| 1. |
|
|
| 2. |
- Bernardy, Jean-Philippe, 1978, et al.
(författare)
-
Modelling prosodic structure using Artificial Neural Networks
- 2017
-
Ingår i: ExLing 2017. Proceedings of 8 th Tutorial and Research Workshop on Experimental Linguistics, 19-22 June 2017, Heraklion, Crete, Greece / edited by Antonis Botinis. - Athens : University of Athens. - 2529-1092. - 9789604661626
-
Konferensbidrag (refereegranskat)abstract
- The ability to accurately perceive whether a speaker is asking a question or is making a statement is crucial for any successful interaction. However, learning and classifying tonal patterns has been a challenging task for automatic speech recognition and for models of tonal representation, as tonal contours are characterized by significant variation. This paper provides a classification model of Cypriot Greek questions and statements. We evaluate two state-of-the-art network architectures: a Long Short-Term Memory (LSTM) network and a convolutional network (ConvNet). The ConvNet outperforms the LSTM in the classification task and exhibited an excellent performance with 95% classification accuracy.
|
|
| 3. |
- den Ouden, Dirk B., et al.
(författare)
-
Comparison of Automated Methods for Vowel Segmentation and Extraction of Acoustic Variables
- 2018
-
Ingår i: Clinical Aphasiology Conference, CAC 2018, Austin, Texas USA..
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- Introduction: Primary Progressive Aphasia (PPA) is a neurodegenerative syndrome in which linguistic abilities become gradually impaired. There are three primary variants of PPA: the non-fluent agrammatic PPA, the fluent type semantic PPA, and the logopenic PPA, which is also considered an atypical form of Alzheimer’s disease (Mesulam et al., 1982; Gorno-Tempini et al., 2011). Along with the three main variants, a fourth variant has been proposed, a non-fluent apraxia of speech (AOS), though this is currently the subject of an open debate (e.g., Duffy et al., 2017; Henry et al., 2013). According to sophisticated criteria established a few years ago, PPA subtyping for a given patient presented in clinic requires clinical, neuropsychological, and imaging information (Gorno-Tempini et al., 2011). Nevertheless, quantifying the decline of linguistic abilities and subtyping the variants of PPA manually is both hard and laborious, so there is great demand for algorithms that subtype a given patient automatically. Picture description samples of connected speech and random forests techniques have been used for this purpose (de Aguiar et al., 2017; Wilson et al., 2010, Fraser et al. 2013, 2014). In the present study, we compared existing models and we propose a new one. Aims: In this study, we provide an automated classification model of PPA variants trained on known morphological and acoustic predictors and on predictors related to the clinical and linguistic profile of individuals with PPA (e.g., Mack et al., 2015; Gorno-Tempini et al., 2011; Wilson et al., 2010). Method: Speech materials for this study come from the Transcranial Direct Current Stimulation for Primary Progressive Aphasia study at Johns Hopkins University. Twenty-six individuals with PPA (Mean(SD) age = 68.6 (7.8) years, Mean(SD) education = 16.1 (2.9) years) participated in this study. PPA participants were diagnosed based on the established consensus criteria (Gorno-Tempini et al., 2011), i.e., imaging, clinical, and neuropsychological examination by trained neurologists. Individuals with PPA included non-fluent with AOS (N=5), non fluent without AOS (N=7), logopenic (N=8), and semantic (N=6) variants. Recordings of the Cookie Theft picture description from the Boston Diagnostic Aphasia Examination (BDAE) were computationally analyzed. All speech productions were automatically transcribed and segmented using an end-to-end speech-to-transcription platform. From the speech signals, we measured morphological and acoustic predictors, including vowel formants F1 ... F3, measured at 15%, 50%, and 75% of vowel’s duration, vowel duration, fundamental frequency, and pause duration. The analysis and the statistics were conducted using Python and R programming languages (R Core Team, 2017; Rossum, 1995). Three different machine learning algorithms: C5.0 decision trees, Classification and Regression Trees (CART) and random forests were trained on the predictors (Breiman, 2001; Quinlan, 1993; Hastie et al., 2009). All models were trained on the 80% of the speakers (training set), with 3-fold cross-validation. All predictor variables were centered and scaled. C5.0 was trained with winnowing and without winnowing. (Winnowing facilitates the automatic pre-selection of the predictors that are used in the decision tree.) After the training we evaluated the trained models on the unknown dataset, namely the 20% of the speakers (evaluation set). Results: C5.0 provided 86% (95% CI[81, 88], kappa = 0.76) and Random Forests 85% (95% CI[81, 88], kappa = 0.76) classification accuracy on the test data; CART provided the lowest overall classification accuracy. Overall, C5.0 outperformed both the random forests and CART, with high classification accuracy on unknown data. Non-fluent PPA with AOS was correctly predicted by both C5.0 and random forests. Discussion: The C5.0 classification model provides support for the known predictors employed in the literature. Also, it provides some objective ways to distinguish the presence of AOS in PPA and corroborate research on classification of AOS using acoustic properties especially those related to vowel production (Den Ouden et al. 2017). However, given the low number of participants employed in this study, further research is required, with a larger number of participants. Nevertheless, the proposed methods employed here constitute a promising step towards a computational differential diagnostic tool of PPA that is easy to use, quick and accurate.
|
|
| 4. |
- Fraser, Kathleen, 1984, et al.
(författare)
-
Improving the Sensitivity and Specificity of MCI Screening with Linguistic Information.
- 2018
-
Ingår i: Proceedings of the LREC workshop: Resources and ProcessIng of linguistic, para-linguistic and extra-linguistic Data from people with various forms of cognitive/psychiatric impairments (RaPID-2). 8th of May 2018, Miyazaki, Japan / Dimitrios Kokkinakis (ed.). - 9791095546269
-
Konferensbidrag (refereegranskat)abstract
- The Mini-Mental State Exam (MMSE) is a screening tool for cognitive impairment. It has been extensively validated and is widely used, but has been criticized as not being effective in detecting mild cognitive impairment (MCI). In this study, we examine the utility of augmenting MMSE scores with automatically extracted linguistic information from a narrative speech task to better differentiate between individuals with MCI and healthy controls in a Swedish population. We find that with the addition of just four linguistic features, the F score (measuring a trade-off between sensitivity and specificity) is improved from 0.67 to 0.81 in logistic regression classification. These preliminary results suggest that the accuracy of traditional screening tools may be improved through the addition of computerized language analysis.
|
|
| 5. |
- Fyndanis, Valantis, et al.
(författare)
-
Are there prototypical associations between time frames and aspectual values? Evidence from Greek aphasia and healthy ageing
- 2019
-
Ingår i: Clinical Linguistics & Phonetics. - : Informa UK Limited. - 0269-9206 .- 1464-5076. ; 33:1-2, s. 191-217
-
Tidskriftsartikel (refereegranskat)abstract
- Time reference, which has been found to be selectively impaired in agrammatic aphasia, is often interwoven with grammatical aspect. A recent study on Russian aphasia found that time reference and aspect interact: Past reference was less impaired when tested within a perfective aspect context (compared to when tested within an imperfective aspect context), and reference to the non-past was less impaired when tested within an imperfective aspect context (compared to when tested within a perfective aspect context). To explain this pattern, the authors argued that there are prototypical associations between time frames and aspectual values. The present study explores the relationship between time reference and aspect focusing on Greek aphasia and healthy ageing and using a sentence completion task that crosses time reference and aspect. The findings do not support prototypical matches between different time frames and aspectual values. Building on relevant studies, we propose that patterns of performance of healthy or language-impaired speakers on constrained tasks tapping different combinations of time frames with aspectual values should reflect the relative frequency of these combinations in a given language. The analysis of the results at the individual level revealed a double dissociation, which indicates that a given time frame–aspectual value combination may be relatively easy to process for some persons with aphasia but demanding for some others.
|
|
| 6. |
- Fyndanis, Valantis, et al.
(författare)
-
Morphosyntactic production in agrammatic aphasia: A cross-linguistic machine learning approach.
- 2018
-
Ingår i: Frontiers in Human Neuroscience. Academy of Aphasia 56th Annual Meeting, Montreal, Canada, 21 Oct - 23 Oct, 2018.. - : Frontiers Media SA. - 1662-5161 .- 1662-5161.
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- Introduction Recent studies on agrammatic aphasia by Fyndanis et al. (2012, 2017) reported evidence against the cross-linguistic validity of unitary accounts of agrammatic morphosyntactic impairment, such as the Distributed Morphology Hypothesis (DMH) (Wang et al., 2014), the two versions of the Interpretable Features’ Impairment Hypothesis (IFIH-1: Fyndanis et al., 2012; IFIH-2: Fyndanis et al., 2018b), and the Tree Pruning Hypothesis (TPH) (Friedmann & Grodzinsky, 1997). However, some of the features/factors emphasized by the accounts above (i.e. involvement of inflectional alternations (DMH), involvement of integration processes (IFIH-1), involvement of both integration processes and inflectional alternations (IFIH-2), position of a morphosyntactic feature/category in the syntactic hierarchy (TPH)) may still play a role in agrammatic morphosyntactic production. These features may act in synergy with other factors in determining the way in which morphosyntactic production is impaired across persons with agrammatic aphasia (PWA) and across languages. Relevant factors may include language-independent and language-specific properties of morphosyntactic categories, as well as subject-specific and task/material-specific variables. The present study addresses which factors determine verb-related morphosyntactic production in PWA and what is their relative importance. Methods We collapsed the datasets of the 24 Greek-, German-, and Italian-speaking PWA underlying Fyndanis et al.’s (2017) study, added the data of two more Greek-speaking PWA, and employed machine learning algorithms to analyze the data. The unified dataset consisted of data on subject-verb agreement, time reference (past reference, future reference), grammatical mood (indicative, subjunctive), and polarity (affirmatives, negatives). All items/conditions were represented as clusters of theoretically motivated features: ±involvement of integration processes, ±involvement of inflectional alternations, ±involvement of both integration processes and inflectional alternations, and low/middle/high position in the syntactic hierarchy. We included 14 subject-specific, category-specific and task/material-specific predictors: Verbal Working Memory (WM), (years of formal) Education, Age, Gender, Mean Length of Utterance in (semi)spontaneous speech (Index 1 of severity of agrammatism), Proportion of Grammatical Sentences in (semi)spontaneous speech (Index 2 of severity of agrammatism), Words per Minute in (semi)spontaneous speech (Index of fluency), Involvement of inflectional alternations, Involvement of integration processes, Involvement of both integration processes and inflectional alternations, Position of a given morphosyntactic category in the syntactic hierarchy (high, middle, low), Item Presentation mode (cross-modal, auditory), Response mode (oral, written), and Language (Greek, German, Italian). Different machine learning models were employed: Random Forest, C5.0 decision tree, RPart, and Support Vector Machine. Results & Discussion Random Forest model outperformed all the other models achieving the highest accuracy (0.786). As shown in Figure 1, the best predictors of accuracy on tasks tapping morphosyntactic production were the involvement of both integration processes and inflectional alternations (categories involving both integration processes and inflectional alternations were more impaired than categories involving one or neither of them), verbal WM capacity (the greater the WM capacity, the better the morphosyntactic production), and severity of agrammatism (the more severe the agrammatism, the worse the morphosyntactic production). Results are consistent with IFIH-2 (Fyndanis et al., 2018b) and studies highlighting the role of verbal WM in morphosyntactic production (e.g., Fyndanis et al., 2018a; Kok et al., 2007).
|
|
| 7. |
- Fyndanis, Valantis, et al.
(författare)
-
Time reference and aspect in agrammatic aphasia: Evidence from Greek
- 2017
-
Ingår i: Front. Hum. Neurosci. Conference Abstract: Academy of Aphasia, 55th Annual Meeting, Baltimore, United States, 5 Nov - 7 Nov, 2017.. - : Frontiers Media SA. - 1662-5161.
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- Time reference, which has been found to be selectively impaired in agrammatic aphasia (e.g., Bastiaanse et al., 2011), is often interwoven with grammatical aspect. Dragoy and Bastiaanse (2013) investigated the relationship between time reference/tense and aspect focusing on Russian aphasia and found that the two interact: past reference was less impaired when tested within perfective aspect (compared to when tested within imperfective aspect), and reference to the nonpast was less impaired when tested within imperfective aspect (compared to when tested within perfective aspect). To account for this pattern, Dragoy and Bastiaanse (2013: 114) claimed that “perfectives primarily refer to completed, past events while imperfectives prototypically describe ongoing, non-past events”. This study explores the relationship between time reference and aspect focusing on Greek aphasia. In Greek, verb forms referring to the past and future encode the perfective-imperfective contrast. Dragoy and Bastiaanse (2013) would make predictions PR1–PR4 for Greek. (PR1) past reference within perfective aspect > past reference within imperfective aspect; (PR2) future reference within perfective aspect < future reference within imperfective aspect; (PR3) perfective aspect within past reference > imperfective aspect within past reference; (PR4) perfective aspect within future reference < imperfective aspect within future reference. Methods Eight Greek-speaking persons with agrammatic aphasia (PWA) and eight controls were administered a sentence completion task consisting of 128 experimental source sentence (SS)-target sentence (TS) pairs. There were eight subconditions, each of which consisted of 16 items: past reference within perfective aspect; past reference within imperfective aspect; future reference within perfective aspect; future reference within imperfective aspect; perfective aspect within past reference; imperfective aspect within past reference; perfective aspect within future reference; imperfective aspect within future reference. Participants were auditorily presented with a SS and the beginning of the TS, and were asked to orally complete the TS producing the missing Verb Phrase. We fitted generalized linear mixed-effect models and employed Fisher’s exact tests to make within-participant comparisons. Results Overall, the aphasic group fared significantly worse than the control group (p < 0.001). At the group level, none of the four relevant comparisons (see PR1–PR4) yielded significant differences for PWA (Table 1). Four PWA (P1, P3, P7, P8) exhibited dissociations, with three of them making up a double dissociation: P1 performed better on imperfective aspect-future reference than on perfective aspect-future reference (p < 0.001), and P7 and P8 exhibited the opposite pattern (p = 0.016 and p < 0.001 for P7 and P8, respectively). Discussion Results are not consistent with Dragoy and Bastiaanse’s (2013) findings, which challenges the idea of prototypical and non-prototypical associations between time reference and aspect. The double dissociation that emerged in the aspect condition indicates that a given time reference-aspect combination may be relatively easy to process for some PWA but demanding for some others. Thus, studies investigating tense/time reference in aphasia should ensure that this grammatical/semantic category is not confounded by aspect.
|
|
| 8. |
- Grohmann, Kleanthes, et al.
(författare)
-
Acquiring Clitic Placement in Bilectal Settings: Interactions between Social Factors
- 2017
-
Ingår i: Frontiers in Communication. - : Frontiers Media SA. - 1050-9046 .- 2297-900X. ; 2
-
Tidskriftsartikel (refereegranskat)abstract
- This paper examines the development of object clitic placement by children acquiring Cypriot Greek. Greek-speaking Cyprus is sociolinguistically characterized by diglossia between two varieties of Greek, the local Cypriot Greek and the official Standard Modern Greek. Arguably as a result of this situation, clitics may be placed post- (enclisis) or preverbally (proclisis) in the same syntactic environment; while the former is a property of Cypriot Greek, the latter is typically considered an effect of the standard language. The following issues are investigated here: (a) how such bilectal speakers distinguish between the two Greek varieties with respect to clitic placement; (b) how the acquisition of clitics develops over time; (c) how, and which, sociolinguistic factors determine clitic placement; and (d) how schooling may affect clitic placement. To address (a)–(d), a sentence completion task was used to elicit clitic productions, administered to 431 children around Cyprus ranging from 2;8 to 8;11. The C5.0 machine-learning algorithm was employed to model the interaction of (socio-)linguistic factors on the development of clitic placement. The model shows that speakers acquire the relevant features very early, yet compartmentalization of form and function according to style emerges only as they engage in the larger speech community. In addition, the effects of sociolinguistic factors on clitic placement appear gradually.
|
|
| 9. |
- Kokkinakis, Dimitrios, 1965, et al.
(författare)
-
A Multifaceted Corpus for the Study of Cognitive Decline in a Swedish Population
- 2019
-
Ingår i: CLARe4 : Corpora for Language and Aging Research, 27 February – 1 March 2019, Helsinki, Finland.
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- A potential, early-stage diagnostic marker for neurodegenerative diseases, such as Alzheimer’s disease, is the onset of language disturbances which is often characterized by subtle word-finding difficulties, impaired spontaneous speech, slight speech hesitancy, object naming difficulties and phonemic errors. Connected speech provides valuable information in a non-invasive and easy-to-assess way for determining aspects of the severity of language impairment. Data elicitation is an established method of obtaining highly constrained samples of connected speech that allows us to study the intricate interactions between various linguistic levels and cognition. In the paper, we describe the collection and content of a corpus consisting of spontaneous Swedish speech from individuals with Mild Cognitive Impairment (MCI), with Subjective Cognitive Impairment SCI) and healthy, age-matched controls (HC). The subjects were pooled across homogeneous subgroups for age and education, a sub-cohort from the Gothenburg-MCI study. The corpus consists of high quality audio recordings (including transcriptions) of several tasks, namely: (i)a picture description task – the Cookie-theft picture, an ecologically valid approximation to spontaneous discourse that has been widely used to elicitate speech from speakers with different types of language and communication disorders; (ii)a read aloud task (including registration of eye movements) – where participants read a text from the IREST collection twice, both on a computer screen (while eye movements are registered), and the same text on paper; (iii)a complex planning task – a subset of executive functioning that tests the ability to identify, organize and carry out (complex) steps and elements that are required to achieve a goal; (iv)a map task – a spontaneous speech production/semi-structured conversation in which the participants are encouraged to talk about a predefined, cooperative task-oriented topic; (v)a semantic verbal fluency task – category animals: where participants have to produce as many words as possible from a category in a given time (60 seconds). The fluency tests require an elaborate retrieval of words from conceptual (semantic) and lexical (phonetic) memory involving specific areas of the brain in a restricted timeframe. All samples are produced by Swedish speakers after obtaining written consent approved by the local ethics committee. Tasks (i) and (ii) have been collected twice in a diachronically apart period of 18 months between 2016 and 2018. The corpus represents an approximation to speech in a natural setting: The material for elicitation is controlled in the sense that the speakers are given specific tasks to talk about, and they do so in front of a microphone. The corpus may serve as a basis for many linguistic and/or speech technological investigations and has being already used for various investigations of language features.
|
|
| 10. |
- Kokkinakis, Dimitrios, 1965, et al.
(författare)
-
Kan textforskning bidra till tidigare och säkrare demensdiagnostik?
- 2018
-
Ingår i: Forum för textforskning 13 , Lund 7 – 8 juni 2018.
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- Tidigare forskning har visat att subtila språkstörningar kan finnas vid de tidigaste förstadierna till demens, flera år innan en klinisk diagnos kan ställas. Inom ramen för projektet ”Språkliga och extra-lingvistiska parametrar för tidig upptäckt av kognitiv svikt” (finansierat av Riksbankens Jubileumsutlysning, 2016-19) undersöker vi med hjälp av språkteknologi och språkanalysstudier hur dessa språkstörningar yttrar sig. Kan språkteknologi användas för att upptäcka dessa tidiga språkrelaterade symtom och därmed bidra med nyanserad, komplementär och användbar kunskap? Kan användning av språkteknologi särskilja personer med de allra tidigaste kognitiva avvikelserna från personer med mer godartad, åldersrelaterad kognitiv svikt? Vilka språkliga förmågor drabbas? Hur yttrar sig dessa förändringar och vilka slags empiriska material finns att tillgå? Dessa är några av de frågor vi söker svar på. Vi gör inspelningar som vi analyserar för att kunna ta fram ny kunskap om subtila språkliga kännetecken som kan föregå demensutveckling. Denna kunskap kan användas för att eventuellt kunna förutsäga vilka individer som befinner sig i riskzonen för att utveckla demens, och kan vara användbar som komplementerande beslutsunderlag till domänexperter. Vi utvinner, analyserar och undersöker om det finns samband mellan olika språkrelaterade parametrar från spontan talinteraktion, transkriptioner men även ögonrörelser och neuropsykologiska tester från personer med subjektiv eller lindrig kognitiv nedsättning och friska kontrollpersoner. Många gånger är det svårt att avgöra huruvida lindriga kognitiva symtom är en del av det normala åldrandet eller början på en neurodegenerativ process. Vi förväntar oss inte heller att varje enskild person med kognitiv nedsättning kommer att uttrycka sig eller läsa på samma sätt utan snarare att dessa personer tidigt i sjukdomsförloppet kommer att börja uppvisa olika slags avvikande läsmönster, eller göra fonologiska, lexikala, syntaktiska eller semantiska fel. I studien utvecklar vi verktyg för att automatiskt hitta dessa avvikelser, och målet är att detta sedan ska kunna användas som komplement till tidig diagnostik samt som prognostiskt eller screeningverktyg. Deltagarna i vår studie har rekryterats från en pågående longitudinell studie, ”Demens i Tidigt Skede”, (eng. ”The Gothenburg MCI study”) på Minnesmottagningen i Göteborg, och vårt projekt har godkänts av den lokala etiknämnden. Alla deltagare i studien (kontrollgruppen [HC], personer med subjektiv kognitiv nedsättning [SCI] och personer med mild kognitiv nedsättning [MCI]) har genomgått baslinjeundersökning och gett informerat skriftligt samtycke (demografisk information finns i tabell 1). Vårt projekt är f.n. pågående och vi kommer presentera resultat baserade på inspelningstillfälle nr ett (aug. 2016-mars 2017). En ny inspelningsomgång, med samma deltagare, började i februari 2018 och förväntas vara avslutat i december 2018. Under presentationen kommer vi ge exempel på olika tal-, text- och ögonrörelseanalyser vi har genomfört och diskutera metodval och resultat baserade på studiens första fas. Vi kommer vidare ge en kort inblick i den nya, pågående inspelningsomgången och de nya testmoment vi använder. Vi vill med vårt arbete visa hur språkteknologisk analys kan bidra till att utöka vår kunskap inom området så att den kan vara användbar för tidig diagnostik och optimal omvårdnad. Enligt Socialstyrelsen (2017) finns det i Sverige över 160 000 personer med någon demenssjukdom. Våra resultat kan ha en betydelse för vårdpersonal som snabbare vill diagnostisera och identifiera individer med olika former av kognitiv funktionsnedsättning innan allvarliga symtom blir påtagliga. Utvecklingsmöjligheterna är många: nya eller förbättrade kognitiva screeningtester som skulle kunna användas inom primär- och specialistvården, samt utveckling och tillämpning av insatser som kan påverka beteendemönster och träna upp individens kommunikativa förmåga, kan på sikt leda till positiva konsekvenser som minskade vårdköer samt effektivare behandling avseende kostnader och behandlingsutfall.
|
|
