<?xml version="1.0" standalone="yes"?>
<Paper uid="C02-1158">
  <Title>Study of Practical Effectiveness for Machine Translation Using Recursive Chain-link-type Learning</Title>
  <Section position="3" start_page="0" end_page="2" type="metho">
    <SectionTitle>
2 Basic Idea
</SectionTitle>
    <Paragraph position="0"> RCL is a method with an ability that automatically acquires translation knowledge in a computerwithoutanyanalyticalknowledge, suchas GA-ILMT. This is the ability to extract corre- null sponding parts from pairs of objects with which itcorresponds. Inthispaper,weapplythisability to a translation example that consists of SL and TL sentences. A system with RCL can acquire translation rules from sparse translation examples. Figure 2 shows how translation rules are acquired using this method  .</Paragraph>
    <Paragraph position="1"> Figure 2 shows the process where translation rules B, C and D are acquired one after another using RCL. In this paper, source parts arethose parts that are extracted from the SL sentences of translation examples, and target parts are those parts that are extracted from the TL sentences of translation examples. Moreover, part translation rules are pairs of source parts and  In Figure 2, the use of a Greek character means that all language characters correspond to unknown character strings for a computer.</Paragraph>
    <Section position="1" start_page="1" end_page="1" type="sub_section">
      <SectionTitle>
4.1 Translation process
</SectionTitle>
      <Paragraph position="0"> In the translation process, the system generates translation results using acquired translation rules. First, the system selects the sentence translation rules that can be applied to the SL sentence. Second, the system generates thetranslationresultsbyreplacingthevariables in the sentence translation rules with the part translation rules.</Paragraph>
    </Section>
    <Section position="2" start_page="1" end_page="1" type="sub_section">
      <SectionTitle>
4.2 Feedback process
</SectionTitle>
      <Paragraph position="0"> In the feedbackprocess, the system evaluates the translation rules used. First, the system evaluatesthetranslationruleswithoutvariables by using the results of combinations between the translation rules with variables and the translation rules without variables(Echizen-ya et al., 1996). Next, the system evaluates translationruleswithvariablesbyusingtheprocesses null of combinations between the translations rules with variables and the translation rules without variables(Echizen-ya et al., 2000). As a result, the system increases the correct translation frequencies, or the erroneous translation frequencies, of the translation rules by using these evaluation methods for the translation rules.</Paragraph>
    </Section>
    <Section position="3" start_page="1" end_page="2" type="sub_section">
      <SectionTitle>
4.3 Learning process
4.3.1 GA-ILMT
</SectionTitle>
      <Paragraph position="0"> In this paper, by using the process of acquisition of translation rules in GA-ILMT, the system acquires both sentence and part translation rules. These rules are then used as starting points when the system performs RCL.</Paragraph>
      <Paragraph position="1">  Learning(RCL) In this section, we describethe processofacquisition of translation rules using RCL. The details of the process of acquisition of part translation rules are as follows.</Paragraph>
      <Paragraph position="2"> (1)The system selects translation examples that have common parts with the sentence translation rules.</Paragraph>
      <Paragraph position="3"> (2)The system extracts the parts that correspond to the variables in the source parts and in the target parts of the sentence translation rules from the SL sentences, and the TL sentences of the translation examples. null (3)The system registers pairs, of the parts extractedfromtheSLsentencesandtheparts null extracted from the TL sentences, as the part translation rules.</Paragraph>
      <Paragraph position="4"> (4)The system gives the correct and erroneous frequencies of sentence translation rules to the acquired part translation rules.</Paragraph>
      <Paragraph position="5"> Figure 4  shows an example of the acquisitionofaparttranslationruleusingthesentence null translation rule. In Figure 4, (thirty;30[sanju]) as the part translation rule is acquired because &amp;quot;thirty&amp;quot; corresponds to the variable in the source part of sentence translation rule and &amp;quot;30[sanju]&amp;quot; corresponds to the variable in the target part of sentence translation rule.</Paragraph>
      <Paragraph position="6">  translation rule using the sentence translation rule.</Paragraph>
      <Paragraph position="7"> The details of the process of acquisition of sentence translation rules are as follows: (1)Thesystemselectstheparttranslationrules in which the source parts are included in theSLsentencesofthetranslationexample or in the source parts of sentence translationrules, andinwhichthetargetpartsare included in the TL sentences of the translation examples or in the target parts of sentence translation rules.</Paragraph>
      <Paragraph position="8"> (2)The system acquires new sentence translation rules by replacing the parts which are same as the part translation rules with the variablestothetranslationexamplesorthe sentence translation rules.</Paragraph>
      <Paragraph position="9"> (3)The system gives the correct and erroneous frequencies of the part translation rules to the acquired sentence translation rules.</Paragraph>
      <Paragraph position="10">  Italics are the pronunciation in Japanese. Figure 5 shows examples of the acquisition of the sentence translation rules using the part translation rules. In Figure 5, the system acquiresIt starts in @0 minutes.f/x /@0//ho/y//b}[Sore wa @0 pun tate ba hajimari masu.]as a sentence translation rule by using the part translation rule (thirty;30[sanju]) acquired in Figure4, and@1 starts in @0 minutes.@1/x/@0//ho/y/ /b}[@1 wa @0 pun tate ba hajimari masu.]as the sentence translation rule, that is more abstracted, is acquired by using the part translation rule (it;f[sore]).</Paragraph>
    </Section>
  </Section>
  <Section position="4" start_page="2" end_page="2" type="metho">
    <SectionTitle>
5 Experiments for performance
</SectionTitle>
    <Paragraph position="0"> evaluation</Paragraph>
    <Section position="1" start_page="2" end_page="2" type="sub_section">
      <SectionTitle>
5.1 Experimental procedure
</SectionTitle>
      <Paragraph position="0"> There are two kinds of data as experimental data. One is learning data and the other is evaluation data. In these experiments, 1,759 translationexampleswereusedaslearningdata.</Paragraph>
      <Paragraph position="1"> These translation examples were taken from textbooks(Nihon Kyozai(1), 2001; Nihon Kyozai(2), 2001; Hoyu Shuppan, 2001) for second-grade junior high school students. As well, 1,097 translation examples were used as evaluation data. These translation examples were taken from textbooks(Bunri, 2001; Sinko Shuppan, 2001) for second-grade junior high school students. Allofthesetranslationexampleswere processed by the method outlined in Figure 3.</Paragraph>
      <Paragraph position="2"> The initial condition of the dictionary is empty.</Paragraph>
      <Paragraph position="3">  Moreover,weusedthreeothercommercialRule-Based MT systems, comparing our system with those systems. We call these three MT systems A, B and C respectively.</Paragraph>
    </Section>
    <Section position="2" start_page="2" end_page="2" type="sub_section">
      <SectionTitle>
5.2 Evaluation standards
</SectionTitle>
      <Paragraph position="0"> The correct translation results are grouped into two categories:</Paragraph>
    </Section>
  </Section>
  <Section position="5" start_page="2" end_page="2" type="metho">
    <SectionTitle>
(1) The correct translation
</SectionTitle>
    <Paragraph position="0">  Thismeansthatthetranslationresultscorrespond to the correct translation results taken from textbooks respectively(Bunri, 2001; Sinko Shuppan, 2001).</Paragraph>
    <Paragraph position="1"> (2) A correct translation which includes unknown words This means that the translation results with substituted nouns or adjectives as variables correspond to the correct translation results taken from textbooks respectively(Bunri, 2001; Sinko Shuppan, 2001). In this paper, the effective translation results are the translation results that correspond to (1) and (2), and the ineffective translation results are the translation results that do not correspond to (1) and (2). Moreover, the effective translation rate is the rate of the effective translation results in all the evaluation data.</Paragraph>
    <Paragraph position="2"> The translation results are ranked when several translation results are generated. The translation results using the translation rules whose rate of correct translation frequency is high, are ranked at the top. We evaluated the translation results that are ranked from No.1 to No.3.</Paragraph>
    <Section position="1" start_page="2" end_page="2" type="sub_section">
      <SectionTitle>
5.3 Experimental results and discussion
</SectionTitle>
      <Paragraph position="0"> Table 1 shows examples of effective translation results in our system with RCL. Table 2 shows the results of comparative experiments of our system and the three Rule-Based MT systems.</Paragraph>
      <Paragraph position="1"> We excluded 309 SL sentences from 1,097 SL sentences used as evaluation data in Table 2. In oursystem, the309SLsentencesbecametheineffective translation results because of a lackof learning data. Therefore, the 309 SL sentences are not inadequate as evaluation data. Table 2 shows the effective translation rates in 788 SL sentences, which were left after excluding 309 SL sentences from the 1,097 SL sentences used as evaluation data. In the other three Rule-Based MT systems, the same 788 SL sentences were used as evaluation data and the translation results which correspond to (1) and (2)  Examples of the correct translation results SL sentences TL sentences This bag was made in France.\wx apb}[Kono baggu wa furansu sei desu.] We went there to playh`hjxbhf\V`h} baseball. [Watashi tachi wa yakyu wo suru tame soko e iki mashi ta.]  ythe noun &amp;quot;Kyoto&amp;quot;.</Paragraph>
      <Paragraph position="2"> described in section 5.2 were evaluated as the correct translation results. The effective translation rate in the system with only GA-ILMT was 45.1%. In Table 2, the effective translation rate of system with RCL is almost the same as the effective translation rates of system A, but is higher than systems B and C.</Paragraph>
      <Paragraph position="3">  Moreover, we evaluated translation results more strictly in terms of the quality of translation. Meaning that only translation results that had almost the same character strings as the correct translation results taken from the textbooks(Bunri, 2001; Sinko Shuppan, 2001) were effective translation results. For example, &amp;quot;fx10TTb[Sore wa yaku juppun kakari masu.]&amp;quot; is an ineffective translation result because of the correct translation results for &amp;quot;It takes about ten minutes.&amp;quot; is &amp;quot;10TTb[Yaku juppun kakari masu.]&amp;quot; in textbook(Bunri, 2001; Sinko Shuppan, 2001). In this Japanese sentence, phrase &amp;quot;fx[sore wa]&amp;quot; results in needlessly long. Therefore, we evaluate the translation results that have different phrases to the correct translation results as the ineffective translation results in terms of the quality of translation. Table3showsacomparisonofeffectivetranslation null ratesbasedon quality. InTable3, weconfirmed that the system with RCL can generate more high-quality translation results than the three other Rule-Based MT systems.</Paragraph>
      <Paragraph position="4"> In the system with RCL, the erroneous translation rules are also acquiredlike a linked chain. For example, in Figure 2, the translation rules B, C and D are acquired as the erroneous translation rules when the translation rule A is the erroneous translation rule. Namely, a chain reactioncauses theacquisition oferroneous translation rules. In learning data, the rate of erroneous part translation rules to the acquired part translation rules was 47.9%, and the rate of erroneous sentence translation rules to the acquired sentence translation rules was 38.2%.</Paragraph>
      <Paragraph position="5"> However, such erroneous translation rules are automatically decided as being erroneous translation rules in the feedbackprocess resulting from the ineffective translation results.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>