（節(jié)選）外文翻譯--潛在關系映射引擎算法與實驗（英文）

1、Journal of Artificial Intelligence Research 33 (2008) 615-655 Submitted 09/08; published 12/08The Latent Relation Mapping Engine: Algorithm and ExperimentsPeter D. Turney peter.turney@nrc-cnrc.gc.caInstitute for Informat

2、ion TechnologyNational Research Council CanadaOttawa, Ontario, Canada, K1A 0R6AbstractMany AI researchers and cognitive scientists have argued that analogy is the core of cognition. The most influential work on computati

3、onal modeling of analogy-making is Structure Mapping Theory (SMT) and its implementation in the Structure Mapping Engine (SME). A limitation of SME is the requirement for complex hand-coded representations. We introduce

4、the Latent Relation Mapping Engine (LRME), which combines ideas from SME and Latent Relational Analysis (LRA) in order to remove the requirement for hand- coded representations. LRME builds analogical mappings between li

5、sts of words, using a large corpus of raw text to automatically discover the semantic relations among the words. We evaluate LRME on a set of twenty analogical mapping problems, ten based on scientific analogies and ten

6、based on common metaphors. LRME achieves human-level performance on the twenty problems. We compare LRME with a variety of alternative approaches and find that they are not able to reach the same level of performance.1.

7、IntroductionWhen we are faced with a problem, we try to recall similar problems that we have faced in the past, so that we can transfer our knowledge from past experience to the current problem. We make an analogy betwee

8、n the past situation and the current situation, and we use the analogy to transfer knowledge (Gentner, 1983; Minsky, 1986; Holyoak Hofstadter, 2001; Hawkins & Blakeslee, 2004).In his survey of the computational mode

9、ling of analogy-making, French (2002) cites Structure Mapping Theory (SMT) (Gentner, 1983) and its implementation in the Structure Mapping Engine (SME) (Falkenhainer, Forbus, & Gentner, 1989) as the most influential

10、work on modeling of analogy-making. In SME, an analogical mapping M : A → B is from a source A to a target B. The source is more familiar, more known, or more concrete, whereas the target is relatively unfamiliar, unknow

11、n, or abstract. The analogical mapping is used to transfer knowledge from the source to the target.Gentner (1983) argues that there are two kinds of similarity, attributional similarity and relational similarity. The dis

12、tinction between attributes and relations may be under- stood in terms of predicate logic. An attribute is a predicate with one argument, such as large(X), meaning X is large. A relation is a predicate with two or more a

13、rguments, such as collides with(X, Y ), meaning X collides with Y .The Structure Mapping Engine prefers mappings based on relational similarity over mappings based on attributional similarity (Falkenhainer et al., 1989).

14、 For example, SME is able to build a mapping from a representation of the solar system (the source) to ac ?2008 National Research Council Canada. Reprinted with permission.The Latent Relation Mapping Engine(defEntity nuc

15、leus :type inanimate) (defEntity electron :type inanimate)(defDescription rutherford-atomentities (nucleus electron) expressions (((mass nucleus) :name mass-n)((mass electron) :name mass-e) ((greater mass-n mass-e) :name

16、 >mass) ((attracts nucleus electron) :name attracts-form) ((revolve-around electron nucleus) :name revolve) ((charge electron) :name q-electron) ((charge nucleus) :name q-nucleus) ((opposite-sign q-nucleus q-electron)

17、 :name >charge) ((cause >charge attracts-form) :name why-attracts)))Figure 2: The Rutherford-Bohr model of the atom in SME (Falkenhainer et al., 1989).However, the CogSketch user interface requires the person who d

18、raws the sketch to iden- tify the basic components in the sketch and hand-label them with terms from a knowledge base derived from OpenCyc. Forbus et al. (2008) note that OpenCyc contains more than 58,000 hand-coded conc

19、epts, and they have added further hand-coded concepts to OpenCyc, in order to support CogSketch. The Gizmo system requires the user to hand-code a physical model, using the methods of qualitative physics (Yan & Forbu

20、s, 2005). Learning Reader uses more than 28,000 phrasal patterns, which were derived from ResearchCyc (Forbus et al., 2007). It is evident that SME still requires substantial hand-coded knowledge.The work we present in t

21、his paper is an effort to avoid complex hand-coded representa- tions. Our approach is to combine ideas from SME (Falkenhainer et al., 1989) and Latent Relational Analysis (LRA) (Turney, 2006). We call the resulting algor

22、ithm the Latent Re- lation Mapping Engine (LRME). We represent the semantic relation between two terms using a vector, in which the elements are derived from pattern frequencies in a large corpus of raw text. Because the

23、 semantic relations are automatically derived from a corpus, LRME does not require hand-coded representations of relations. It only needs a list of terms from the source and a list of terms from the target. Given these t

24、wo lists, LRME uses the corpus to build representations of the relations among the terms, and then it constructs a mapping between the two lists.Tables 1 and 2 show the input and output of LRME for the analogy between th

25、e solar system and the Rutherford-Bohr model of the atom. Although some human effort is involved in constructing the input lists, it is considerably less effort than SME requires for its input (contrast Figures 1 and 2 w

26、ith Table 1).Scientific analogies, such as the analogy between the solar system and the Rutherford- Bohr model of the atom, may seem esoteric, but we believe analogy-making is ubiquitous in our daily lives. A potential p

27、ractical application for this work is the task of identifying semantic roles (Gildea & Jurafsky, 2002). Since roles are relations, not attributes, it is appropriate to treat semantic role labeling as an analogical ma

28、pping problem.For example, the Judgement semantic frame contains semantic roles such as judge, evaluee, and reason, and the Statement frame contains roles such as speaker, ad- dressee, message, topic, and medium (Gildea