are expected.

It is not that there is universal agreement within linguistics about lexical categories, even for a language as commonly studied as English. In fact, the very distinction between lexical categories has been called into question, with any attempt to justify such categories described as “methodological opportunism.”

To address this gap, I demonstrate novel graphical and statistical methods to quantify the level of confidence in the diagnostics and their resulting categories.

The methods involve encoding the morphological, phonological, semantic, and syntactic features of the English determinatives and pronouns in a 138-words × 232-features table. One method uses multiple correspondence analysis – correspondence analysis for high-dimensional, categorical data – to reduce the dimensionality of the resulting data, which allows the display of the relative position of the words in a two-dimensional graph. The other uses the distance components test from the “energy” class of statistics to measure the distance between the two categories relative to their average internal distance and to estimate the probability that they are more distinct than would be expected by chance. It can also identify the features contributing the most to the categorizations.

Using these methods on the full data set, I find that determinatives and pronouns are significantly different categories in English. Moreover, rerunning the tests while limiting the data to only the morphological features, only the semantic features, or only the syntactic features did not change the basic results, undercutting the claims of methodological opportunism.
5Dec 2020 03:30 UTC
5Dec 2020 00:30 Local Time *

Emoji based reactions to the Said Construction

Alicia Stevers (San Diego State University - USA)

Said used in place of a determiner seems to carry a unique social meaning. I present results of a study where participants were more likely to react to said containing sentences with 😂. Identical sentences with a standard det got a 👍 reaction
My research focuses on the said construction (SC), a standard English construction usually characterized by the use of ​said in place of a determiner, followed by a noun (N2), typically given (in some sense) and licensed by an antecedent noun (N1):

When arguing a point with an opponentN1, she was accustomed to swift deference (provided that said opponentN2 was male, straight, and his eyes worked). [Freisner 2011]

I present results of a social media based experiment, focusing on the way participants react to SC-containing sentences using the “reactions” buttons provided on many social media sites. Participants were presented with stimuli that looked like a Facebook profile; each item in the experiment had a “profile photo”, a sentence that contained either a standard determiner like ​the or ​that, ​or SC, as well as an emoji-based reactions schema like the buttons provided on Facebook. Participants were told to read each sentence and react to it using the provided emoji buttons. Results showed that SC-containing sentences were more likely to receive the “😂/haha” reaction, whereas sentences containing a standard determiner received the “👍/like” reaction. These findings suggest that SC may have a humorous meaning that is not conveyed by a standard determiner.
5Dec 2020 03:45 UTC
5Dec 2020 00:45 Local Time *

#AboriginalEnglish: BE LIKE, stability and change

Celeste Rodriguez Louro (Australian Research Council/The University of Western Australia - Australia)

Co-author(s): Glenys Collard (The University of Western Australia - Australia)

Our synchronic/diachronic dataset offers broad view of #AboriginalEnglish quotation. Findings based on 48 speakers b. 1907-2005 allow tackling ‘actuation problem’: why changes occur in some varieties, not others.
In settler Englishes, including standarised Australian English, the quotative system has shifted from introducing third person speech in the past tense to expressing first person thought in the historical present. Despite these findings, little is known about whether and how these longitudinal trends apply to Indigenised Englishes, including Australian Aboriginal English (AAE).

In this paper I draw on synchronic and diachronic datasets to offer a broad view of AAE quotation. My findings, based on the speech of 48 AAE speakers born between 1907 and 2005, offer a continuous link between present-day AAE quotation and the early 20th century. Analysis of 2,404 quotative tokens reveals an overwhelmingly stable system across the generations. Those born after 1983 add BE LIKE to their quotative repertoire but BE LIKE does not take over say which persists as the prime speech encoder across the board. Neither does the ingress of BE LIKE bring about radical change in the types of quote content introduced. Both before and after the arrival of BE LIKE, the most frequent content type encoded is speech (95% [2283/2404]).
The persistence of direct speech reporting in AAE ‘yarning’ (storytelling) may be explained by considering narrative syntax, the role of hearing in Aboriginal culture and a strong oral tradition underpinning the dataset in the study. Taken together, these factors help us tackle the well known ‘actuation problem’ in sociolinguistics: the question of why linguistic changes occur in some varieties, but not in others.
5Dec 2020 04:00 UTC
5Dec 2020 01:00 Local Time *

Perception of American English pure vowels

Naeimeh Afshar (University of Pannonia - Hungary)

Co-author(s): Vincent J. van Heuven (University of Pannonia - Hungary), Judit Navracsics (University of Pannonia - Hungary)

We study the perception of American-English (AE) vowels by Persian and Persian/Azeri (Turkic) learners. We test their perceptual assimiltion and compare the mapping of AE vowel space by L1 listeners and our learners, using a synthesized vowel space varying formants and duration.
The idea of measuring distance between languages seems to have its roots in the work of the French explorer Dumont D’Urville. He collected comparative word lists of various languages during his voyages aboard the Astrolabe from 1826 to 1829 and, in his work about the geographical division of the Pacific, he proposed a method to measure the degree of the relation among languages. The method used by modern glottochronology, developed by Morris Swadesh in the 1950s, measures distances from the percentage of shared cognates, which are words with a common historical origin. Recently, this study will propose a method when it gets into an application that uses normalized Levenshtein distance among words with the same meaning and averages on the words contained in a list. In related studies by several researchers proposed an advanced method of the intended method including a second normalization. In this project, we will compare the different dialects of the Arabic Language to decide on which is the closest to Modern Standard Arabic (MSA) to decide which of the two closely related dialects can be applied with greater success to resolve relationships among languages.
Expected results will show the most accurate dialect to MSA that can be used as a reference to anyone interested in reading or writing in the Arabic language that can be generalized as much as people and not to be specified for a specific group of people and excluding those who can find a problem understanding the full message while using the least used or far terms and words from the common knowledge of the Arabic language. And that closer related Dialects have smaller distance and spread through the Arab world. This means that not only do they have more similar words for the same meaning, but the general occurrence and order of characters in words is more similar. That will also provide a slight insight to think about the future of the language if it is turning towards global spread or attrition, by comparing the number of the places and its speakers for each of the main Dialects representing the Arabic Language that will be selected.
References
D. D’Urville, Sur les ˆıles du Grand Oc´ean, Bulletin de la Soci´et´e de
Go´egraphie 17, (1832), 1-21.
Keywords: lexical distance, dialects, Arabic language, phonotactic features.
5Dec 2020 04:15 UTC
5Dec 2020 01:15 Local Time *

Interactionally situating the power scream

Emily Hofstetter (Linköping University - Sweden)

Language emerges from bodies, but some sounds are treated by interlocutors as particularly ‘bodily motivated’. This study uses conversation analysis to study how a corpus of grunts (365) blur the lines between ‘physiological byproduct’ & ‘social display’ #nonlexicals #emca
Everyday interaction is filled with sounds that are connected to bodily events: breathing, sniffing (Hoey, 2020), crying (Hepburn, 2004), grunting, and so on. Most linguistic theories, however, exclude the body from consideration, arguing its contributions are insufficiently symbolic or conventionalized, and most of all, merely unintentional by-products. In contrast, ethnomethodological and interactional linguistics studies show how speakers orient to such sounds as meaningful and accountable (e.g. Keevallik 2020).

This paper begins to address the lack of research concerning ‘bodily motivated’ sounds, specifically the diverse vocalizations involved in physical strain, using a corpus of 25 hours of recorded naturally occurring rock climbing as data (368 strain vocalizations). It uses the term ‘bodily motivated’ to capture the dual nature of these sounds as both physiological and social events. The analysis proceeds using multimodal conversation analysis, that is, it elucidates the pragmatics of the vocalizations as demonstrated through the participants’ own orientations to the phenomena in situ. The participants in the data treat the vocalizations as simultaneously physiological and socially meaningful for their ongoing interaction, which makes available their understandings of the vocalizations for both co-participants at the time of recording, and the analyst for research.

Participants distinguished between three variations: 1) ‘power screams’, laryngeally constricted vocalizations made during intense physical strain, 2) ‘power hups’, shorter bursts with glottal onsets that accompany motion, often jump-like moves, and 3) ‘strain releases’, outbreaths that accompany a relaxation of muscles. The former are treated as projecting motion, whereas the latter projects cessation. Climbers have a particular need to distinguish strain vocalizations as safety partners (belayers) must anticipate the climber’s motions in order to safely manage the rope equipment.

This study proposes the concept of ‘bodily motivated’ sounds as an emically defined form of vocalization that blurs the boundaries between the ‘merely physiological’ and ‘socially meaningful display’. Such vocalizations provide a perspicuous opportunity to examine how the body and language are interwoven.

Hepburn, A. (2004). Crying: Notes on Description, Transcription, and Interaction. Research on Language and Social Interaction, 37(3), 251–290.
Hoey, E. M. (2020). Waiting to Inhale: On Sniffing in Conversation. Research on Language and Social Interaction, 53(1), 118–139.
Keevallik, L. 2020. Linguistic structures emerging in the synchronization of a Pilates class. In C. Taleghani-Nikazm, E. Betz & P. Golato (eds), Mobilizing others: Grammar and lexis within larger activities (pp.147-174). John Benjamins.
5Dec 2020 04:30 UTC
5Dec 2020 01:30 Local Time *

Why are we *quick* to point out, but not *fast*?

Martin Schäfer (Universität Tübingen - Germany)

We are *quick*, but not *fast*, to point out, because of 'quick's inceptive readings. These and a volitional agent are typical of the construction, as opposed to the competing '-ly' forms, but also alternative ways of combining with an infinitival ('Pasta is fast to prepare').
All English speed adjectives have attributive ('a quick swimmer'), predicative ('she is quick'), and adverbial usages ('she moved quickly'). But while quick/slow/swift are fine with a following infinitival, cf. (1), fast/rapid/speedy almost never occur in this construction.

(1) She was quick/slow/swift to condemn the attack.

This is surprising in view of their similar versatility with regard to the other usages, and in view of theoretical accounts holding all these adjectives to be event predicates. The existence of the construction itself is surprising, as it seems very similar to adverbial constructions such as (2).

(2) She quickly condemned the attack.

My thread shows the following:
a) Quick/slow/swift also pattern together in
other constructions, and in particular lend themselves to inchoative and holistic readings. Thus, 'a quick walk' and 'a fast walk' can refer to the same event, but typically, the former refers to a walk that took only a short time (= holistic reading), while the latter refers to a walk at a higher speed.

b) The core difference between the ADJ-to-infinitival construction and the competing -ly forms lies in the preferences with regard to inchoative and rate-of-speed readings on the one hand and a strict requirement on volitional, controlling agents on the other hand. Normal stones can quickly roll down a slope, but stones that are quick to roll down a slope must be animate (or belong to the construction illustrated in 3a)

c) If adjectives from the other set come with a to infinitival, they typically represent other patterns, like (3a) where pasta is not the agent of the action but the patient, or are part of larger other constructions coming with their own semantics, like (3b), where too ADJ-INF leads to an interpretation where the event referred to in the to-verb is actually not carried out.

(3a) Pasta is fast to prepare.

(3b) She was too fast to stop.

Overall, my aim is to lay bare the amazing interconnectness between the usages of the same words in different environments, and the surprising factors that play a role in the choices for a specific word and construction.
5Dec 2020 04:45 UTC
5Dec 2020 01:45 Local Time *

What the Italian subjunctive "actually" means

Salvio Digesto (University of Ottawa - Canada)

Is the Italian subjunctive used to convey doubt, desire, non-assertion? Is it disappearing from the language? Do southern Italians use it less/differently than northern Italians? Quantitative and statistical examination of spontaneous speech data says no
The use of the subjunctive (S) in Italian remains a matter of debate in the literature and there is little agreement on what triggers its use in discourse and where exactly it “must” be used in the context of completive clauses governed by verbs. Essentially, the S is claimed to convey readings of doubt, desire, uncertainty, non-assertion, speaker’s intent or state of mind. When it is not selected in subjunctive-selecting contexts, scholars suggest a shift in the meaning conveyed. On the other hand, the S has become a feature for speakers to parade as a sign of polished Italian, whereas its avoidance is often condemned as a feature of popular, uneducated speech. It is also considered to undergo loss, and that this supposed change is considered more advanced in southern Italian speakers. However, these issues have never been tested systematically under an accountable empirical methodology against a corpus of vernacular speech data. I address these issues by examining S usage in two corpora (LIP and C-ORAL-ROM; for a total of 607,487 words), providing a fair representation of everyday speech, as well as a variety of speech styles and speakers from different the four main urban centres of Florence, Milan, Rome, and Naples. Following previous variationist research, I identified the contexts of use corpus-internally, by locating all the contexts where the S is actually used to determine where it could be used, i.e., by limiting to matrix verbs that governed at least one S in the data. Quantitative analysis of 1,663 tokens of S selecting-contexts extracted from corpora reveals robust variability, with an overall rate of the S of 68% and a total of 141 governors that triggered it in discourse. However, despite its apparent productivity, comparison of the conditioning of S selection suggests otherwise: quantitative and statistical evidence shows that its selection is largely determined by the lexical identity of the matrix verbs (governors: bisognare, volere, credere, sembrare, pensare) as well as suppletive forms of 'essere', which suggest lexicalization, contra the assumption that its selection is semantically motivated. The highly lexicalized pattern on variability was found to be largely shared amongst the four main urban centres of Florence, Milan, Rome, and Naples, thus countering the assumption of divergent linguistic behaviour between northern and southern varieties of Italian. On a more socio-linguistic aspect, this study confirms the linguistic prestige that the S has acquired in contemporary speech, being selected with a wider range of infrequent and singleton governors by highly educated speakers. In fact, speakers with little or no formal education do not use the S in casual conversation to the extent that highly educated speakers do, whereas highly educated speakers are almost single-handedly providing the rich set of governors observed in our quantitative analysis and the amount of subjunctives is almost three times bigger in formal speech
5Dec 2020 05:00 UTC
5Dec 2020 02:00 Local Time *

Empirical methods for describing TAM

Ana Krajinovic (University of Düsseldorf - Germany)

I discuss different types of evidence for TAM descriptions offered by empirical methods, such as corpus work, storyboards and translation-based questionnaires, by focusing on irrealis mood and perfect aspect in Nafsan (Vanuatu, Oceanic).
The description of tense, aspect, and mood (TAM) categories is often considered to be one of the more difficult tasks in language description, due to the proliferation of TAM categories in the literature and complex relationships between different language-internal processes. In this paper I report on available empirical methods for studying TAM categories and my experience with using them in the study of Nafsan (Vanuatu, Oceanic). By focusing on case studies of relevant categories in Nafsan, such as irrealis mood and perfect aspect, I argue that corpus work, storyboards with targeted TAM contexts (Burton & Matthewson, 2015) and translation-based questionnaires accompanied by meta-linguistic discussions offer different kinds of evidence that are all necessary for a successful description.

A language corpus, if available, is a good start for finding the basic distribution of the grammatical element in question. For Nafsan, I studied the corpus data (Thieberger, 1995–2018) and the grammar of Nafsan (Thieberger, 2006), from where I identified functions of irrealis and perfect found in the corpus and I hypothesized about which functions would be expected from these categories, given their cross-linguistic properties. For instance, perfect is expected to have resultative, experiential, universal, and anterior (past perfect) functions, and be incompatible with temporal adverbs in present perfect (Comrie, 1976; Klein, 1994). Since some of these functions were not attested in the corpus with the Nafsan perfect, only the combination of questionnaires (e.g. Dahl, 2000), storyboards (e.g. Matthewson, 2014), and additional elicitation revealed the presence of experiential and universal functions, as well as the incompatibility with temporal adverbs in present perfect meanings and compatibility with temporal adverbs in past perfect meanings, similarly to the English perfect.

In my tweets, I will discuss the different types of evidence that were provided by corpus data, storyboards and elicitations, and highlight how these methods can offer insight into relationships between language-specific and cross-linguistic properties of TAM categories.

References:
Burton, Strang & Lisa Matthewson. 2015. Targeted construction storyboards in semantic fieldwork. In M. Ryan Bochnak & Lisa Matthewson (eds.), Methodologies in semantic fieldwork, 135–156. New York: Oxford University Press.
Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press.
Dahl, Östen. 2000. Tense and aspect in the languages of Europe. Berlin: Mouton de Gruyter.
Klein, Wolfgang. 1994. Time in language. London: Routledge.
Matthewson, Lisa. 2014. Miss Smith’s Bad Day. Totem Field Storyboards http://www.
totemfieldstoryboards.org.
Thieberger, Nicholas. 2006. A grammar of South Efate: An Oceanic language of Vanuatu. Honolulu:
University of Hawai’i Press.
Thieberger, Nick. 1995–2018. Guide to the Nafsan, South Efate collection.
http://www.nthieberger.net/sefate.html.
5Dec 2020 05:15 UTC
5Dec 2020 02:15 Local Time *

Positional Preference of Emotion Phrase in Hindi

Spandan Chowdhury (Jadavpur University - India)

Taking permutations of phrases denoting Stimulus, Response & associated Emotion in an event involving an emotional component, the paper finds emotion-specific preference of phrase ordering in Hindi and classifies Ekman’s six basic emotions in Hindi based on these characteristics
The conceptualization and expression of emotion is a natural function of Language (Foolen 1997, 2012). The conceptualization of a certain experience including its components, namely stimulus, response and the associated emotion, are reflected in the construal used in the language. The study aims to find if there is any specific preference between the available construals describing an event involving some emotional component (based on Ekman’s (1993) six basic emotion types).

The target language considered for this study is Hindi. Data has been collected from 56 healthy native Hindi-speaking participants (28 male and 28 female) between ages 16 years to 53 years (mean age = 25.46 years, SD = 6.66 years) after taking their consent. Construals of situations involving a Stimulus Phrase (S-phrase), a Response Phrase (R-phrase) and an Emotion Phrase (E-phrase) were presented and a grammaticality judgement test was carried out. Taking into account that stimulus always precedes response, the position of the E-phrase was varied sentence-initially (ESR ordering), sentence-medially (SER ordering) and sentence-finally (SRE ordering) for each of the six basic emotions (i.e., total 18 sentence combinations were obtained). It was observed that the SER ordering is the most preferred structure [p-value<<0.0001]. Further, among the lesser preferred structures, an interesting pattern was observed –the two emotions of fear and surprise have been found to be structurally opposite in nature, showing lesser syntactic variability with respect to the other four emotions. These, along with other observations made in this paper, point towards possible typological findings involving emotion realization in language.

References:
Croft, W., & Cruse, D. A. (2004). Cognitive linguistics. Cambridge University Press.

Damasio, A. (1994). Descartes’ error. Emotion, reason, and the human brain. London: Picador.

Ekman, P. (1993). Facial expression and emotion. American psychologist, 48(4), 384.

Ekman, P. (1999). Handbook of cognition and emotion in T. Dalgleish and M. Power (Eds). Wiley and Sons. Ltd. Sussex, U.K.

Foolen, A. (1997). The expressive function of language: towards a cognitive semantic approach. In: S. Niemeier & R. Dirven (Eds.), The language of emotions(15–31). Amsterdam: Benjamins.

Foolen, A. (2012). The relevance of emotion for language and linguistics. Moving ourselves, moving others: Motion and emotion in intersubjectivity, consciousness and language, 349-369.
5Dec 2020 05:30 UTC
5Dec 2020 02:30 Local Time *

We don’t agree (only) upwards

András Bárány (Bielefeld University - Germany)

Co-author(s): Jenneke van der Wal (Leiden University Centre for Linguistics - Germany)

Is agreement upward or downward? In our tweets we argue that Bjorkman & Zeijlstra’s (2019) proposal that Agree is upward checking and downward valuation is empirically not adequate. Data from Bantu languages, German, and Nez Perce suggest that upward valuation is not exceptional.
Bjorkman & Zeijlstra (2019; B&Z) argue that Agree involves two operations: checking and valuation. Checking always involves [uF] probing upwards for [iF] which c-commands it and can check it. Valuation generally happens downwards: [iF] c-commands [uF] and values it. In typical subject agreement, the subject DP has its Case feature [uT] checked by T, moves to SpecTP, and checks and values T’s [uϕ].

We argue that this approach is not supported empirically by data for subject agreement in Matengo, German, and Serbo-Croatian, object agreement in Sambaa and Liko, and complementiser agreement in Nez Perce. These agreement phenomena all involve configurations in which the agreement controller does not move to the probe’s specifier.

In Matengo, German, and Serbo-Croatian, subject agreement can involve both preverbal and postverbal subjects. For Matengo and German, evidence from quantifier scope, for example, shows that postverbal subjects are indeed lower than T. In Serbo-Croatian, preverbal and postverbal coordinated subjects lead to different agreement marking on the agreeing participle, with either the last conjunct (preverbal subject) or the first conjunct (postverbal subject). These agreement patterns are straightforwardly derived if the agreement controller need not move to the agreeing head’s specifier.

The Bantu languages Sambaa and Liko do not show any evidence for object movement in relation to object agreement. In Bantu languages, extraposition of agreeing objects and focus movement have been argued to motivate object movement: neither of these correlates with object agreement in Sambaa and Liko, however. We argue that the head agrees with in-situ objects without movement. Finally, in Nez Perce complementiser agreement, first and second person arguments can control agreement on the complementiser, but they need not move to its specifier either.

B&Z acknowledge scenarios of this kind, in which an agreement controller is lower than the head that spells out agreement. When the head’s specifier is filled by there, they argue that there checks the head’s [uϕ] features but cannot value them because it is ϕ-defective. This allows a lower phrase to exceptionally value the head’s [uϕ]. When the head’s specifier is empty but the head nevertheless agrees, B&Z propose that a null expletive pro fills its specifier. The failure of this defective pro to value the probe allows for exceptional valuation of the probe by a lower argument. We reject this argument, however, on the grounds that, first, null expletives cannot be motivated for languages such as Matengo, and second, positing a null, semantically empty element is unfalsifiable.

In sum, our data suggest that Agree relations without movement, involving upward valuation, are not exceptional and much more common than B&Z suggest. In addition, we argue that conceptual arguments about the nature of Agree and its directionality are premature if the theoretical model cannot account for the empirical reality.
5Dec 2020 05:45 UTC
5Dec 2020 02:45 Local Time *

The F2 Robot Interaction System

Liudmila Zaidelman (Kurchatov Institute - Russia)

Co-author(s): Anna Zinina (Kurchatov Institute - Russia), Nikita Arinkin (Kurchatov Institute - Russia), Artemiy Kotov (Kurchatov Institute - Russia)

Our multi-layered text analysis system allows the robot F2 to interpret participants of the situation as protagonists or antagonists, and construct a list of possible speech and gesture reactions to be selected according to the simulated mood of the robot.
The first thing that is expected from robot assistants is a reaction to the user's remarks. No matter how the robot's creators position it (as a teacher, a simulator, or a home assistant), there must be some (possibly, rather primitive) dialog system inside it. Such systems usually do not use deep syntactic analysis and are often based on the "bag of words" model. The dialog system that we use for the F2 companion robot includes a step-by-step linguistic analysis of the incoming utterance consisting of three stages: morphological, syntactic, and semantic.

Pronounced or typed user’s utterance comes to the first step of analysis – the morphological one. At this stage of the analysis, all words included in the utterance are lemmatized and get morphological markup (part of speech, gender, time, etc.). At the second, syntactic stage of analysis, all participants in the situation described in the utterance are automatically distributed according to the roles they perform in this situation. We have a limited list of such roles, or valences, formed on the basis of the list proposed by Fillmore – agent, patient, instrument, source etc., and predicate – a specific valency for the verb. Syntactic analysis relies on a formal grammar of Russian stored in the SyntXML format and containing over 600 syntactic rules.

At the final, semantic stage of linguistic analysis, we move from the words of the utterance to their meaning, each valence is assigned a set of semantic features. For example, a statement "the robot hit the Professor" will turn into a semantic structure [agent: robot, egocentric, like an alien] [predicate: to touch, intensely, to harm, to manipulate, body action, like to hit] [patient: someone, profession, like docent]. After forming such semantic structures, the most appropriate script for this structure is selected from the list of robot reaction scripts, and the robot makes a response. Thus, the reactions to statements "the robot hit the Professor" and "the Professor hit the robot" will be different, since the actants "robot" and "Professor" in these statements fall into different valences, and therefore different scripts will work. In one case, the robot will feel aggression from the Professor, the DANGER script will be triggered, and, depending on the mood settings, the robot may begin to ask for mercy or respond with retaliatory aggression. In another situation, the WE DANGER script will be triggered, where the robot associates itself with the aggressor, after which, again depending on the settings, it can start apologizing or continue aggressive polemics.
Program