Mirjam Ernestus
Centre for Language Studies |
Main goalAs a psycholinguist, I am interested in all aspects of language processing. My main focus is on the immense pronunciation variation that is typical of casual speech: the types of variation, how and why speakers produce this variation, and how native and non-native listeners cope with this variation. In addition, I focus on how speakers and listeners process morphologically complex words (words which can be broken up into smaller meaningful units, e.g., the word meaningful can be broken up in mean, ing and ful). My final focus, which is based on the other two, is on specific theories of speech comprehension. Pronunciation variation in casual speechInformal (casual) conversations differ from formal speech in many respects. An important difference is that, in casual speech, words are often produced with fewer segments and even with fewer syllables. Here are a few examples of reduced pronunciation variants of American English and Dutch words.
When I started my studies on reduced words, everyday speech was hardly studied, and no one had studied reduced words as a characteristic of everyday speech before. Over the years, I have documented reduced pronunciation variants of words and of sound sequences, mostly in Dutch, French, Spanish, and Czech. I have provided both detailed qualitative as well as quantitative descriptions, based on large corpora of spontaneous speech, most of which I compiled myself. I have established the roles of many variables that favor reduction, providing insights in the articulatory and cognitive mechanisms underlying speech production. While many reduction patterns can simply be explained as dramatic effects of co-articulation, which may be more pronounced for words that speakers have uttered more often, other variables favoring reduction provide insights into the higher levels of the production system. For instance, the role of the likelihood of the following word shows how speakers may ensure that the speed with which they produce words is aligned with how quickly they can (articulatory) plan the next words, and the role of the position of stress in the following word shows that speakers consider the prosodic properties of the next word before planning the detailed articulation of the present word. Furthermore, reduction patterns are language specific. For instance, while Spanish mostly shows consonant reduction, in French, vowel reduction is much more common. Healthy native listeners seem to easily process reduced word pronunciation variants. Over the years, I have documented informational cues that native listeners use to recognize reduced words. We have found that listeners heavily rely on the sound properties and meanings of the preceding words, which help them to guess what the reduced words may be. Nevertheless, listeners also need to hear the reduced words themselves to identify them, even when these words are extremely reduced. This shows that listeners extract indispensable valuable information from very subtle acoustic cues. Another important informational cue is the frequency of occurrence of the reduction patterns: listeners more easily recognize reduced pronunciation variants that are more frequent, even when these high frequency variants are more reduced than low frequency variants. Notwithstanding all these cues, we have found that listeners process the meanings of reduced words somewhat less easily than those of full words. In contrast to native listeners, non-native listeners have great difficulties understanding reduced words. This holds for both second language and foreign language learners. Learners benefit less from the sound properties and meanings of the preceding words. Moreover, they have difficulties interpreting subtle acoustic cues to reduced sounds, especially when their native languages do not display the same sound and reduction patterns. Like native listeners, learners are better in recognizing reduced words that they have heard before more often.Morphologically complex wordsMy work on morphologically complex words has focused on two research questions. One central question is how speakers determine which affix to attach to a given stem (e.g., should the plural of Dutch aardappel be aardappel+en or aardappel+s?) and how these factors can make speakers choose affixes that are incorrect according to the standard language rules (e.g., Dutch dub+te instead of standard dub+de). Addressing this question, I was the first to show how paradigmatic effects result in synchronic variation in the choice of affixes for real words and in their durations. We showed how speakers’ choices are determined by the affixes that are taken by phonologically similar words (e.g., klopte, dopte, lapte, etc. supporting dubte), and how the influences of these similar words are constrained by the frequencies of the word itself and of the words with the same stem. My second central question is how morphologically complex words are processed by readers and listeners. We have shown that the paradigmatic effects that play a role in speakers’ choices for a given affix also play a role in reading: they read complex words more quickly -- also those that violate the standard language rules -- the more their affixes are supported by the phonologically similar words. In addition, we have shown that, also during listening, the processing of a word is affected by the properties of all words with the same stem. Taken together, these results show that, both during language production and comprehension, language users not only consider the word that has to be produced or understood, but also the phonologically and morphologically related words.Theories of spoken word comprehensionOver the years, we have directly investigated the plausibility of one important theory of spoken word recognition and developed another one ourselves (in addition to testing LDL based models). We have conducted many behavioral experiments studying the role of examplars (i.e. the memory traces of word tokens) in spoken word recognition. We have shown that exemplar effects are only visible in simple and highly controlled experiments. Once experiments start approaching normal listening situations (e.g. by presenting word tokens that differ in two instead of one acoustic characteristic), exemplar effects disappear. This shows that examplars do not play an important role in everyday speech comprehension. Together with Louis ten Bosch and Lou Boves, I have developed Diana. This computational model of human spoken word recognition takes the acoustic signal as its input and produces as its output 1) the word that it recognized, 2) whether this is a real word , and 3) how long it takes to recognize the word. We have tested Diana by comparing its outputs with human listeners’ behavior in several experiments, in Dutch and English, and showed that Diana behaves like an average human participant. Diana is based on assumptions that better conform with what is known about auditory processing than other models of spoken word recognition. Nevertheless, also Diana could still be improved in several respects. |