Professional Documents
Culture Documents
Textbook Algorithmic Aspects in Information and Management 12Th International Conference Aaim 2018 Dallas TX Usa December 3 4 2018 Proceedings Shaojie Tang Ebook All Chapter PDF
Textbook Algorithmic Aspects in Information and Management 12Th International Conference Aaim 2018 Dallas TX Usa December 3 4 2018 Proceedings Shaojie Tang Ebook All Chapter PDF
https://textbookfull.com/product/algorithmic-aspects-in-
information-and-management-12th-international-conference-
aaim-2018-dallas-tx-usa-december-3-4-2018-proceedings-shaojie-
tang/
https://textbookfull.com/product/brain-informatics-international-
conference-bi-2018-arlington-tx-usa-
december-7-9-2018-proceedings-shouyi-wang/
https://textbookfull.com/product/algorithmic-aspects-in-
information-and-management-14th-international-conference-
aaim-2020-jinhua-china-august-10-12-2020-proceedings-zhao-zhang/
https://textbookfull.com/product/combinatorial-optimization-and-
applications-12th-international-conference-cocoa-2018-atlanta-ga-
usa-december-15-17-2018-proceedings-donghyun-kim/
Combinatorial Optimization and Applications 14th
International Conference COCOA 2020 Dallas TX USA
December 11 13 2020 Proceedings Weili Wu
https://textbookfull.com/product/combinatorial-optimization-and-
applications-14th-international-conference-cocoa-2020-dallas-tx-
usa-december-11-13-2020-proceedings-weili-wu/
https://textbookfull.com/product/information-systems-
security-14th-international-conference-iciss-2018-bangalore-
india-december-17-19-2018-proceedings-vinod-ganapathy/
https://textbookfull.com/product/neural-information-
processing-25th-international-conference-iconip-2018-siem-reap-
cambodia-december-13-16-2018-proceedings-part-ii-long-cheng/
https://textbookfull.com/product/neural-information-
processing-25th-international-conference-iconip-2018-siem-reap-
cambodia-december-13-16-2018-proceedings-part-iii-long-cheng/
https://textbookfull.com/product/neural-information-
processing-25th-international-conference-iconip-2018-siem-reap-
cambodia-december-13-16-2018-proceedings-part-v-long-cheng/
Shaojie Tang · Ding-Zhu Du
David Woodruff · Sergiy Butenko (Eds.)
Algorithmic Aspects
LNCS 11343
in Information
and Management
12th International Conference, AAIM 2018
Dallas, TX, USA, December 3–4, 2018
Proceedings
123
Lecture Notes in Computer Science 11343
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen
Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zurich, Zurich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology Madras, Chennai, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7407
Shaojie Tang Ding-Zhu Du
•
Algorithmic Aspects
in Information
and Management
12th International Conference, AAIM 2018
Dallas, TX, USA, December 3–4, 2018
Proceedings
123
Editors
Shaojie Tang David Woodruff
The University of Texas at Dallas University of California, Davis
Richardson, TX, USA Davis, CA, USA
Ding-Zhu Du Sergiy Butenko
University of Texas at Dallas Texas A&M University
Richardson, TX, USA College Station, TX, USA
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
General Chairs
Ding-Zhu Du University of Texas at Dallas, USA
David Woodruff University of California, Davis, USA
Program Chairs
Shaojie Tang University of Texas at Dallas, USA
Sergiy Butenko Texas A&M University, USA
1 Introduction
Given an undirected graph G = (V, E) and a subset T ⊆ V of terminals, a
Steiner tree is referred to as a connected subgraph of G that spans all the termi-
nals in T [6]. A Steiner forest is a group of disjoint connected branches spanning
T , where every branch is a Steiner tree [1,2,5,7,11–14]. They have a wide variety
of applications in communication networks, computational biology, and etc. [6].
2 Preliminaries
and
r(T , X ∗ ) = min r(T , X ). (2)
X ⊂P
and
diam(F) = max diam(Fj ). (4)
1≤j≤k
Since ASkCP and MDkSFP are trivial when p = 2, we focus on the cases
with p ≥ 3. On the other hand, the optimum is obvious (every branch contains
one terminal) when k = p. Therefore, we focus on the cases where k is (quite)
smaller than p in general.
Lemma 1. There are surely at least one facility x∗j ∗ ∈ X ∗ and at least two
terminals, ti∗1 and ti∗2 , such that d(ti∗1 , x∗j ∗ ) = d(ti∗2 , x∗j ∗ ) = r(T , X ∗ ).
3 Approximation to ASkCP
In this section, we study ASkCP in edge-weighted graph G = (V, E, w), and use
the dual approximation framework [9,10] to devise an approximation algorithm
with a factor of 2. Let opt be the optimal value (of k-radius), X ∗ and X A denote
an optimal solution and an algorithm solution with no duplicate, respectively.
Fig. 2. Two possible values of k-radius for given ti1 , ti2 and edge e = {v , v }.
Minimum Diameter k-Steiner Forest 7
Obviously, TEST1 (lλ ) returns YES. Our task is to find the smallest one, ls∗ ,
from all the possible values ls , 1 ≤ s ≤ λ which make TEST1 (ls ) return YES.
In Step 2, we find ls∗ by applying a binary search to ls , 1 ≤ s ≤ λ. So, it is
sufficient to apply TEST1 O(log λ) times. Recall that applying TEST1 once takes
at most O(pk) time. Since λ ≤ mp(p − 1) and p ≤ n ≤ m, Step 2 takes at most
O(pk log m) time. Therefore, ASkCP-ALG takes at most O(mp2 log m) time. The
details of Step 2 are presented as follows. Our search begins with l1 . If TEST1 (l1 )
outputs YES, then we claim that ls∗ = l1 and get an approximation X A = X (l1 ).
Otherwise, we find ls∗ by applying a binary search to l1 , l2 , . . . , lλ . Let LB and
UB be the lower bound and upper bound on the index s∗ , respectively, and let
M = LB+UB 2 . Initially, set LB = 1 and UB = λ. If TEST1 (lM ) outputs NO,
then M becomes a new lower bound on s∗ and UB is also its upper bound. If
TEST1 (lM ) outputs YES, then M becomes a new upper bound on s∗ and LB is
also its lower bound. Above operation is repeated until LB and UB become two
consecutive integer numbers. During the whole operation, TEST1 (lLB ) always
outputs NO while TEST1 (lUB ) always outputs YES. As a result, the final UB
is s∗ . The binary search ends with an approximation, X A = X (ls∗ ), to ASkCP
in a vertex-unweighted graph G. According to above discussions, we obtain the
following theorem.
Algorithm ASkCP-ALG:
Input: a graph G = (V, E, w) and distance matrix M ,
a set T ⊂ V of terminals;
Output: a k-set X A .
Step 1: Delete duplicates of all the possible values of
k-radius, and then arrange the values left into
an increasing sequence, l1 < l2 < · · · < lλ .
Step 2: if TEST1 (l1 ) = YES then Return X (l1 );
else LB ← 1; UB ← λ;
while UB − LB = 1 do
M ← LB+UB2
;
if TEST1 (lM ) = NO then LB ← M;
else UB ← M; endif
end Return X (lUB );
endif
4 Approximations to MDkSFP
In this section, we design a 2-approximation algorithm for MDkSFP, and further
develop an improved approximation algorithm with a factor of 2ρ, where ρ < 1
in general, based on the approximation to ASkCP.
Let FjA be the SPT (branch) spanning CjA with ti as the origin. By Theo-
rem 1 and the definition of CjA , we claim that all FjA , ∀j are both edge-disjoint
k
and vertex-disjoint. Hence, the collection of j=1 FjA forms a Steiner forest
(algorithm solution), denoted by F A . This leads to an approximation algorithm
MDkSFP-ALG for MDkSFP. Its performance analysis is shown in Theorem 5. Let
ji be the index such that d(ti , tji ) = min1≤j≤k d(ti , tj ), for any 1 ≤ i ≤ p.
Algorithm MDkSFP-ALG:
Input: a graph G = (V, E, w) and distance matrix M ,
a set T ⊂ V of terminals;
Output: a Steiner forest F A .
Step 1: Call ASkCP-ALG to get X A = {t1 , t2 , . . . , tk };
CjA ← ∅, 1 ≤ j ≤ k;
for i := 1 to p do
ji ← arg min1≤j≤k d(ti , tj );
CjAi ← CjAi ∪ {i};
end
Step 2: for j := 1 to k do
Compute a SPT (branch) FjA spanning CjA ;
end
F A ← kj=1 FjA ; Return F A ;
Minimum Diameter k-Steiner Forest 9
The center of CjA is perturbed from tj to xj , for all 1 ≤ j ≤ k, resulting
in a better solution to ASkCP, denoted by X B = {x1 , x2 , . . . , xk }. It is certain
that tj is closest to all the terminals in CjA , but it is uncertain that xj is closest
to all of them. Therefore, if the MDST (a.k.a., the SPT with xj as the origin)
spanning CjA is taken as one branch, then two different branches may have at
least a common edge. In other words, the resulting solution may be a pseudo
Steiner forest. Let FjP denote the MDST spanning CjA , and F P denote the union
of all FjP , ∀j.
In order to avoid a pseudo solution, we re-cluster all the terminals with X B
as centers. Let CjB be the cluster of terminals to which xj is the closest facility,
and let r(CjB ) denote the maximum distance to xj in CjB . The SPT spanning
CjB with xj as the origin is taken as one branch, denoted by FjB . By Theorem 1,
we conclude from the definition of CjB that all FjB , 1 ≤ j ≤ k are both edge-
disjoint and vertex-disjoint. As a consequence, all the branches FjB , ∀j form a
Steiner forest, denoted by F B . Based on above discussions, we devise an improved
approximation algorithm MDkSFP-IMP for MDkSFP. Its performance analysis
is shown in Theorem 6. For any 1 ≤ s ≤ p, we let js be the index such that
d(ts , xjs ) = min1≤j≤k d(ts , xj ).
10 W. Ding and K. Qiu
Algorithm MDkSFP-IMP:
Input: a graph G = (V, E, w) and distance matrix M ,
a set T ⊂ V of terminals;
Output: a Steiner forest F B .
Step 1: The same as MDkSFP-ALG;
Step 2: for j := 1 to k do
Use the Ding and Qiu’s algorithm in [4] to
compute the AS1C, xj , of CjA ;
end
CjB ← ∅, 1 ≤ j ≤ k;
for s := 1 to p do
js ← arg min1≤j≤k d(ts , xj );
CjBs ← CjBs ∪ {s};
end
Step 3: for j := 1 to k do
Compute a SPT (branch) FjB spanning CjB ;
end
F B ← kj=1 FjB ; Return F B ;
diam(F B )
ρ= . (7)
diam(F A )
In most cases, ρ < 1. This is because ρ = 1 if and only if the center (point)
corresponding to the k-radius is at a terminal (vertex). Therefore, the approxi-
mation of MDkSFP-IMP improves that of MDkSFP-ALG by a factor of 2(1 − ρ).
5 Conclusions
This paper studied MDkSFP in undirected graphs and established the rela-
tionship between MDkSFP and ASkCP. First, we obtained an approximate
ASkC and then designed a 2-approximation algorithm for MDkSFP. Further,
we achieved a better approximate ASkC by modification, and then developed
Minimum Diameter k-Steiner Forest 11
References
1. Agrawal, A., Klein, P., Ravi, R.: When trees collide: an approximation algorithm
for the generalized Steiner problem in networks. SIAM J. Comput. 24(3), 440–456
(1995)
2. Bateni, M., Hajiaghayi, M., Marx, D.: Approximation schemes for Steiner forest
on planar graphs and graphs of bounded treewidth. J. ACM 58(5), 1–21 (2011)
3. Bui, M., Butelle, F., Lavault, C.: A distributed algorithm for constructing a mini-
mum diameter spanning tree. J. Parallel Distrib. Comput. 64(5), 571–577 (2004)
4. Ding, W., Qiu, K.: Algorithms for the minimum diameter terminal Steiner tree
problem. J. Comb. Optim. 28(4), 837–853 (2014)
5. Dinitz, M., Kortsarz, G., Nutov, Z.: Improved approximation algorithm for Steiner
k-forest with nearly uniform weights. In: 17th APPROX, pp. 115–127 (2014)
6. Du, D., Hu, X.: Steiner Tree Problems in Computer Communication Networks.
World Scientific Publishing Co., Pte. Ltd., Singapore (2008)
7. Hajiaghayi, M., Jain, K.: The prize-collecting generalized steiner tree problem via
a new approach of primal-dual schema. In: Proceedings of the 17th SODA, pp.
631–640 (2006)
8. Hassin, R., Tamir, A.: On the minimum diameter spanning tree problem. Inf.
Process. Lett. 53, 109–111 (1995)
9. Hochbaum, D.S., Shmoys, D.B.: A best possible heuristic for the k-center problem.
Math. Oper. Res. 10(2), 180–184 (1985)
10. Plesnı́k, J.: A heuristic for the p-center problem in graphs. Discrete Appl. Math.
17, 263–268 (1987)
11. Ravi, R.: A primal-dual approximation algorithm for the Steiner forest problem.
Inf. Process. Lett. 50(4), 185–189 (1994)
12. Segev, D., Segev, G.: Approximate k-Steiner forests via the Lagrangian relaxation
technique with internal preprocessing. Algorithmica 56, 529–549 (2010)
13. Winter, P.: Generalized Steiner problem in outerplanar graphs. BIT Numer. Math.
25(3), 485–496 (1985)
14. Winter, P.: Generalized Steiner problem in series-parallel networks. J. Algorithms
7(4), 549–566 (1986)
Factors Impacting the Label Denoising
of Neural Relation Extraction
1 Introduction
Relation extraction is an active research task in information extraction [7] and
natural language processing, which aims to predict the relation between two
entities mentioned in plain text. It can benefit a variety of artificial intelligence
tasks such as question answering [1], information search [16] and knowledge base
construction [22]. However, supervised learning for relation extraction requires
a large amount of annotated data, which is costly to obtain.
Therefore, a distant supervision [12] strategy for relation extraction is pro-
posed to automatically label large-scale training data by heuristically aligning
entity pairs in plain text with the Knowledge Base (KB). However, distant
supervision tends to have the noisy labeling problem due to the incompletion
c Springer Nature Switzerland AG 2018
S. Tang et al. (Eds.): AAIM 2018, LNCS 11343, pp. 12–23, 2018.
https://doi.org/10.1007/978-3-030-04618-7_2
Factors Impacting the Label Denoising of Neural Relation Extraction 13
of the exploited KB. As shown in Fig. 1, for a relational triple born in(Donald
Trump,United States) in KB, all the sentences including two entities Donald
Trump and United States are labeled as the instances of relation born in, despite
the sentence “Donald Trump is the 45th president of the United States” fails to
express the relation. A similar noisy labeling problem exists in the entity pair
(Donald Trump,New York ), thus it is a great challenge for distantly supervised
relation extraction.
To handle the noisy labeling issue, many researchers have made great efforts
to distantly supervised relation extraction. Recently, deep learning and neural
networks have achieved an excellent performance on this task. Zeng et al. [19]
selected the most likely sentence for each entity pair as a valid instance to train
the neural model. Then Lin et al. [10] selected multiple valid sentences to predict
the relation by assigning higher weight for more informative sentences. Subse-
quently, Liu et al. [11] used a soft label as the ground-truth to correct the noisy
labels, which directly achieved label denoising. Further, we also investigate some
common methods that may affect the label denoising in relation extraction such
as word-level prior knowledge [3,21] and class imbalance [6,9].
On the whole, existing works focus on specific algorithms but fail to explore
the factors impacting the label denoising of neural relation extraction. Hence, we
did a lot of research and experiments and found there are some commonalities
between different denoising methods. In summary, we propose three fundamental
and influential factors as follows:
(1) Labeling assumption. We mainly consider two labeling assumptions: one-
valid-instance assumption and multiple-valid-instance assumption. Specifi-
cally, one-valid-instance assumption [18,19] indicates that we only choose the
most effective sentence within all the sentences that mention an entity pair
to predict the relation. Multiple-valid-instance assumption [4,10] assumes
that one-valid-instance assumption may lose a lot of information from the
neglected sentences, thus we select multiple informative sentences to benefit
the relation prediction. Based on these assumptions, we can achieve sentence
denoising at different levels with respect to the distantly supervised labels.
14 T. Sun et al.
(2) Prior knowledge. We regard prior knowledge as all the information about
the target in training instances, which are objective facts and independent of
the knowledge base. Generally speaking, distant supervision mainly depends
on informative prior knowledge in plain text. For instance, as shown in Fig. 1,
in addition to entity mentions Donald Trump, United States and New York,
the contextual words born and president are more important to relation
extraction than the other words such as the and was. There are a variety
of methods to capture influential prior knowledge, including feature engi-
neering [2,13,14], neural networks [4,10,19], attention-based models [3,21].
Thus, how to effectively obtain informative prior knowledge is a challenging
factor for relation extraction.
(3) Confidence level. We define two confidence levels – complete confidence
and partial confidence in Distantly Supervised (DS) labels. With respect
to complete confidence, we treat DS labels as the gold labels of relation
extractor, ignoring the effect of false negatives and false positives. As for
partial confidence, we hope to learn a new soft label [11] as the ground-truth
by partly trusting the DS labels. At this time, we will exploit the semantic
information obtained from the correct relational facts. For example, as shown
in Fig. 1, the sentence “Donald Trump was born in New York ” will be labeled
as the instance of relation Place lived between two entities Donald Trump
and New York because the triple exists in KB, even it fails to describe the
relation. However, if we choose partial confidence in DS labels, we can correct
the noisy labels according to the similar semantic patterns (blue fonts) in
our corpora, and find the right relation Place of birth. Therefore, choosing
different confidence levels for DS labels is an important factor that impacts
the denoising work.
The main contributions of this paper are to:
(1) propose three main factors that impact the label denoising of distantly super-
vised relation extraction, including labeling assumption, prior knowledge and
confidence level.
(2) analyze these factors in a unified neural framework with word, sentence and
label denoising modules. Through the combination of different methods in
each module, we can test the denoising effectiveness of the proposed factors.
(3) conduct experiments to evaluate and compare these factors with ten neural
schemes, discuss the typical cases and discover that the important word-level
prior knowledge and partial confidence for distantly supervised labels can
remarkably improve the denoising effect.
The remainder of this paper is structured as follows. Section 2 describes our
proposed methodology in detail. Section 3 reports our experimental results and
analysis. Section 4 gives a conclusion of the whole paper.
2 Methodology
As shown in Fig. 2, in order to analyze the denoising impact of our proposed fac-
tors for relation extraction, we design a unified neural model and employ some
Factors Impacting the Label Denoising of Neural Relation Extraction 15
typical methods to model the factors. With respect to two labeling assump-
tions, we use the corresponding labeling schemes to reduce the influence of noisy
sentences for each entity pair. To evaluate the prior knowledge factor, we also
use two methods to verify the word-level denoising performance. As for differ-
ent confidence levels, we respectively apply DS label (complete confidence) and
soft label (partial confidence) as the ground-truth of relation extractor. Besides,
based on the partial confidence idea, we propose a dynamic soft-label (DySoft)
method to gradually learn the corrected labels during training. By combining
three modules – word, sentence and label denoising, we build a neural relation
extraction framework in an end-to-end training way.
exp(ui )
αi = (1)
k exp(uk )
ui = Ww tanh([wi , e1 , e2 ]) + bw (2)
where wi , e1 and e2 are the embeddings of the i-th word, entity pair e1 and e2
in the sentence. Ww and bw are training parameters. Here we follow the work
of Huang et al. [3] to utilize the attention mechanism.
where p(τ ) represents the relation score predicted by relation extraction model
at τ -th training step. The initial label confidence c(0) is fed with one-hot vector
y of the DS label y.
We simply learn the label confidence of the current training step by a linear
combination of previous label confidence and current relation score. The iteration
operation can model the correlation between label confidences at different train-
ing steps, and use preceding obtained soft labels to learn current new soft labels.
Factors Impacting the Label Denoising of Neural Relation Extraction 17
3 Experiments
Our proposed factors are compared and evaluated on a widely used distantly
supervised relation extraction benchmark dataset, which was developed by
Riedel et al. [13] via aligning the plain text of the New York Times corpus with
relational triples of Freebase. The dataset has 52 relation classes and a non-
relation NA class. The training set includes 522,611 sentences, 281,270 entity
pairs and 18,252 relational facts. The test set includes 172,448 sentences, 96,678
entity pairs and 1,950 relational facts. Similar to previous work [5,10,11,19],
we adopt the held-out evaluation with aggregate precision-recall curves and top
N Precision (P@N). The held-out evaluation compares the predicted relations
of entity pairs with the distantly supervised relations without requiring human
evaluation. As for parameter settings, we follow the settings of Liu et al. [11].
Specially, our proposed soft label is utilized after 3 training epochs.
From Fig. 4, we present the impact of different denoising modules. For both
ONE and MUL assumptions, the Soft-label [11] method performs well when the
recall is less than 0.3, but it declines notably when the recall is greater than 0.4.
Our proposed DySoft-label method performs stable and achieves higher precision
in the MUL assumption. In particular, combining word denoising (WA) with
label denoising (Soft-label) module fails to improve the baseline. Consider that
correcting labels may have a bad impact on the attention of prior knowledge,
we get that the combination effect of WA+DySoft scheme may be not as good
as we expect.
corresponds to the area under the precision-recall curve. Compared with the
neural baselines PCNN+ONE and PCNN+MUL which use DS labels as the
ground-truth, We can observe that adding WA or adding DySoft can achieve
the better performance than the other neural schemes. Next we will analyze the
effect of each factor in detail.
Table 3. Top N precision (P@N) for relation extraction in the entity pairs with different
number of sentences.
results of Liu et al. [10]. The average precision for different settings of multiple-
valid-instance assumption has 1% to 4% improvements compared with one-valid-
instance assumption, which is also less than the gap of 5% to 9% in previous
work [10]. Specially, the case study of labeling assumption for sentence-level
denoising has been reported [10], thus we omit it in our paper.
Since the denoising impact of the prior knowledge factor in neural networks is
usually invisible and directly reflected in model performance, we evaluate this
factor in the precision-recall curve and visualization of word attention. From
Fig. 4 and Table 2, we can find that adding a word-level denoising WA mechanism
can greatly benefit both ONE and MUL assumptions. Particularly, MUL+WA
model achieves the best AP performance compared with all the other models.
As shown in Table 2, combining WA mechanism makes the average precision
improve 3% and 4% for ONE and MUL baselines. These results demonstrate the
effectiveness of adding the word-level importance for the prior knowledge factor.
Relation Instance
place of birth Roscoe Lee Browne was born on May 2, 1925, in Woodbury .
place of death His most recent book is “Murder in Amsterdam: the death of
Theo Van Gogh and the limits of tolerance.”
company “I like Eli Manning,” the NBC football analyst John
Madden said in a conference call Thursday.
founder Still , some early investors like Peter Munk , the founder and
chairman of Barrick Gold, were skeptical at first.
Table 4 presents the visualization of word attention mechanism, the bold font
and size of a word denote its importance, the italic words are entity mentions.
For example, with respect to the instance of relation place of birth, the word born
Another random document with
no related content on Scribd:
“Fear not a constraining measure!
—Yielding to the gentle spell,
Lucida! from domes of pleasure,
Or from cottage-sprinkled dell,
Comes to regions solitary,
Where the eagle builds her aery,
Above the hermit’s long-forsaken cell!
—She comes!—behold
That figure, like a ship with silver sail!
Nearer she draws; a breeze uplifts her veil;
Upon her coming wait
As pure a sunshine, and as soft a gale,
As e’er, on herbage-covering earthly mold,
Tempted the bird of Juno to unfold
His richest splendour,—when his veering gait,
And every motion of his starry train,
Seem governed by a strain
Of music, audible to him alone.”
The next lady that he invokes before the astonished youth is his own
daughter—sweet Dora—the previous one was Miss Southey.
“Come, if the notes thine ear may pierce,
Come, youngest of the lovely three,
Submissive to the mighty verse
And the dear voice of harmony,
By none more deeply felt than thee!
I sang; and lo! from pastures virginal
She hastens to the haunts
Of Nature, and the lonely elements.
Air sparkles round her with a dazzling sheen;
And mark, her glowing cheek, her vesture green!
And, as if wishful to disarm
Or to repay the potent charm,
She bears the stringed lute of old romance,
That cheered the trellissed arbour’s privacy,
And soothed war-wearied knights in raftered hall.
How vivid, yet how delicate, her glee!
So tripped the muse, inventress of the dance;
So, truant in waste woods, the blithe Euphrosyne!”
Then follows that beautiful description of her moral graces, already quoted
in these pages, beginning—
A fairer subject than this, for the imagination of the true painter, does
scarcely exist in poetry. The gorgeous magnificence of Miss Southey—the
wild, bird-like nature of Dora, the mystic, spiritual, meditative beauty of
Miss Coleridge. Here is material enough for the highest effort of art.
A number of poems followed this exquisite “Triad”—viz., “The Wishing
Gate,” in 1828—“The Lawn,” “Presentiments,” “The Primrose on the
Rock,” “Devotional Incitements;” these last were written between 1828 and
’32. A number of gold and silver fishes presented to the poet by Miss H. J.
Jewsbury, who subsequently died of the cholera in India,—and afterwards
removed to the pond already alluded to, under the oak in “Dora’s field,”
suggested the verses “Gold and Silver Fishes in a Vase,” and likewise
“Liberty,” and “Humanity;” “The Poet and the caged Turtle Dove” was
likewise suggested by real circumstances. Miss Jewsbury had given Dora a
pair of these beautiful birds; one of them was killed by an un-necessary cat,
and not a “harmless one;” the other survived many years, and had a habit of
cooing the moment Wordsworth began “booing” his poems, as the country
people called it.
Wordsworth gives an amusing account of a visit which he paid about this
time to “Chatsworth.” He had undertaken to ride his daughter’s pony from
Westmoreland to Cambridge, that she might have the use of it during a visit
she was about to make to her uncle at Trinity; and on his way from
Bakewell to Matlock, he turned off to see the splendid mansion of the great
Duke of Devonshire. By-and-bye a tremendous storm came on, and the poet
was drenched through to the very skin, whilst the pony, to make his rider’s
seat the more easy, went “slantwise” all the way to Derby. Notwithstanding
this, however, and the pelting of the pitiless storm, Wordsworth managed to
hold sweet and sad converse with his muse, and composed his “Lines to the
Memory of Sir George Beaumont,” who died 7th February, 1827. It is a
picture which we cannot readily forget, and shows how completely the poet
was master of himself. Sir George Beaumont and his lady were friends and
benefactors of Wordsworth—he loved them both intensely. Walking
through the grounds and gardens of Coleorton with Sir George—the
successor in the Baronetcy to his friend—and after the death of Lady
Beaumont, which took place in 1822, he comes suddenly to her ladyship’s
grotto, near the fountain, and is overwhelmed with his feelings, and the
recollection of the dead, and the happy memories which rush over his mind
in connection with this place, so that he cannot speak for tears. On his
return home he wrote the elegiac musings, already mentioned in these
memoirs, which are full of love, and the sanctity of a sweet sorrow. In the
same year (1831) were composed “The Armenian Lady’s Love,” “The
Egyptian Maid,” and “The Russian Fugitive,” poems in which all the
beauties of language are pressed, along with the simplicity which marks the
old English ballads. Lines on his portrait, painted by Pickersgill, and
preserved with sacred veneration in St. John’s College, Cambridge, were
likewise written in this year, as well as the inscription already quoted, for
the stone at Rydal.
Besides these poetical compositions, however, Wordsworth interested
himself in public affairs; and having fixed principles of political and social
economy in his own mind, regarded all public measures at variance with
them, as fatal errors, and subversive in their consequences of the highest
human concerns. In 1806, he wrote a letter to a friend, who had consulted
him respecting the education of his daughter—in which he gives some
sound and excellent advice respecting the training and development of
youthful minds. For Wordsworth had at an early period devoted his
attention to the subject of education, and had his own views respecting it—
views which were marked by the spiritual peculiarity of his mind. When he
wrote “The Excursion,” he seems to have had the highest hopes for man,
when education should become universal; and insisted that the State should
teach those to obey, from whom she exacted allegiance:—
Wordsworth does not seem, during any period of his life, to have been on
intimate terms with any of his contemporaries. He preferred the flower of
the literateurs, Coleridge, Scott, Southey; and these, with the exception
perhaps of Rogers, were his chief friends. We have letters of his, however,
to much smaller fry; to Mrs. Hemans, and Miss Jewsbury, for example—
and to sundry editors of other men’s wares; but there is little or no
recognition of Byron, Shelly, Keats, Tennyson, Baily, Campbell, Moore, nor
yet of Dickens or Bulwer. His letters represent his character even better than
his poetry; they are Wordsworth in undress, without the “garland and
singing robe,” and are worthy to be studied. I like much what he says to the
Rev. Robert Montgomery, author of “The Devil and Father Luther,” and
pious Robert would do well even at this late day to think on it. Montgomery
had sent Wordsworth a copy of his poems, and in reply, the poet answers: “I
cannot conclude without one word of literary advice which I hope you will
deem my advanced age entitles me to give. Do not, my dear sir, be anxious
about any individual’s opinion concerning your writings, however highly
you may think of his genius, or rate his judgment. Be a severe critic to
yourself; and depend upon it no person’s decision upon the merit of your
works will bear comparison in point of value with your own. You must be
conscious from what feeling they have flowed, and how far they may or
may not be allowed to claim on that account, permanent respect; and above
all I would remind you, with a view to tranquillise and steady your mind,
that no man takes the trouble of surveying and pondering another’s writings
with a hundredth part of the care which an author of sense and genius will
have bestowed upon his own. Add to this reflection another, which I press
upon you, as it has supported me through life—viz.: That posterity will
settle all accounts justly, and that works which deserve to last will last; and
if undeserving this fate, the sooner they perish the better.”
In the year 1836 the sister of the poet’s wife—Miss Sarah Hutchinson,
who had resided with the family at Rydal, died, and was buried in Grasmere
church, “near the graves of two young children removed from a family to
which through life she was devoted.”
In the following year, 1837, Wordsworth, accompanied by his friend H.
C. Robinson, Esq., set off from London for Rome, returning in August. The
“Itinerary” of the travellers is contained in the “Memoirs,” along with some
memoranda by the poet; but they are not of much interest. Many fine
pieces, however, sprung as usual from the journey, as well as a goodly
number of sonnets. They originally appeared in a volume entitled “Poems,
chiefly of Early and Late Years,” in 1842. In 1839, Wordsworth received
the degree of D.C.L., from the University of Oxford, which was conferred
on him in the Sheldonian Theatre, amidst shouts of rejoicing such as had
never before been heard in that city, except upon the occasion of an
unexpected visit of the Duke of Wellington. In 1838, Wordsworth prepared
a new edition of his poems, to be published by Moxon, and continued to
live at Rydal, in his quiet and musical manner, writing poems, taking
rambles, and conducting his correspondence until 1843, when he was
appointed Poet Laureate of England, Southey having died on the 21st of
March of that year, and the appointment having been offered to Wordsworth
on the 31st of the same month. One occurrence only broke the even tenor of
the poet’s life in the interim alluded to, and this was an accident by which
he was upset from his gig, and thrown violently into a plantation. The
accident was owing to the carelessness and want of skill in the driver of a
coach, which they met on the road. No serious consequences followed,
however, and inquiries and congratulations flowed in on all sides, from the
peasant up to Queen Adelaide.
From the time of Wordsworth’s appointment as Laureate,—which it
ought to be said he at first refused, and only accepted with the
understanding that it should be an honorary office,—he wrote very little
poetry. His work, indeed, was done, his mission accomplished; and his old
days were spent in rambling over the hills, and in the quiet enjoyment of his
family, friends, fame, and fortune. Honours of a high order were
subsequently heaped upon him. In the year 1838, the University of Durham
took the initiative in conferring an academic degree on the poet; then the
grand old Mother, Oxford, followed,—and in 1846 he was put in
nomination, without his knowledge, for the office of Lord Rector of the
University of Oxford, and gained a majority of twenty-one votes, in
opposition to the premier, Lord John Russell. “The forms of election,
however,” says Wordsworth, in a letter to Sir W. Gomm, of Port Louis,
Mauritius, dated November 23, 1846, “allowed Lord John Russell to be
returned through the single vote of the sub-rector voting for his superior. To
say the truth, I am glad of this result, being too advanced in life to
undertake with comfort any considerable public duty, and it might have
seemed ungracious to have declined the office.”
On the 20th of January, 1847, Mr. William Wordsworth, the younger son
of the poet, was married at Brighton, to Fanny Eliza Graham, youngest
daughter of Reginald Graham, Esq., of Brighton, who was a native of
Cumberland; and whilst the joy of this event was still fresh in the hearts of
the Rydal household, a dread calamity awaited them in the death of Mrs.
Quillinan—the sweet Dora so often spoken of in these pages, the beloved
daughter of the poet. As previously stated, she had accompanied her
husband to Portugal for the benefit of her health,—and although the change
seemed at first to have operated favourably upon her, it was soon evident,
on her return home, that she was doomed for the silent bourne of all
travellers in this world. She died on the 9th day of July, 1847, and was
buried in Grasmere church-yard. Her death was a terrible blow to the
venerable poet, now in his eightieth year,—but he bore up patiently, with
the heart and hope of a Christian.
Three years after this sad loss, Wordsworth himself was summoned
away. On Sunday, the 10th of March, 1850, he attended at Rydal chapel for
the last time, visiting, during the day, a poor old woman, who had once
been his servant, and another person who was sick, and as the poet said,
“never complained.”
“On the afternoon of the following day, he went towards Grasmere, to
meet his two nieces, who were coming from Town End. He called at the
cottage near the White Moss Quarry, and the occupant being within, he sat
down on the stone seat of the porch, to watch the setting sun. It was a cold,
bright day. His friend and neighbour, Mr. Roughsedge, came to drink tea at
Rydal, but Mr. Wordsworth not being well, went early to bed.”
From this time he gradually grew worse; and in order to convey to him
the impressions of his physicians, Mrs. Wordsworth whispered in a soft
voice, full of deep devotion, “Dear William, you are going to Dora.” How
delicate, how affectionate, how poetical! But the poet did not hear, or did
not seem to hear; and yet, twenty-four hours after, when one of his nieces
came into the room, and gently drew aside the curtains of his bed, he caught
a glimpse of her figure, and asked, “Is that Dora?”
On the 23rd of April—the birth-day, and death-day of Shakspeare, the
great-hearted Wordsworth went back again to God.
He was buried on the 27th, in Grasmere church-yard.
Those who would know more of the poet must go to his writings; and, I
may add, that the “Memoirs” of Dr. Wordsworth are indispensable to a full
understanding both of the Poet and the Man. His letters, containing his most
private thoughts, are printed there with plentiful profuseness; and the
“Memoranda” respecting the origin of his poems are intensely interesting
and important to all students of Wordsworth. The reminiscences of various
persons who knew him, set the character of the poet before us in strong
relief. All agree in speaking of him as a most kindly, affectionate, and
hospitable man, living with the simple tastes and manners of a patriarch, in
his beautiful home. My limits prevent me from entering into an analysis of
his mind and character, as I had intended to do; I must reserve this work,
therefore, for another occasion, and will conclude with a few quotations
from the poet’s “Table-Talk,” respecting his cotemporaries.—Speaking of
Goethe, he says:—
“He does not seem to me to be a great poet in either of the classes of
poets. At the head of the first I would place Homer and Shakspeare, whose
universal minds are able to reach every variety of thought and feeling,
without bringing his own individuality before the reader. They infuse, they
breathe life into every object they approach, but you cannot find themselves.
At the head of the second class, those whom you can trace individually in
all they write, I would place Spenser and Milton. In all that Spenser writes,
you can trace the gentle, affectionate spirit of the man; in all that Milton
writes, you find the exalted, sustained being that he was. Now, in what
Goethe writes, who aims to be of the first class, the universal, you find the
man himself, the artificial man, where he should not be found; so that I
consider him a very artificial writer, aiming to be universal, and yet
constantly exposing his individuality, which his character was not of a kind
to dignify. He had not sufficiently clear moral perceptions to make him
anything but an artificial writer.
And again:—
“I have tried to read Goethe. I never could succeed. Mr.—— refers me to
his ‘Iphigenia,’ but I there recognise none of the dignified simplicity, none
of the health and vigour which the heroes and heroines of antiquity possess
in the writings of Homer. The lines of Lucretius describing the immolation
of Iphigenia are worth the whole of Goethe’s long poem. Again there is a
profligacy, an inhuman sensuality, in his works, which is utterly revolting. I
am not intimately acquainted with them generally. But I take up my ground
on the first canto of ‘Wilhelm Meister;’ and as the attorney-general of
human nature, I there indict him for wantonly outraging the sympathies of
humanity. Theologians tell us of the degraded nature of man; and they tell
us what is true. Yet man is essentially a moral agent, and there is that
immortal and unextinguishable yearning for something pure and spiritual
which will plead against these poetical sensualists as long as man remains
what he is.”
Of Scott he says:—
“As a poet, Scott cannot live, for he has never in verse written anything
addressed to the immortal part of man. In making amusing stories in verse,
he will be superseded by some newer versifier; what he writes in the way of
natural description is merely rhyming nonsense. As a prose writer, Mr.
Wordsworth admitted that Scott had touched a higher vein, because there he
had really dealt with feeling and passion. As historical novels, professing to
give the manners of a past time, he did not attach much value to those
works of Scott’s, so called, because that he held to be an attempt in which
success was impossibility. This led to some remarks on historical writing,
from which it appeared that Mr. Wordsworth has small value for anything
but contemporary history. He laments that Dr. Arnold should have spent so
much of his time and powers in gathering up, and putting into imaginary
shape, the scattered fragments of the history of Rome.”
And again:—
“He discoursed at great length on Scott’s works. His poetry he
considered of that kind which will always be in demand, and that the supply
will always meet it, suited to the age. He does not consider that it in any
way goes below the surface of things; it does not reach to any intellectual or
spiritual emotion; it is altogether superficial, and he felt it himself to be so.
His descriptions are not true to nature; they are addressed to the ear, not to
the mind. He was a master of bodily movements in his battle-scenes; but
very little productive power was exerted in popular creations.”
Moore:—
“T. Moore has great natural genius; but he is too lavish of brilliant
ornament. His poems smell of the perfumer’s and milliner’s shops. He is
not content with a ring and a bracelet, but he must have rings in the ear,
rings on the nose—rings everywhere.”
Shelley:—
“Shelley is one of the best artists of us all: I mean in workmanship of
style.”
Tennyson:—
“I saw Tennyson, when I was in London, several times. He is decidedly
the first of our living poets, and I hope will live to give the world still better
things. You will be pleased to hear that he expressed in the strongest terms
his gratitude to my writings. To this I was far from indifferent, though
persuaded that he is not much in sympathy with what I should myself most
value in my attempts—viz., the spirituality with which I have endeavoured
to invest the material universe, and the moral relations under which I have
wished to exhibit its most ordinary appearances.”
Hartley Coleridge—
He spoke of with affection. “There is a single line,” he added, “in one of
his father’s poems, which I consider explains the after life of the son. He is
speaking of his own confinement in London, and then says,—
FINIS.
——————
J. S. Pratt, Stokesley, Yorkshire.
FOOTNOTES:
[A] These remarks do not of course apply to Cowper and Burns, to whom our
modern literature is so deeply indebted, but to their predecessors, from Pope
downwards.
[B] Memoirs of William Wordsworth, by Dr. Wordsworth, vol. 1, page 7.
[C] De Quincy, Tait’s Magazine, for 1839.
[D] It is related by De Quincy, that during Wordsworth’s early residence in the
lake country—after his return from Cambridge—his mind was so oppressed by the
gloomy aspect of his fortunes, that evening card-playing was resorted to, to divert
him from actual despondency.
[E] Dr. Wordsworth’s Memoir, page 53.
[F] Memoir, page 71-2.
[G] Chambers’ Papers for the People, article Wordsworth.
[H] Vol. 1, page 149 to 154.
[I] Memoirs, Vol. 1., page 156
[J] Dr. Wordsworth’s “Memoirs,” vol. 2, p. 94.
[K] “Memoirs,” vol. 2, p. 102-105.
[L] Memoirs, vol. ii., p. 121.
[M] This view of Coleridge is confirmed by Carlyle, in his “Life of John Sterling,”
just published.
“Coleridge sat on the brow of Highgate-hill, in those years, looking down on
London and its smoke tumult, like a sage escaped from the inanity of life’s battle;
attracting towards him the thoughts of innumerable brave souls still engaged there.
His express contributions to poetry, philosophy, or any specific province of human
literature or enlightenment, had been small and sadly intermittent; but he had,
especially among young inquiring men, a higher than literary,—a kind of prophetic,
or magician character. He was thought to hold,—he alone in England,—the key of
German and other transcendentalisms; knew the sublime secret of believing by ‘the
reason’ what ‘the understanding’ had been obliged to fling out as incredible; and
could still, after Hume and Voltaire had done their best and worst with him, profess
himself an orthodox Christian, and say and point to the Church of England, with its
singular old rubrics and surplices at Allhallowtide, Esto perpetua. * * * * He
distinguished himself to all that ever heard him as at least the most surprising talker
extant in this world,—and to some small minority, (by no means to all,) the most
excellent. The good man, he was now getting old, towards sixty perhaps,—and gave
you the idea of a life that had been full of sufferings; a life heavy-laden, half-
vanquished, still swimming painfully in seas of manifold physical and other
bewilderment. * * * * * I still recollect his ‘object’ and ‘subject,’ terms of continual
recurrence in the Kantean province; and how he sung and snuffled them into ‘om-m-
mject’ ‘sum-m-mject,’ with a kind of solemn shake or quaver, as he rolled along. No
talk, in his century or in any other, could be more surprising.
* * * * *
“He had knowledge about many things and topics,—much curious reading; but
generally all topics led him, after a pass or two, into the high seas of theosophic
philosophy, the hazy infinitude of Kantean transcendentalism, with its ‘sum-m-
mjects’ and ‘om m-mjects.’ Sad enough, for with such indolent impatience of the
claims and ignorances of others, he had not the least talent for explaining this or
anything unknown to them; and you swam and fluttered in the mistiest, wide,
unintelligible deluge of things,—for most part in a rather profitless, uncomfortable
manner. Glorious islets, too, I have seen rise out of the haze; but they were few, and
soon swallowed in the general element again. * * * * * * One right peal of concrete
laughter at some convicted flesh-and-blood absurdity, one burst of noble indignation
at some injustice or depravity rubbing elbows with us on this solid earth,—how
strange would it have been in that Kantean haze-world, and how infinitely cheering
amid its vacant air-castles, and dim-melting ghosts and shadows! None such ever
came. His life had been an abstract thinking and dreaming, idealistic one, passed
amid the ghosts of defunct bodies and of unborn ones. The mourning sing-song of
that theosophico-metaphysical monotony left on you, at last, a very dreary feeling.”