m7n/nomic-embed-philosophy-triplets_v5

SentenceTransformer based on nomic-ai/nomic-embed-text-v1

This is a sentence-transformers model finetuned from nomic-ai/nomic-embed-text-v1. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: nomic-ai/nomic-embed-text-v1
• Maximum Sequence Length: 8192 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Cosine Similarity

Model Sources

• Documentation: Sentence Transformers Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 8192, 'do_lower_case': False}) with Transformer model: NomicBertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("m7n/nomic-embed-philosophy-triplets_v5")
# Run inference
sentences = [
    '1993]. So being a "visual display" or being dependent on visual processing cannot be what distinguishes pictures from other modes of representation. Second, the distinction between the visual and the cognitive is not only unclear and empirically suspect [Schwartz 1994, 1996], it does not seem up to the job. Standard proposals for drawing a visual/cognitive boundary fail to support the surrogate theorist\'s core intuition. For example, it is often said picture perception is non-cognitive, since it does not involve active deliberation or contemplation. But the same holds for everyday cases of language comprehension. Without delay, pondering, or conscious inference we read and understand immediately most sentences we encounter. Identifying the cognitive with the learned, and the visual with the innate, will not work either. Many animal signal systems are instinctive, but they are not pictorial. And were humans equipped at birth with Latin, or as some claim "Mentalese," it would not mean these representational systems were pictorial.\' Still, I think, proponents of the symbolic model can do more to accommodate the intuition of there being something more "visual" about pictures than natural languages, music notation, and various other forms of representation. And it is to this task I wish to turn my attention. The account I shall sketch makes use of ideas Goodman employs in his taxonomy of symbolic systems. Goodman argues that pictures, in contrast, say, to sentences in language or scores in music notation, belong to dense schemes of representa2 Variations of this criticism are',
    "leveled against the model not only by vision theorists, but by art historians and philosophers who balk at what they take to be the symbolic paradigm's conventionalist implications. 3 Much is often made of evidence suggesting that infants [Hochberg and Brooks, 1962] and people from non-Western cultures [Derogowski, 1989] can understand pictures without training. My point here is not to challenge these empirical findings but to call attention to the need to separate issues of learning and innateness from claims about the form and conventionality of symbols. 708 ROBERT SCHWARTZ tion. Pictorial symbols are analog, while English sentences constitute a digital system. Goodman further claims pictorial schemes are comparatively replete. Many more of the symbols' own properties function in determining its representational content. A given line may be understood as a graph, plotting the height of a mountain range, or as a picture of the same mountain (fig. 1). Fig. 1 Read as a graph, the thickness of the line, its color, and background have no significance. Interpreted as a picture, all these properties go to constitute the displays representational force. Notice, something phenomenal, akin to a Gestalt switch or aspect change, occurs when shifting between the two readings. And experience of the line takes on another much different character when read in the context of figure 2. Fig. 2 Simply making a graph more replete, however, will not turn it into a picture. Nor will assigning representational significance to the background do the trick. For if these additional features of the graph",
    'Goodman\'s account of depiction may be described as minimalist at best. Pictures denote or refer. Among his undisputed contributions is his discussion of what he calls the routes of reference-denotation, predication, exemplification, and expression. Pictures represent in all these ways. But Goodman is not interested in the roots of reference-how referential relationships are established.5 His remarks about what determines what a picture represents are entirely negative: resemblance is neither necessary nor sufficient for depiction.6 Beyond rejecting the resemblance theory, Goodman refuses to give "general instructions for determining what a work describes or represents."7 Every system of pictures includes a plan of correlation mapping 228 pictures onto what they represent, but Goodman is silent about what determines or constrains these plans of correlation. Others have been tempted to fill the lacuna, drawing on resources internal or external to Goodman\'s account of pictures.8 In particular, it is widely held that Goodman favors a convention theory of depiction. Some go so far as to ascribe to him the view that pictures are verbal symbols!9 But Goodman denies that he holds either view, insisting that "there is no vocabulary of picturing as there is of saying," and depiction does not depend on "rule-following."\'0 Indeed, the convention view is incompatible with the claim that pictures belong to analog schemes, as conventions are rules operating upon disjoint and differentiated characters. "I That Goodman refuses to give an account of depiction does not mean that he thinks none can be',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Triplet

• Dataset: nomic
• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.96
dot_accuracy	0.04
manhattan_accuracy	0.956
euclidean_accuracy	0.96
max_accuracy	0.96

Triplet

• Dataset: nomic
• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.963
dot_accuracy	0.037
manhattan_accuracy	0.9625
euclidean_accuracy	0.963
max_accuracy	0.963

Triplet

• Dataset: nomic
• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.963
dot_accuracy	0.037
manhattan_accuracy	0.9625
euclidean_accuracy	0.963
max_accuracy	0.963

Training Details

Training Dataset

Unnamed Dataset

• Size: 10,000 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 271 tokens mean: 333.67 tokens max: 570 tokens	min: 272 tokens mean: 333.9 tokens max: 727 tokens	min: 263 tokens mean: 332.95 tokens max: 571 tokens

• Samples:

  anchor	positive	negative
  such as Argentina, Brazil and Mexico (Sagasti and Guerrero 1974; Stepan 1981; Chambers 1987; Lafuente and Sala Catala 1989). However, recent studies on the sociology of science, philosophy of science, and scientific literature which stress and validate a more local and comparative perspective have lead to identify key elements in the diffusion process of © 2013 Stazione Zoologica Anton Dohm 70 ANA BARAHONA scientific knowledge and to more accurate approaches with respect to local complexities. There is, nevertheless, a need for historical studies that acknowledge the complex interactions generated after the contact between imported scientific novelties and local cultural traditions, which have yielded different results in different countries and cultural settings. Many problems have caught my attention influenced by the innovative perspective of Hans-Jörg Rheinberger 's project A Cultural History of Heredity. Especially the perspective on "the emergence of specific practices [...] the shaping of standards, and the conjunctions of these elements in a variety of social arenas" (MüllerWille and Rheinberger 2007, 13) has proved to be a useful lens through which to study many problems related to the history of genetics in Mexico. In my reconstructions I have analyzed the conditions for scientific research and the social relationships that allowed the establishment of genetics in Mexico, in the laboratory, the clinic and in agronomy. I included in this historical reconstruction the institutions, the interests, the bodies of norms, and the practices that contributed to the circulation of knowledge and methods imported from abroad, the effects that small local communities	had during the early stages of the consolidation of genetics in Mexico, and the idiosyncrasies involved in the uses, diffusion and acceptance of genetics by agricultural, academic, and professional communities. In MüllerWille and Rheinberger 's words, my research thus took "a fresh approach to the history of heredity, drawing on a wealth of well-researched case studies, in order to embrace the cultural history of heredity [in Mexico]. This involves tracing the emergence of the knowledge of heredity in broader social and historical contexts, from a wide synchronic and diachronic perspective" (MüllerWille and Rheinberger 2007, 8). In this paper I give three examples to illustrate how the cultural history of heredity has enlightened my work: the introduction and institutionalization of Mendelism in Mexico in the early twentieth century, the hereditarian ideas of medical doctors in the late nineteenth century, and the introduction of medical genetics in Mexico. I have shown that Mendelism in Mexico was introduced taking into account the cultural particularities of Mexican peasants (Barahona et al. 2005; Barahona 2008). I have also explored the distinctive characteristics of Mexico's society, politics and history that impacted on the establishment of genetics in Mexico, as a new disciplinary field. The beginnings of this process of introduction and institutionalization in the early twentieth century, finally, have been traced by following the work of agronomist Edmundo Taboada and his group during the government of General Lázaro Cardenas (1934-1940), and the role played by the Rockefeller Foundation in Mexican agricultural research. This process was consoliTHE	Exeter. This project included five large-scale conferences, which have thus far produced four preprint volumes (Staffan Müller-Wille, ed., A Cultural History of Heredity I: 17th and 18th Centuries [Berlin: Max-Planck-Institut für Wissenschaftsgeschichte, 2002]; Hans-Jörg Rheinberger and Staffan Müller-Wille, eds., A Cultural History of Heredity II: 18th and 19th Centuries [Berlin: Max-Planck-Institut für Wissenschaftsgeschichte, 2003]; Staffan Müller-Wille and Hans-Jörg Rheinberger, eds., A Cultural History of Heredity III: 19th and Early 20th Centuries [Berlin: Max-Planck-Institut für Wissenschaftsgeschichte, 2005]; Staffan Müller-Wile, Hans-Jörg Rheinberger, and John Dupré, eds., A Cultural History of Heredity IV: Heredity in the Century of the Gene [Berlin: Max-Planck-Institut für Wissenschaftsgeschichte, 2008]), one published edited volume (Staffan Müller-Wille and Hans-Jörg Rheinberger, eds., Heredity Produced: At the Crossroads of Biology, Politics, and Culture, 1500–1870 [Cambridge, MA: MIT Press, 2007]), and a further edited volume in preparation (Staffan Müller-Wille and Christina Brandt, eds., Heredity Explored: Between Public Domain and Experimental Science, 1850–1930). This book, then, is the authors’ own narrative synthesis, as project directors, of the results of this long and extremely fruitful international scholarly collaboration. The work begins with an introductory chapter that presents the authors’ broad scope by reference to the writings of Francis Galton, whose metaphors of heredity as a “post office” and a “parliament”—in which the hereditary materials are packed and shipped in mailbags before competing for a limited number of “places” in the developed character of an individual—immediately situate that work in a broader sociocultural context. The
  differentiated from mentoring in that it does not require a Springer close, personal relationship between models and observers. In fact, there are wide varieties of potentially important people who can be selected as role models such as distant leaders, co-workers, and inspiring individuals from all walks of life (e.g., teachers, sports heroes, religious figures, family members). We focus on role models because prior research grounded in social learning theory has demonstrated their impact on moral judgment and action (Bandura 1991). Modeling influences have been associated with the development of prosocial behavior in children (Eisenberg and Fabes 1998) and ethical behavior in the workplace (Moberg 2000; Sims and Brinkmann 2002; Weaver et al. 2005). Within the leadership literature, both transformational leaders (Avolio 1999) and ethical leaders (Brown et al. 2005) have been described as ethical role models for others. The assumption is that having been exposed to ethical role models contributes to the development of one's ethical leadership (Brown and Treviño 2006; Weaver et al. 2005). In this research, we investigate whether the ethical role models of leaders are related to employees' ratings of their ethical leadership. Because ethical role models can take many forms, we consider three different types of ethical role models that are potential influences on the development of ethical leadership (a) childhood models (e.g., parents, teachers, and coaches); (b) career models (e.g., mentors or supervisors), and (c) top managers who model ethics for employees in the organization (Treviño et al. 2000). We ground our hypotheses in social learning	theory (Bandura 1986, 1991) and the influence of modeling on the acquisition of moral reasoning and standards. Three main questions guided our research. Are role models related to ethical leadership? If so, what types of models (i.e., childhood models, career mentors, top managers) are influential? Given that the types of role models selected as well as the lessons learned from such models change over the lifetime of the learner (Gibson 2003), does leader age moderate the relationships between different types of role models and ethical leadership? Theory and Hypotheses Modeling, Social Learning, and Ethical Leadership Social learning theory helps explain why individuals are likely to seek guidance from role models, and how role modeling might be related to ethical leadership. Social learning theory posits that individuals learn what to do and how to behave largely by observing and emulating role models. Most adults are not ethically self-sufficient. Rather, they look outside themselves to peers and significant others for Role Modeling and Perceived Ethical Leadership 589 ethical guidance (Kohlberg 1969; Treviño 1986). This is particularly true because ethical dilemmas often involve ambiguity and individuals attempt to reduce such ambiguity by turning to others for guidance. The social learning process begins when individuals focus their attention on modeled behaviors. Among the potential models to choose from, attractive models capture a learner's attention. Attractiveness is based on a number of model characteristics such as nurturance (Yussen and Levy 1975), status (Lippitt et al. 1952; Lefkowitz et al. 1955), competence (Kanareff and Lanzetta	interpersonal relationships and the promotion of such conduct to followers through two-way communication, reinforcement and decision-making.' ' These authors take a social learning perspective (e.g., Brown and Treviño, 2006). Social learning theory highlights that leaders are role models of appropriate behaviors and emphasizes that people learn from reward and punishment (Bandura, 1986). In other words, ethical leaders use transactional efforts (i.e., communication, rewarding, and punishing) as well as role modeling of desired behavior to stimulate subordinates ethical behavior (Brown et al., 2005; Treviño et al., 2003). Other authors conceptualize ethical leadership more in terms of a basic tension between altruistic and egoistic motives (e.g., Aronson, 2001; Turner et al., 2002). Kanungo and Mendonca (2001) expect an ethical leader to be driven by a system of accepted beließ and appropriate judgments instead of self-interest, which is beneficial for followers, organizations, and society. Brown et al. (2005) developed a 10-item unidimensional measure of ethical leadership combining various ethical leader behaviors (e.g., acting fairly, allowing voice, and rewarding ethical conduct). Walumbwa and Schaubroeck also use this unidimensional measure in their study of ethical leadership and traits. Other authors see the ethical leader behaviors that are combined in this measure as theoretically different and argue that these behaviors may have different antecedents and consequences and should ideally be measured separately (e.g., De Hoogh and Den Hartog, 2008; Kalshoven et al., in press; Resick et al., 2006). As with other leadership styles (e.g., transactional, transformational), the identification of multiple dimensions increases the comprehension
  is ultimately trival. Based upon such an explication of deduction, Popper or a Popperian could then simply admit that deductive rules are apriori justified without creating problems for the justification of deduction analogous to the problems of an apriori defense of induction. One way to show that the rules of deduction are ultimately trivial analytic truths and as such acceptable candidates for apriori knowledge is to explain how the truths of deductive logic depend solely on the meanings or a conceptual analysis of the logical constants, 'not', 'if...then', 'either...or'. This task turns out to be more difficult than it sounds, however. Tarski showed that if one presuppposes an account of some logical constants, then one can define the notion of logical conse quence. But one's definition of logical consequence was relative to one's presuppositions about which terms are logical constants. Popper's idea was that this problem could be solved without any nontrivial assumptions, and so in this way deduction might be justified apriori. In retrospect Popper explains: My papers were...inspired by the hope of solving a problem which Tarski...indicates as insoluble; rightly, I now suspect. This was the problem of distinguishing between logical (or as I prefer to call them, 'formative') signs and descriptive signs...Tarski showed that the concept of logical consequence can easily be elucidated (with the help of the concept of truth of a 'model') once we have decided upon a list of logical or formative signs. My idea was very simple: I have suggested we take	the concept 'logical consequence' as primitive and try to show that those signs are logical or formative which can be defined with the help of this primitive concept. It is only fair to say that my papers did not succeed in this (as emerges from Lejewski's analysis).4 (emphasis added) Popper's primitive was actually the relation of 'being deducible from', which supposedly he takes to be the same as the technical notion of logical consequence. Taking the meaning of the deducibility relation to be clear and trivial, he then attempts to define the logical constants without any presuppositions. What Lejewski shows is that Popper must presuppose exactly those notions he does not want to presuppose. Lejewski concludes, 'The upshot seems to be that Popper's claim to have constructed a logic without assumptions or a logic without axioms; i.e. a logic based on defi nitions alone, can hardly be upheld'.5 Given this failure, one might think that Popper could still fall back on Tarski's account of logical consequence as an analysis which justifies the apriori defense of deduction. Tarski rigo rously shows that the logical truths are analytic and thus apriori justified. That Tarski's 'definition' does this or is capable of doing this is far from obvious. As John Etchemendy argues A REFUTATION OF PURE CONJECTURE 59 ...his [Tarski's] account of logical truth and logical consequence does not capture, or even come close to capturing, any pretheoretic conception of the logical properties...Applying the model-theoretic account of consequence, I claim, is no more reliable	In the view of many authors, the intuitive notion of logical truth is closely tied to the notion of analytic truth. A clear recent example is John Etche mendy, in his book The Concept of Logical Consequence (an all-out attack on Tarski's theory of logical consequence, whose main argument against the extensional adequacy of the theory will be discussed at length in Sec tion 3.1). On Etchemendy's view, the intuitive notion of logical truth is especially linked to the notion of analyticity.9 In fact, in his book he often (though not always) assumes that a sentence S is logically true in the intuitive sense when it is true by virtue of the meanings of some of its expressions, its standard logical constants. One may justifiably wonder whether TVMC is a clearly delimited in tuitive notion within the languages in the classical hierarchy that Tarski intended the standard definition to be applicable to. One may wonder this without having any skeptical worries about the notion of meaning in gen eral. These worries might not be justified for natural language and the kind of worry that concerns us here would still remain. The issue is whether there is a sufficiently clear pretheoretical use of the notion of TVMC as applied to the formal logico-mathematical languages for which Tarski's definition is especially intended. A way to approach the issue is through examples that seem to generate conflicting intuitions about truth by virtue of meaning in formal languages. Consider, to begin with, one of those

• Loss: TripletLoss with these parameters:

  {
      "distance_metric": "TripletDistanceMetric.COSINE",
      "triplet_margin": 0.05
  }
  

Evaluation Dataset

Unnamed Dataset

• Size: 500 evaluation samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 285 tokens mean: 336.93 tokens max: 569 tokens	min: 279 tokens mean: 336.47 tokens max: 633 tokens	min: 263 tokens mean: 335.47 tokens max: 560 tokens

• Samples:

  anchor	positive	negative
  silence. In Koffler's terms, the feminine is engendered, then expelled, and finally pathologized. And Melville's purpose, in this version of the narrative, is to fight against the mechanisms of repression. By presenting charac ters who merge conventional gender characteristics, the novel sub verts those very conventions which are part of the repressive power structures that Melville fights against. The narrative itself presents the tragedy of repression, since the progressive expulsion and pathologiz ing of the feminine mark the ultimate "feminine absence." The final censorship, in this reading, is the silencing of the feminine. So to those who would say to Professor Koffler, "But there aren't any women in Billy Budd," I'd reply on her behalf, "Yes, and that's pre cisely the point." A doubt arises, however, concerning this version of the narra tive and whether we can escape the very mechanisms of repression we might want to resist. Foucault raises that same doubt concerning his own "repressive hypothesis": 21 Does the critical discourse that addresses itself to repression come to act as a roadblock to a power mechanism that has operated unchallenged up to that point, or is it not in fact part of the same histori cal network as the thing it denounces (and doubt less misrepresents) by calling it "repression"?3 The power of Foucault's doubt concerns the possibility of a metalan guage. That is, Foucault is primarily interested in describing linguis tic/discursive mechanisms of repression ? especially censorship, law, and silence. And his point is that any attempt at describing	the discur sive mechanisms of repression may well find itself becoming part of these very discursive mechanisms themselves. Foucault carries this argument a step further, too, by doubting whether the historical net work of sexuality and power is accurately represented as repressive. In other parts of the History of Sexuality, for instance, Foucault shows that the discourse on sex during the past two or three centuries has in fact proliferated: so the conspiracy of silence and censorship becomes instead a strategy for producing more and more discourse. The final point I want to make about Foucault here is that this "incitement to discourse" is a mechanism of power. The dis courses Foucault cites tend to be medical, psychiatric, and juridical, and the function these discourses serve is, broadly speaking, "to drive sex out of hiding and constrain [it] to lead a discursive existence." So-called "perversions" are thus persecuted by being incorporated within the discursive institutions; individuals are specified by entomologizing, categorizing discourses. This means, in short, that Foucault's version of discourse as the all-pervasive medium of power, sexuality, and silence calls into question our own ability to stand outside or apart from the very mech anisms of power we denounce. Foucault recognizes, of course, that this applies equally well to his own History of Sexuality. I wish through my introductory paragraphs on Foucault to raise a doubt concerning the two discourses before us (three, if you count mine, and I think this appropriate). First, then, does Melville successfully	about that each one of us embodies a particular sexuality that needs to be known, taken care of, and expressed positively and fully? How and why did this region of human experience come to exist and get organized as it has? To understand Foucault's answer-and, thus, to understand my claim about the "intersection" of sex and race-it is necessary first to understand Foucault's conception of power. For the last several hundred years, Westerners have understood power as analogous to tools. What does it take to change a tire? It takes a jack and a lug wrench. What does it take to change a law? It takes money for communication with representatives, a piece of proposed legislation, and the power to pressure opponents to cooperate. Power is a tool like any other. Some people have that tool in their tool kit, and others lack it. A shift in the balance of power is simply a transferal of some of those tools from some people's tool kits to others'. Power is an object, then, under the control of subjects with whom its deployment originates. Those subjects who have power-tools sometimes choose to bring them out and turn them on, and other times they choose to leave them in the box.5 As he studied the history of sexuality and also the histories of criminality, military discipline, and pedagogy, Foucault found that conceiving of power as a tool under the ownership of a controlling subject simply did not work. He realized it would be necessary to rethink the entire concept, construing it
  consists in outdoing him, in espousing "a more thorough pragmatism" (FLPV, p. 46). Because any statement can be accepted or rejected as a result of our decision, the search for a distinction between single statements that are known a priori and ones that are known a posteriori is doomed to failure. There is 'empirical slack' in all our beliefs, and since the whole body of our knowledge is 'under-determined' by experience, accepting or rejecting particular items will always be a matter of decision. Is Quine's general epistemological theory true? Is he justified in rejecting the a priori-a posteriori distinction on the grounds that no statement is immune to revision? In this paper I examine the extent to which the more recent arguments in chapter two of Word and Object support his position. 83 BARRY STROUD I think there are important consequences of those arguments that have not been clearly recognized. The aim of the chapter is to make plausible the following thesis about how much of language can be made sense of in terms of its stimulus conditions: ... the infinite totality of sentences of any given speaker's language can be so permuted, or mapped onto itself, that (a) the totality of the speaker's dispo sitions to verbal behavior remains invariant, and yet (b) the mapping is no mere correlation of sentences with equivalent sentences, in any plausible sense of equivalence however loose. Sentences without number can diverge drastically from their respective correlates, yet the divergences can systematically so offset	one another that the overall pattern of associations of sentences with one another and with non-verbal stimulation is preserved, (p. 27) Putting it interlinguistically, translation between languages is said to be 'indeterminate' in the sense that: ... manuals for translating one language into another can be set up in divergent ways, all compatible with the totality of speech dispositions, yet incompatible with one another. In countless places they will diverge in giving, as their res pective translations of a sentence of the one language, sentences of the other language which stand to each other in no sort of equivalence however loose. (P. 27) It is tempting to take the thesis of the indeterminacy of translation as a less metaphorical and more precise way of making Quine's earlier point that there is always some 'slippage' or 'empirical slack' between our full blown language and the stimuli which give rise to our verbal behavior. If we can produce two or more incompatible manuals for translating a foreigner's remark, all of which square with his dispositions to respond verbally to non-verbal stimuli, then the choice of one or another of those manuals is underdetermined by those dispositions and stimuli. Or, in the domestic case, if two non-equivalent sentences S1 and S2 are correlated in the mapping of the sentences, our saying Sx rather than S2 is not deter mined by sense-experience or by non-verbal stimulation alone. Since we are bound only by the need to square the totality of our utterances with experience	genuine evidential basis: namely, native assent and dissent to the linguist's queries under concurrent observable circumstances. If Evans' attack is sound, it may prove fatal since Quine will not be able to reply by claiming that Evans' criticism relies on non-factual considSemantic Perversity 67 erations. Evans' anti-Quinean line of argument is a powerful one, and I shall spend some time in this section to review it2. Evans starts by pointing out the divergencies between the task of a translator and the task of a semanticist. The aim of the former is simply to facilitate communication between two linguistic communities. In order to do so, she must devise a manual of translation. Evans does not manifest any concern with the claim that translation suffers from indeterminacy. The reason is simply that a translator is not devoted to revealing any semantic truth. The translator's aim is simply to find smooth vehicles of communication, and insofar as this target is achieved, the way the translator dissects native utterances is completely irrelevant to her task. By contrast, the semanticist is involved in the project of constructing a theory of meaning. She is not concerned merely with correlating expressions of native with lumps of home language, but rather with stating what the native expressions actually mear i3. The sentences of Native are potentially infinite in number. The semanticist, similarly to the translator, will be obliged to dissect native sentences. The target now, however, is to account for the meaning of those previously unencountered
  any definition or justification he gave of any particular democratic ideal, such as can be found in the Public and Its Problems or Liberalism and Social Action, are not solutions to be applied but hypotheses to be evaluated in a manner outlined by his methodological work. There are many possible democratic values and procedures, problems for which they might be useful, and peoples who might share them. Thus, it is vital to understand when, where, why, how, and for whom they are valid. For this task Dewey provides an account of experimental political inquiry. This framework can best be understood by—and requires—rethinking the problem of legitimacy as understood by many philosophers. Of course, since the value of an experimental approach must itself be shown experimentally, I will conclude by outlining the ways it suggests that the “current economic crisis” might be addressed. 160 T R A N S A C T I O N S Volume 47 Number 2 Experimental Method Before we turn to Dewey’s suggestions for realizing experimental political inquiry, it will be useful to rehearse his characterization of experimental inquiry more broadly. Like Charles S. Peirce, Dewey understood inquiry as a means by which organisms deal with changes in their environments (“The Fixation of Belief ”; LW 12). Humans might rely on forms of authority—whether that of experts or past experience—to make decisions about what to do, but, as no problem is exactly the same as any previous one, there are likely to be situations for which authorities provide unhelpful advice. For this reason,	the successes of the past ought to have some bearing on our conduct without exclusively determining what should be done now. Avoiding mechanical habits cannot be as simple as following a strict set of rules, though outlining some characteristics of experimental inquiry may help us create the conditions for dealing with problems more effectively when they arise. Most generally, Dewey suggests that experimental inquiry should (1) be based in specific conflicts; (2) trace out the conditions and effects of possible ends and means; and (3) involve deliberation and action undertaken to address a problem. He writes, experimental inquiries should (1) grow out of actual social tensions, needs, ‘troubles’; (2) have their subject-matter determined by the conditions that are material means of bringing about a unified situation, and (3) be related to some hypothesis, which is a plan and policy for existential resolution of the conflicting social situation. [LW 12:493] Reflection meant to deal with problems should undertake to explore tensions among different courses of action and the goals they are likely to realize by investigating their conditions and effects. One should not assume any particular relationship between ends and means, as any set of ends—like happiness or duty, community or difference, truth or justice—will make possible a variety of courses of action in different contexts, and their effects may have bearing upon the validity of such values. Acting so as to maximize happiness may not lead to happiness, just as reforming the basic	engage and fulfill their operative interests. Inquiry into the constituents and possibilities of the situation will, if successful, yield a plan of action to accomplish 2. John Dewey, "The Construction of Good," in The Quest for Certainty (New York: G. P. Putnam's Sons, 1929). 3. White, "Value and Obligation in Dewey and Lewis," p. 322. 4. White, Social Thought in America, p. 213. 5. White, "Value and Obligation in Dewey and Lewis," p. 329. 6. See John Dewey, Reconstruction in Philosophy (New York: Henry Holt & Co., 1920), esp. chaps. 1 and 2; Experience and Nature (New York: Dover Publications, 1958), esp. chap. 10; and the article "Philosophy," in The Encyclopaedia of the Social Sciences (New York: Macmillan Co., 1934), 12:118-28. 220 Ethics the transformation. The features of the situation which are experienced as immediately attractive (or unattractive) prior to inquiry Dewey calls problematic goods. The particular action that is expected to integrate the situation he calls the end-in-view, and the unified activity itself he calls value proper, or consummatory experience. The plan of action which would transform the situation from problematic to consummatory he designates, idiosyncratically, a moral judgment.7 A clarification of the generic features of these situations is necessary for the agents involved to avail themselves of the intellectual and material instrumentalities at their disposal. Dewey's intent as a moral philosopher was precisely this: He wished to clarify the nature of the human situation in such a way that its resources and limitations could be identified and

• Loss: TripletLoss with these parameters:

  {
      "distance_metric": "TripletDistanceMetric.COSINE",
      "triplet_margin": 0.05
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 4
• per_device_eval_batch_size: 4
• warmup_ratio: 0.1
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 4
• per_device_eval_batch_size: 4
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• learning_rate: 5e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 3
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.1
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: True
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• no_cuda: False
• use_cpu: False
• use_mps_device: False
• seed: 42
• data_seed: None
• jit_mode_eval: False
• use_ipex: False
• bf16: False
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 0
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• fsdp_transformer_layer_cls_to_wrap: None
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• skip_memory_metrics: True
• use_legacy_prediction_loop: False
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: False
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• eval_do_concat_batches: True
• fp16_backend: auto
• push_to_hub_model_id: None
• push_to_hub_organization: None
• mp_parameters:
• auto_find_batch_size: False
• full_determinism: False
• torchdynamo: None
• ray_scope: last
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• dispatch_batches: None
• split_batches: None
• include_tokens_per_second: False
• include_num_input_tokens_seen: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss	nomic_max_accuracy
0	0	-	-	0.932
0.04	100	0.0072	0.0070	0.946
0.08	200	0.0047	0.0061	0.95
0.12	300	0.0051	0.0061	0.956
0.16	400	0.0051	0.0076	0.944
0.2	500	0.0084	0.0071	0.946
0.24	600	0.0076	0.0079	0.95
0.28	700	0.0097	0.0076	0.95
0.32	800	0.0092	0.0110	0.92
0.36	900	0.01	0.0125	0.9
0.4	1000	0.0114	0.1029	0.714
0.44	1100	0.012	0.0091	0.932
0.48	1200	0.0121	0.0088	0.936
0.52	1300	0.0096	0.0117	0.912
0.56	1400	0.008	0.0080	0.938
0.6	1500	0.0079	0.0088	0.928
0.64	1600	0.0105	0.0106	0.92
0.68	1700	0.0079	0.0077	0.936
0.72	1800	0.0095	0.0083	0.938
0.76	1900	0.0095	0.0075	0.946
0.8	2000	0.0083	0.0094	0.93
0.84	2100	0.0117	0.0097	0.912
0.88	2200	0.0105	0.0085	0.944
0.92	2300	0.0103	0.0110	0.906
0.96	2400	0.0092	0.0096	0.92
1.0	2500	0.0101	0.0103	0.92
1.04	2600	0.0074	0.0098	0.922
1.08	2700	0.0056	0.0084	0.94
1.12	2800	0.0038	0.0074	0.948
1.16	2900	0.0033	0.0071	0.938
1.2	3000	0.0045	0.0076	0.93
1.24	3100	0.0029	0.0089	0.936
1.28	3200	0.0024	0.0076	0.94
1.32	3300	0.0024	0.0076	0.942
1.3600	3400	0.0015	0.0068	0.946
1.4	3500	0.0027	0.0068	0.948
1.44	3600	0.0029	0.0076	0.944
1.48	3700	0.0027	0.0070	0.954
1.52	3800	0.0035	0.0077	0.944
1.56	3900	0.0034	0.0056	0.96
1.6	4000	0.0022	0.0068	0.952
1.6400	4100	0.002	0.0062	0.958
1.6800	4200	0.0029	0.0074	0.948
1.72	4300	0.004	0.0067	0.95
1.76	4400	0.0024	0.0066	0.956
1.8	4500	0.0018	0.0066	0.95
1.8400	4600	0.0034	0.0067	0.948
1.88	4700	0.0014	0.0057	0.952
1.92	4800	0.0015	0.0054	0.96
1.96	4900	0.0016	0.0061	0.95
2.0	5000	0.0017	0.0069	0.942
2.04	5100	0.0015	0.0059	0.956
2.08	5200	0.0013	0.0062	0.942
2.12	5300	0.0016	0.0061	0.944
2.16	5400	0.0013	0.0055	0.952
2.2	5500	0.0005	0.0055	0.95
2.24	5600	0.0012	0.0047	0.96
2.2800	5700	0.0017	0.0052	0.956
2.32	5800	0.0001	0.0056	0.948
2.36	5900	0.0007	0.0055	0.952
2.4	6000	0.0005	0.0052	0.964
2.44	6100	0.001	0.0050	0.968
2.48	6200	0.0008	0.0051	0.966
2.52	6300	0.0005	0.0049	0.964
2.56	6400	0.0009	0.0050	0.962
2.6	6500	0.0006	0.0050	0.956
2.64	6600	0.0006	0.0049	0.96
2.68	6700	0.0002	0.0049	0.954
2.7200	6800	0.0004	0.0050	0.95
2.76	6900	0.0004	0.0050	0.952
2.8	7000	0.0004	0.0048	0.966
2.84	7100	0.0006	0.0049	0.958
2.88	7200	0.0003	0.0049	0.964
2.92	7300	0.0004	0.0048	0.964
2.96	7400	0.0001	0.0048	0.96
3.0	7500	0.0	0.0048	0.963

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.0.1
• Transformers: 4.42.4
• PyTorch: 2.3.1+cu121
• Accelerate: 0.32.1
• Datasets: 2.21.0
• Tokenizers: 0.19.1

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

TripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification}, 
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}