TABLE OF CONTENTS
2 LITERATURE REVIEW
2.1 WHAT IS HOMONYMY AND POLYSEMY?
2.2 PREVIOUS STUDIES ON LEXICAL AMBIGUITY
3.3 DATA COLLECTION PROCEDURE
4 RESULTS AND DISCUSSION
4.1 RESULTS EVALUATION
4.2 COMPARISON AND CRITICAL DISCUSSION OF THE RESULTS
4.3 A MODEL FOR THE NATURE OF THE MENTAL LEXICON
APPENDIX A TABLE OF TARGETS
APPENDIX B COMPARISON IN A BAR CHART
APPENDIX C COMPARISON IN A TABLE
APPENDIX D RESPONSE TIMES COLLECTED
Decoding how people store and process words with multiple meanings has become a widely probed domain in psycholinguistics. Since earlier linguistic studies agreed on the point that ambiguous words produced a processing advantage none of them asked if this advantage differs between homonymy and polysemy. Are semantically unrelated words, which have (due to the same etymology) the same notation, stored and processed like words that have one meaning but many senses? The storage at least differs in a dictionary. In a dictionary the homonymous word bank, for example, provides two different entries - once as a bank which can mean river bank and again an entry for bank as an institute. Polysemy in contrast refers to relatedness in meaning and thereby only holds one entry. A sample would be the term tongue which once can mean the physical organ or tongue as the dialect one might speak. Thereby, two questions arise. When homonymous meanings are assumed to have separate representations in the human mind does this count for polysemous senses and does the processing advantage differ between homonymous and polysemous words?
Since this is a very new perspective in psycholinguistics, not many results have been achieved by now. Thus, this study aims to go further and explain ‘the source of the processing advantage’ which could have been observed in previous lexical decision studies with ambiguous words (Klepousniotou & Baum 2007: 8). In addition this study will focus on the diverse processing advantages for homonyms and polysemes and attempts to provide a model for the word representation of homonymous and polysemous words. Accordingly, the experiment is constituted as a lexical decision task. The used corpus was adopted from the study of Rodd et al. (2002: 263-264) which built the base for this study. This corpus was used because the present study only included 20 participants which thus only allows a predication to a limited extent. But by choosing this corpus parallels can be drawn between those two experiments and thereby a more general and meaningful statement can be provided.
Before approaching the present study it is important to give a brief overview of previous studies and their findings. Therefore the main terms and works, which are relevant for the present study, will be clarified and described in the next section.
2 Literature review
2.1 What is homonymy and polysemy?
As mentioned earlier, this study tries to present an alternative view on ambiguity. In particular, the aim will be to discover a processing difference between homonymous and polysemous words. Due to the differentiation between homonyms and polysemes it is appropriate to clarify this distinction in order to facilitate comprehension. Peter H. Matthews (2014: 178) a wellknown British linguist defines homonymy as ‘The relation between words whose forms are the same but whose meanings are different and cannot be connected: e.g. between pen ‘writing instrument’ and pen ‘enclosure’.’ In contrast to this concept Matthews (2014: 308) defines polysemy as ‘The property of a single word which has two or more distinct but related senses. Thus the noun screen is polysemous [pɒlɪˈsiːməs], since it is used variously of a fire screen, a cinema screen, a television screen, and so on.’ The fundamental distinction is hence that homonymous senses are entirely separated and thereby often seen as independent words whereas polysemes are treated as being one word with similar senses. However, this distinction is not always solid because in some cases a clear determination is not possible. Since this distinction is not naturally given, it is necessary to keep in mind that semantic ambiguity can shift between those two concepts and that our mental representations of homonymy and polysemy might be very different (Rodd et al. 2002: 246).
2.2 Previous studies on lexical ambiguity
In the past 40 years, many researchers have published journals about semantic ambiguity which have tried to investigate ambiguity effects in the mental lexicon. Three of the first scientists who occupied themselves with mental ambiguity were Rubenstein et al. (1970: 490). On the basis of a simple lexical decision task they found out that the response time (RT) for homographs is faster than for non-homographs. This effect is nowadays also known as the ambiguity advantage, which is the technical term for this finding. It simply means that the greater the meanings of a word the faster the response time (Klepousniotou & Baum 2007: 2). A further influential study was published by Azuma and Van Orden (1997: 492) who investigated, for the first time, the NOM effect which is concerned about the number of word meanings. They could show that the response time for words with multiple NOMs was slower to low-related words than to highrelated words. Almost 11 years later, James E. Jastrzembski (1981: 280) improved Rubenstein et al.’s experiment. He was able to show that a more meaningful result is reflected if one compared words with high number of meanings and low number of meanings. Thereby, he took a new effect into consideration, not just word frequency but also the number of meanings. He was able to demonstrate a processing advantage for ambiguous words with high number of meanings. In addition, James E. Jastrzembski (1981: 301) hypothesised that ambiguous words have in contrast to unambiguous words separated representations in the mental lexicon. This view was based on the basic assumption that one of the multiple entries of an ambiguous word will be located faster than the entry of the unambiguous word. This assumption was later on confirmed by several studies. Hino and Lupker (1996: 1351), for example, interpreted this finding as a result of semantic activation. They were persuaded that all representations are activated together and thus produce a processing advantage.
Nevertheless, this ambiguity advantage effect could just been ascertained in lexical decision tasks. Susan A. Duffy et al. (1988: 442) reported that ‘contextual information influences processing by varying the availability of alternative meanings of ambiguous words.’ Hence a processing disadvantage for ambiguous words could be recognised if the neighbourhood was unambiguous. Further support was given by the study of Alan H. Kawamoto et al. (1994: 12411242) who argued that a processing advantage is determined by lexical ambiguity in contrast to semantic or syntactic ambiguity.
Interestingly, all previous lexical decision studies commonly agreed that many meanings produce a processing advantage. But none of these studies distinguished between the different types of lexical ambiguity and thus utilised them interchangeably. A more recent study by Rodd et al. (2002: 259) investigated this new approach with surprising results. They were able to verify a new theory of word recognition. Rodd et al. used a simple lexical decision task which included homonymous and polysemous words. Thereby they differed between four different ‘word’ types. The distinguishing factors were ambiguity and number of senses. They found out that the ambiguity advantage is favoured by multiple senses opposed to the common assumption of multiple meanings. Furthermore, in 2007, Ekaterini Klepousniotou and his colleague (2007: 20) achieved the same results and supported the finding that a processing advantage can be recognised for words with multiple senses compared to words with multiple meanings. Thereby, they rejected the common assumption that all ambiguous words have separate representations. They hypothesised that only homonyms have separate representations and polysemes have a single core meaning (Klepousniotou & Baum 2007: 19). ‘Since only one meaning is stored, there is no competition among meanings for activation, as might happen in the case of homonymous words’ (Klepousniotou & Baum 2007: 17). On this account, the polysemous words will be localised faster and provide a processing advantage.
Since these findings are quite new and might be very controversial the present study will aim to investigate them. In particular, this study will replicate the study of Rodd et al. by using the same method and corpus. Thus a comparison can be realised in the discussion which could support this newly developed theory.
For this study, the experiment was designed with OpenSesame, which is a free and publicly available computer program. OpenSesame is a user-friendly, graphical and open-source experiment builder for social science which enables the building of complex and scientific experiments (Mathôt 2012: 314). Its easy handling allows the creating of many different experiments within a short period of time and with little effort. The program includes several tools to specify experiments for the conditions of scientists. It provides a precise time measurement, a low deflexion factor and permits an easy evaluation.
The study consisted of 200 words, whereof 128 were actual words and 72 non-words (cf. Appendix A). The actual words were identical with the stimuli of the second experiment of the study of Rodd et al. (2002: 263-264) which was the basis of the present study. Thereby the number of homonyms and polysemes were equal. The non-words were randomly composed to conceal the real purpose of the study. Rodd at al. adopted these words from the Online Wordsmyth dictionary which is a dictionary for linguists. It was designed by lexicographers and provides a clear overview of ambiguous and unambiguous words. The experiment was constituted as a ‘2 x 2 factorial design’ where the word stimuli were determined by ambiguity and the number of senses (Rodd et al. 2002: 254). In due consideration of these criteria words were classed as being unambiguous if they had only one entry in the Wordsmyth dictionary and as being ambiguous if they had more than one entry (Rodd et al. 2002: 254-255). Further, all words were nouns from different topics to avoid falsification and an objection of the results. Thereby, a general prediction of the representation in the mental lexicon will be reliable.
20 subjects participated in this study of whom ten were woman and ten were men. All informants were advanced learners of English between the ages of 20 to 30. Two of them had English as their first language. All participants were in a physically and mentally capable constitution to perform the experiment.
3.3 Data collection procedure
Generally, the experiment was designed as a lexical decision task. Two programs were used for the data collection, firstly OpenSesame for the execution of the experiment and secondly Microsoft Excel for the evaluation of the results. Participants were individually tested. They were seated in a quiet room in front of a computer screen. The trial lasted approximately 3 minutes. The experiment started with a welcome screen followed by instructions. Instructions were given to decide if the given word was a word or a non-word. Therefore two buttons were required to be pressed - the button 'a' for a word, and the button 'l' for a non-word . The participant was told to decide as quickly and accurately as possible. Next a trial run of words and non-words was started to get the participant used to the contrived task of the real experiment. After this the experiment was paused and the participant had to press any key to continue with the real experiment. Then the 200 items were displayed in an unsorted order. This order differed between every attempt. The participant had to decide if the words on the computer screen were words or non-words. No sounds or feedback were included. Moreover all requested words remained on the computer screen until the participant decided. At the end a thank-you concluded the experiment and the data was saved as a CSV file. For the purpose of evaluating the data the file was imported into Microsoft Excel. This program presented the results tabulated in the order they were requested. After all participants finished the experiment the data was sorted into four classes of words. Ambiguous words with few senses, ambiguous words with many senses, unambiguous words with few senses and finally unambiguous words with many senses (cf. Appendix A). Afterwards average values and standard deviations were calculated. Based on these calculations graphs and tables were produced.
Nevertheless, the experiment also had limitations. While the study was performed with a computer program it is obvious that technical problems can occur. Besides the results might be adulterated because it is an artificial situation which might make the informants nervous and manipulate their performance. Moreover the used corpus only included a specific number of nouns and did not include all existing ambiguous words. Hence an undisputed prediction is only possible to a limited extent.
Before all the research results can be related to previous findings the received results will need to be presented. Therefore, the next section provides the most conspicuous and significant outcomes of the present experiment. In order to facilitate comprehension graphs and tables will be included and explained.
4 Results and discussion
4.1 Results evaluation
The analysis only includes targets with correct responses. All given answers which took a longer response time than 1200ms were not included. For the non-words it means that 6.04% of the targets were removed because of wrong answers and 2.71% were removed because they exceeded the time limit of 1200ms. Since this study is just concerned with the real words and not the non-words the following percentages only deal with the percentages of the real words. Thereby, 2.11% of the targets were excluded from the analysis because their response times exceeded the time limit of 1200ms. The error rate of wrong answers amounted to 3.05% in total. The bar chart in Graph 1 exhibits significant evidence for different kinds of ambiguity. The mean reaction time reveals that the ambiguity advantage effect is not just displayed between ambiguous and unambiguous words but also between words with few and many senses. A comparison between ambiguous words with few senses and unambiguous words with many senses indicates a considerably differential processing advantage for unambiguous words with many senses. The processing advantage thereby amounts to 7.14%. This percentage is quite reliable since the standard deviation for both categories is almost the same. The reaction time
(RP), the standard derivation (SD) and the error rate can be seen in Table 1. Further support for the ambiguity advantage effect for unambiguous words with many senses is revealed by the comparison of unambiguous words with few senses and unambiguous words with many senses. Thereby, the average response time for few senses is 4.00% greater than for many senses. Besides, the ambiguity advantage effect for unambiguous words can also be recognized for unambiguous words with few senses compared with ambiguous words with few senses. In doing so the mean reaction time for the unambiguous targets are 3.01% faster than for the ambiguous targets. However, this advantage cannot be identified as obviously for ambiguous words with many senses compared to unambiguous words with many senses. The ambiguity advantage effect can still be seen but to a limited extent. Unambiguous words with many senses are on average only retrieved 1.05% faster than ambiguous words with many senses. Interestingly, an opposite result is to be seen if one contrasts ambiguous words with many senses to unambiguous words with few senses. The ambiguity advantage effect for unambiguous words is not reflected but the processing advantage for many senses it still revealed. In contrast to the previous findings the unambiguous words with few senses are retrieved 2.93% slower than the ambiguous words with many senses. Statistically, it is striking that words with few senses always indicate a slower mean reaction time than many senses regardless of the ambiguity of the word. The mean reaction time for ambiguous words with many senses is 6.03% faster than for ambiguous words with few senses. The weighting between unambiguous words with many senses and unambiguous words with few senses is nearly equal. They are retrieved 4.00% faster. In proportional terms many senses produce an average processing advantage of 5.02%. Nevertheless, the results need to be examined critically since all word classes demonstrate increased standard deviations and error rates. Especially the error rates of the unambiguous words must be viewed with a critical eye. These are approximately three times as great as the errors of the ambiguous words with few senses.
Abbildung in dieser Leseprobe nicht enthalten
Graph 1: Results of the lexical decision task: average response time (RP) for each word class in milliseconds (ms). Source: own data collection
 Below the response time is abbreviated by RT
 NOM is the abbreviation for number of meanings
 Refer to http://www.wordsmyth.net to access more information about the history and the dictionary itself
 Below the standard derivation is abbreviated by SD