Downloads & Free Reading Options - Results

This article investigates the use of Transformation-Based Error-Driven learning for resolving part-of-speech ambiguity in the Greek language. The aim is not only to study the performance, but also to examine its dependence on different thematic domains. Results are presented here for two different test cases: a corpus on "management succession events" and a general-theme corpus. The two experiments show that the performance of this method does not depend on the thematic domain of the corpus, and its accuracy for the Greek language is around 95%.

We will conduct an artificial language learning study (e.g., Hudson Kam & Newport, 2005; Culbertson, 2012) to investigate the learnability of sound patterns in relation to their morphotactic position. The study is aimed at comparing how well participants learn coda consonant clusters that are produced exclusively in morphologically complex word forms (such as /gz/ in the plural form egg+s) to (i) those that are only found in morphologically simple word stems (such as /mp/ in lamp), and (ii) ambiguous ones that occur in both positions (such as /ks/ in fix or kick+s). Ultimately, we want to know whether the coda consonant sequences are learnt better when they occur exclusively in a certain position (such as only in word stem or in the plural) – and are therefore reliable signals of word structure – than when they are morphotactically ambivalent (Dressler & Dziubalska-Kołaczyk, 2006; Matzinger & Ritt, 2022). HOW: German-L1-speakers will be tasked with learning singular and plural nouns and adjectives from an artificial mini-language, in which some word forms end in consonant clusters. They will then be asked to reproduce the word forms from memory, and their output will be analysed. We manipulate the morphotactic position in which the coda consonant clusters occur in the alien language, as well as the specifics of the coda clusters (/-tk/ or /-fk/). Since some of these manipulations (viz., the ambiguous condition) require a more complex language than others, we have divided the study into two experiments: Experiment 1 will focus on the learnability of coda consonant clusters occurring exclusively in noun stems (e.g., SG / tmɪtk/ – PL / tmɪtk/+/ɪ/) vs. in plural forms (e.g., SG /tmɪt/ – PL /tmɪt/+/k/). Experiment 2 will examine the case of predictable vs. ambiguous consonant clusters. Participants will either learn a language where the same coda cluster can occur in both morphologically simple adjectives and complex noun plurals (e.g., SG /matk tmɪt/ – PL /matk tmɪt/+/k/), or an alien language where clusters indicate complexity unambiguously (e.g., SG /mat tmɪt/ – PL /mat tmɪt/+/k/). Our goal is to compare the learnability of the coda consonant clusters in the different morphotactic conditions. WHY: Speakers are sensitive to the frequency distributions of sounds and sound sequences in language, and rely on them in language use and first language acquisition (e.g. already Jusczyk, Luce & Charles-Luce, 1994; Ambridge, 2015). Extant research demonstrates that words with statistically probable sound shapes are recognized, learnt, and retrieved more quickly and accurately (Kelley & Tucker, 2017; Storkel, 2001; Goldrick & Larson, 2008). Further, the sound shapes of words can be indicative of the morphotactic structure of the word forms they express, albeit to varying degrees (cf., Dressler & Dziubalska-Kołaczyk, 2006). For instance, English monosyllables ending in /-mp/ (e.g., lamp, damp, lump, etc.) are always monomorphemic, while monosyllables ending in /-md/ (e.g., trimmed, blamed, seemed) are always morphologically complex, i.e., produced through past-tense suffixation. In both word form types, their shapes indicate their morphotactic structures reliably and unambiguously. Other word shapes, however, occur both in simple and in complex words and are therefore morphotactically ambiguous: for example, monosyllables word forms ending in /-nd/ can be both inflected (such as sinned, signed, banned, or rained) or simple (such as land, find, bond, or sound). We hypothesise that such ambiguity may be dispreferred and comparably difficult to learn because it impedes the processing and the acquisition of morphological regularities (Post et al., 2008). It follows that coda consonant clusters that predictably signal morphotactic structure should be learnt more successfully than those that occur in both morphologically simple and complex forms (= experiment 2). However, within the research field of morphonotactics, which focuses on the acquisition and diachronic development of consonant clusters in particular, the opposite has also been proposed. This area of research argues that consonant clusters are easier to learn and diachronically more stable when they span a morpheme boundary than when they occur in word stems (= experiment 1) because they noticeably and reliably indicate morphotactic complexity (Dressler & Dziubalska-Kołaczyk, 2006). This signalling function eases word processing and makes up for the difficulties involved in their perception and production. It has even been argued that clusters in morphotactically complex forms may be particularly costly to articulate and perceive, as this makes them more salient signals. The case of morphotactically ambiguous clusters has sparked debate: in agreement with the above line of argumentation, such clusters ought to be especially difficult to acquire and use, since this signalling function is compromised (Post et al., 2008; Calderone et al., 2014; Korecky-Kröll et al., 2014). But it has also been argued that frequent complex clusters might support the emergence and stability of simple counterparts in first-language acquisition via analogy (Hogg & McCully, 1987). In our study, we seek to shed more light on the learnability of ambiguous consonant clusters (= experiment 2, Morphotactic Manipulation), and thereby inform the discussion on the role of morphotactic ambiguity in language acquisition and change (e.g., Baumann, Prömer & Ritt, 2019; Matzinger & Ritt, 2022). In light of the morphonotactics literature, we further examine possible differences in the learnability of predictable clusters depending on their position within stems vs. across morpheme boundaries (= experiment 1, Morphotactic Manipulation). Here, we particularly want to explore possible differences in the learnability of a more costly cluster vs. a cluster that is comparatively easier to produce and perceive (= Phonotactic Manipulation). Our study is strongly informed by a master's thesis (Böhm, 2021), which simulated the cross-generational transmission of coda consonant clusters in an iterated learning experiment. Our hope is to find further confirmation of differential treatment of the different cluster types, and extend the insights gained on artificial language change to artificial language acquisition with a larger number of participants and a simpler set-up.

We will conduct an artificial language learning study (e.g., Hudson Kam & Newport, 2005; Culbertson, 2012) to investigate the learnability of sound patterns in relation to their morphotactic position. The study is aimed at comparing how well participants learn coda consonant clusters that are produced exclusively in morphologically complex word forms (such as /gz/ in the plural form egg+s) to (i) those that are only found in morphologically simple word stems (such as /mp/ in lamp), and (ii) ambiguous ones that occur in both positions (such as /ks/ in fix or kick+s). Ultimately, we want to know whether the coda consonant sequences are learnt better when they occur exclusively in a certain position (such as only in word stem or in the plural) – and are therefore reliable signals of word structure – than when they are morphotactically ambivalent (Dressler & Dziubalska-Kołaczyk, 2006; Matzinger & Ritt, 2022). HOW: German-L1-speakers will be tasked with learning singular and plural nouns and adjectives from an artificial mini-language, in which some word forms end in consonant clusters. They will then be asked to reproduce the word forms from memory, and their output will be analysed. We manipulate the morphotactic position in which the coda consonant clusters occur in the alien language, as well as the specifics of the coda clusters (/-tk/ or /-fk/). Since some of these manipulations (viz., the ambiguous condition) require a more complex language than others, we have divided the study into two experiments: Experiment 1 will focus on the learnability of coda consonant clusters occurring exclusively in noun stems (e.g., SG / tmɪtk/ – PL / tmɪtk/+/ɪ/) vs. in plural forms (e.g., SG /tmɪt/ – PL /tmɪt/+/k/). Experiment 2 will examine the case of predictable vs. ambiguous consonant clusters. Participants will either learn a language where the same coda cluster can occur in both morphologically simple adjectives and complex noun plurals (e.g., SG /matk tmɪt/ – PL /matk tmɪt/+/k/), or an alien language where clusters indicate complexity unambiguously (e.g., SG /mat tmɪt/ – PL /mat tmɪt/+/k/). Our goal is to compare the learnability of the coda consonant clusters in the different morphotactic conditions. WHY: Speakers are sensitive to the frequency distributions of sounds and sound sequences in language, and rely on them in language use and first language acquisition (e.g. already Jusczyk, Luce & Charles-Luce, 1994; Ambridge, 2015). Extant research demonstrates that words with statistically probable sound shapes are recognized, learnt, and retrieved more quickly and accurately (Kelley & Tucker, 2017; Storkel, 2001; Goldrick & Larson, 2008). Further, the sound shapes of words can be indicative of the morphotactic structure of the word forms they express, albeit to varying degrees (cf., Dressler & Dziubalska-Kołaczyk, 2006). For instance, English monosyllables ending in /-mp/ (e.g., lamp, damp, lump, etc.) are always monomorphemic, while monosyllables ending in /-md/ (e.g., trimmed, blamed, seemed) are always morphologically complex, i.e., produced through past-tense suffixation. In both word form types, their shapes indicate their morphotactic structures reliably and unambiguously. Other word shapes, however, occur both in simple and in complex words and are therefore morphotactically ambiguous: for example, monosyllables word forms ending in /-nd/ can be both inflected (such as sinned, signed, banned, or rained) or simple (such as land, find, bond, or sound). We hypothesise that such ambiguity may be dispreferred and comparably difficult to learn because it impedes the processing and the acquisition of morphological regularities (Post et al., 2008). It follows that coda consonant clusters that predictably signal morphotactic structure should be learnt more successfully than those that occur in both morphologically simple and complex forms (= experiment 2). However, within the research field of morphonotactics, which focuses on the acquisition and diachronic development of consonant clusters in particular, the opposite has also been proposed. This area of research argues that consonant clusters are easier to learn and diachronically more stable when they span a morpheme boundary than when they occur in word stems (= experiment 1) because they noticeably and reliably indicate morphotactic complexity (Dressler & Dziubalska-Kołaczyk, 2006). This signalling function eases word processing and makes up for the difficulties involved in their perception and production. It has even been argued that clusters in morphotactically complex forms may be particularly costly to articulate and perceive, as this makes them more salient signals. The case of morphotactically ambiguous clusters has sparked debate: in agreement with the above line of argumentation, such clusters ought to be especially difficult to acquire and use, since this signalling function is compromised (Post et al., 2008; Calderone et al., 2014; Korecky-Kröll et al., 2014). But it has also been argued that frequent complex clusters might support the emergence and stability of simple counterparts in first-language acquisition via analogy (Hogg & McCully, 1987). In our study, we seek to shed more light on the learnability of ambiguous consonant clusters (= experiment 2, Morphotactic Manipulation), and thereby inform the discussion on the role of morphotactic ambiguity in language acquisition and change (e.g., Baumann, Prömer & Ritt, 2019; Matzinger & Ritt, 2022). In light of the morphonotactics literature, we further examine possible differences in the learnability of predictable clusters depending on their position within stems vs. across morpheme boundaries (= experiment 1, Morphotactic Manipulation). Here, we particularly want to explore possible differences in the learnability of a more costly cluster vs. a cluster that is comparatively easier to produce and perceive (= Phonotactic Manipulation). Our study is strongly informed by a master's thesis (Böhm, 2021), which simulated the cross-generational transmission of coda consonant clusters in an iterated learning experiment. Our hope is to find further confirmation of differential treatment of the different cluster types, and extend the insights gained on artificial language change to artificial language acquisition with a larger number of participants and a simpler set-up.