Hidden Semi Markov Models Theory Algorithms and Applications 1St Edition Yu Full Chapter
Hidden Semi Markov Models Theory Algorithms and Applications 1St Edition Yu Full Chapter
Hidden Semi Markov Models Theory Algorithms and Applications 1St Edition Yu Full Chapter
Shun-Zheng Yu
No part of this publication may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, recording, or any information storage and
retrieval system, without permission in writing from the publisher. Details on how to seek
permission, further information about the Publisher’s permissions policies and our arrangements
with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency,
can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the
Publisher (other than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and
experience broaden our understanding, changes in research methods, professional practices, or
medical treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in
evaluating and using any information, methods, compounds, or experiments described herein. In
using such information or methods they should be mindful of their own safety and the safety of
others, including parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors,
assume any liability for any injury and/or damage to persons or property as a matter of products
liability, negligence or otherwise, or from any use or operation of any methods, products,
instructions, or ideas contained in the material herein.
ISBN: 978-0-12-802767-7
developments and emerging topics that have surfaced in this field need
to be summarized.
For these purposes, this book presents nine chapters in three parts.
In the first part, this book defines a unified model of HSMMs, and dis-
cusses the issues related to the general HSMM, which include:
In the second part, this book discusses the state duration distribu-
tions and the observation distributions, which can be nonparametric or
parametric depending on the specific preference of the applications.
Among the parametric distributions, the most popular ones are the
exponential family distributions, such as Poisson, exponential,
Gaussian, and gamma. A mixture of Gaussian distributions is also
widely used to express complex distributions.
Other than the exponential family and the mixed distributions, the
Coxian distribution of state duration can represent any discrete proba-
bility density function, and the underlying series parallel network also
reveals the structure of different HSMMs.
There usually exists a class of HSMMs that are specified for the
applications in an area. For example, in the area of speech synthesis,
speech features (observations to be obtained), instead of the model
parameters, are to be determined. In the area of human activity recog-
nition, unobserved activity (hidden state) is to be estimated. In the
area of network traffic characterization/anomaly detection, perfor-
mance/health of the entire network is to be evaluated. In the area of
fMRI/EEG/ECG signal analysis, neural activation is to be detected.
ACKNOWLEDGMENTS
for all iAS. That is, state i must transit to another state in the
time ½0; NÞ.
Ðτ P
where Wi ðτÞ 5 0 jAS hij ðτ 0 Þdτ 0 is the probability that the process
stays in state i for at most time τ before transiting to another state,
and 1 2 Wi ðτÞ is the probability that the process will not make transi-
tion from state i to any other state within time τ. The likelihood func-
tion corresponding to the sample path (x0 , τ 1 , x1 , . . ., τ n , xn ,
P
T 2 nk51 τ k ) is thus
" # !!
X
n X
n
L x0 ; τ 1 ; x1 ; . . .; τ n ; xn ; T 2 τ k 5 P½x0 U 1 2 Wxn T 2 τk
k51 k51
n
U L hxk21 xk ðτ k Þ:
k51
Suppose the current time is t. The time that has been passed since
last jump is defined by Rt 5 t 2 TNðtÞ . Then the process ðSt ; Rt Þ is a
continuous time homogeneous Markov process.
ÐN
aij P½Xn11 5 jjXn 5 i 5 0 hij ðτÞdτ is the transition probability
P
from state i to state j, s.t. jAS aij 5 1, and
and
aði;τÞj P½Xn11 5 jjXn 5 i; τ n11 5 τ
is the probability that the system will make the next transition to
state j, given time τ and current state i. In this model,
hij ðτÞ 5 wi ðτÞaði;τÞj :
3. The semi-Markov process can also be thought as such a process that
after having entered state i, it randomly draws the pair ðk; dik Þ for all
kAS, based on fik ðτÞ, and then determines the successor state and length
of time in state i from the smallest draw. That is, if dij 5 minkAS fdik g,
then the next transition is to state j and the length of time the process
holds in state i before going to state j is dij . In this model,
Ðτ
where Fik ðτÞ 5 0 fik ðτ 0 Þdτ 0 , and Lk6¼j ð1 2 Fik ðτÞÞ is the probability
that the process will not transit to another state except j by time τ.
This type of semi-Markov process is applied to such as reliability
analysis (Veeramany and Pandey, 2011). An example of this type of
semi-Markov process is as follows.
Introduction 5
Example 1.1
Suppose a multiple-queue system contains L queues, each with known
inter-arrival time distribution and departure time distribution. Let ql
be the length of the lth queue and define the state at time t by
St 5 ðq1 ; . . .; qL Þ. Then every external arrival to a queue will increment the
queue length and every departure from a queue will decrement the queue
length if it is greater than zero. Therefore, each arrival/departure will
result in a transition of the system to a corresponding state. Denote an
external arrival to queue l by e1 l 5 ð0; . . .; 1; . . .; 0Þ and a departure from
queue l with ql . 0 by e2 l 5 ð0; . . .; 21; . . .; 0Þ. Then the next state is
ðq1 ; . . .; qL Þ 1 e1 2
l or ðq1 ; . . .; qL Þ 1 el for l 5 1, . . . ,L. The time to the next
state transition is determined by which e1 2
l or el , for l 5 1, . . . ,L, occurs
first, based on their inter-arrival/departure time distributions.
External
arrival
External Out of
arrival net
Dept.
Out of
net
External Dept.
arrival Out of
net
External
arrival
transition depends on both the current state and the next event. The
next transition epoch Tn11 is determined by the event En11 AEðiÞ that
occurs first among all events of EðiÞ. That is,
where Te0 ;next denotes the time epoch that event e0 will occur. In consid-
ering that each event has an inter-event time distribution, the inter-event
time for event e has passed ce;n at time epoch Tn since event e lastly
occurred. Suppose Te;last and Te;next , for Te;last # Tn , Te;next , are the
epochs that event e lastly occurred and will appear at. The inter-event
time for event e is thus ye;n 5 Te;next 2 Te;last . Then,
and
P½Te0 ;next 2 Tn . τjXn 5 i; En11 5 e0
P½ce0 ;n 1 τ , ye0 ;n jXn 5 i; En11 5 e0
5 :
P½ ye0 ;n . ce0 ;n jXn 5 i; En11 5 e0
Introduction 7
hðmÞ
ij ðτÞ P½τ n1m 5 τ; Xn1m 5 jjXn 5 i
Example 1.2
There are two libraries. Readers can borrow books from and return to
any of them, as shown in Figure 1.2. The statistics of the libraries shows
that the transition mass functions are
Then we can get the probability that a book is borrowed from library
P
i and returned to library j by aij 5 Nτ51 hij ðτÞ. It yields that
where
αt ðjÞ P½St 5 j; o1:t jλ
is the forward variable defined as the joint probability of St 5 j and the
partial observed sequence o1:t , and
S0 S1 S2 … ST
Graphical model
o1 o2 oT
Observation seq o1 o2 oT
State 1 …
State 2 S0 …
.
Trellis
. . . S2 .
. . .
. S1
…
.
State M …
ST
a22
π1 a11 a a2M
⎡ a11 a12 L a1M 12
M
⎢a 2
π2 a22 a2M 1 a32
π= A = ⎢ 21 a13
M ⎢ M a34
M O M
πM ⎢ 3 4 5
⎣ aM1 aM2 L aMM a43 a44
Initial state Transition probability State transition
distribution matrix diagram
with the initial conditions α0 ðjÞ 5 πj and β T ðjÞ 5 1; jAS, where in order
P
to make Eqn (1.2) true for β T21 ðjÞ 5 P½oT jST21 5 j; λ 5 iAS aji bji ðoT Þ,
it is assumed β T ðjÞ 5 1.
for 0 # t # T.
A limitation of the MAP estimation is that the resulting state
sequence may not be a valid state sequence. This can be easily proved.
Let C be a matrix with the elements cij 5 αt ðiÞaij bij ðot11 Þβ t11 ðjÞ and
ci0 j0 5 0, that is, the transition from state i0 to j0 at time t is not valid. But
it is still possible that i0 5 arg maxi Ce and j 0 5 arg maxj eT C, that is,
those two states could be the best choice based on the MAP estimation,
where e is an all-ones vector and eT its transpose. However, this limita-
tion does not affect its successful application in vast areas including the
area of digital communications. A famous algorithm in the digital
communication areaPis the BCJR algorithm (Bahl et al., 1974), which
replaces bij ðot Þ with xt eij ðxt ÞRðot ; xt Þ in the forwardbackward formu-
las (1.1) and (1.2) for decoding, where xt is the output associated with a
state transition at time t, eij ðxt Þ 5 P½xt jSt21 5 i; St 5 j; λ is the output
probability, ot is the observation of xt and Rðot ; xt Þ 5 P½ot jxt is the
channel transition probability. The difference between the BaumWelch
algorithm and the BCJR algorithm is shown in Figure 1.4.
Similar to the derivation of the forward formula for the HMM, the
Viterbi algorithm can be readily derived. Define
δt ðjÞ max P½S1:t21 ; St 5 j; o1:t jλ:
S1:t21
Then from the similarity we can see that by replacing the sumproduct
of Eqn (1.1) with the max-product the Viterbi algorithm of HMM is
yielded by:
δt ðjÞ 5 maxfδt21 ðiÞaij bij ðot Þg; 1 # t # T; jAS;
iAS
Introduction 13
Ot
Markov Detector
source
P[ot | xt ]
Markov xt Ot
Channel Detector
source
Noise
(b) BCJR model
PaðUðkÞ ðkÞ
t Þ 5 fPaðXt Þ : Xt AUt g:
Ut(1)
−1
U(2)
t −1
(K )
Ut −1 Ut
(1) (2)
Ut
(K )
Ut
(1)
Ut +1
(2)
Ut +1 Ut(K)
+1
Intra-slice
Cut-set of arcs
Inter-slice
the original DBN and each “arc” denotes that the child subset has at
least one parent in the parent subset. The solid arrows represent the
intra-slice arcs and the dotted arrows the inter-slice arcs.
where Pat21 ðut Þ are the set of boundary nodes going out of slice t 2 1
to t.
Denote UðEÞ
t as the set of entry nodes from slice t 2 1 into slice t by
letting
UðEÞ
t fXt : Pat21 ðXt Þ 6¼ nullg:
Therefore,
" #
K
[
P½UðIt Þ jUðEÞ
t 5P UðIt Þ - UðkÞ
t jUðEÞ
t
k51
h i
5 P UðIt Þ - Uð1Þ
t jU ðEÞ
t U
" #
[K
P UðIÞ
t - Ut
ðkÞ
jUðEÞ ðI Þ ð1Þ
t , Ut - Ut
k52
...
" #
K X
k21
5 LP UðIÞ
t - UðkÞ ðEÞ
t jUt , UðIÞ
t - UðlÞ
t
k51 l51
K
5L L P½Xt jPaðXt Þ
k51 Xt AUðI Þ - UðkÞ
t t
X K21 X
5 L L P½xt jPaðxt Þ L P½x0t jPaðx0t Þ
uðI Þ ðKÞ
t \ut
k51 xt AuðI Þ
t -ut
ðkÞ
uðKÞ
t
x0t AuðI Þ
t -ut
ðKÞ
X K21
5 L L P½xt jPaðxt Þ
k51 xt AuðI Þ -uðkÞ
uðI Þ ðKÞ
t \ut t t
...
51:
As shown in Figure 1.5, any cut-set of arcs divides the DBN into a
left part and a right part. If the starting/ending nodes of the arcs in the
cut-set are given, such as given uðDÞ ðEÞ
t21 or ut , the transition probabilities
to the nodes in the right part can be completely determined. For
ðEÞ
instance, for given the transition probabilities P½uðIt Þ juðEÞ
P ut , ðIÞ t can be
ðEÞ
determined and uðI Þ P½ut jut 5 1. Based on this observation, we can
t
try to find a cut-set of arcs, which can be inter-slice, intra-slice, or
mixed ones, so that they have the minimum number of starting/ending
nodes. Let UðMinÞ
t denote the set of nodes in one ends of the cut-set of
arcs. Then P½UðMinÞ
t may have the minimum number of dimensions.
then the DBN becomes a HMM. Similarly, the forward variables can
be defined with reduced-dimensions by
αt ðsðEÞ ðEÞ
t Þ P½o1:t ; st
X
5 αt21 ðsðEÞ ðIÞ ðEÞ ðEÞ ðI Þ ðDÞ
t21 ÞP½st21 jst21 P½st ; st jst21 P½ot jPaðot Þ
sðEÞ ;sðI Þ ;sðI Þ
t21 t21 t
X
5 αt21 ðsðEÞ ðIÞ ðEÞ ðEÞ ðDÞ ðI Þ ðEÞ
t21 ÞP½st21 jst21 P½st jst21 P½st jst P½ot jPaðot Þ:
sðEÞ ;sðI Þ ;sðI Þ
t21 t21 t
(1.7)
Compared with Eqns (1.4) and (1.5), we can see that in the discrete
cases the computational complexity of Eqns (1.6) and (1.7) and the
number of model parameters for an HMM are consistent to the
Introduction 19
where se and sv are the component sets of s associated with the vertices
of edge e and vertice v, respectively, the feature functions fk and gk0 are
given and fixed, and ZðoÞ is an instance-specific normalization function
defined by
!
X X X
ZðoÞ 5 exp λk fk ðe; se ; oÞ 1 μk0 gk0 ðv; sv ; oÞ : (1.9)
s eAE;k vAV ;k 0
where s1 is assumed being the initial state. Then the conditional probability
factorizes as
1 1 T
P½sjo 5 P½s; o 5 L P½st jst21 P½ot jst
P½o P½o t51
!
1 XT XT
5 exp log P½st jst21 1 log P½ot jst
P½o t51 t51
1 X T X
5 exp log aij UI ððst21 ; st Þ 5 ði; jÞÞ
P½o t51 i;j
!
XT X
1 log bj ðxÞUI ðst 5 j ÞUI ðot 5 xÞ ;
t51 j;x
1. The first state i1 and its duration d1 are selected according to the state
transition probability aði0 ;d0 Þði1 ;d1 Þ , where ði0 ; d0 Þ is the initial state and
duration. State i1 lasts for d1 5 2 time units in this instance.
2. It produces two observations ðo1 ; o2 Þ according to the emission
probability bi1 ;d1 ðo1 ; o2 Þ.
Introduction 23
Observations o1 o2 o3 o4 o5 o6 •••••• oT
Time 1 2 3 4 5 6 •••••• T
Duration d1 d2 •••••• dn
State seq i1 i2 •••••• in
Transitions
3. It transits, according to the state transition probability aði1 ;d1 Þði2 ;d2 Þ , to
state i2 with duration d2 .
4. State i2 lasts for d2 5 4 time units in this instance, which produces
four observations ðo3 ; o4 ; o5 ; o6 Þ according to the emission probabil-
ity bi2 ;d2 ðo3 ; o4 ; o5 ; o6 Þ.
5. ði2 ; d2 Þ then transits to ði3 ; d3 Þ, and ði3 ; d3 Þ transits to . . ., ðiN ; dN Þ until
the final observation oT is produced. The last state iN lasts for dN time
P
N
units, where dn 5 T and T is the total number of observations.
n51
Example 1.4
The observed workload data of a Web site consists of ot, the number of
user requests per second, with the maximum observed value
K 5 max{ot} 5 74 requests/s. The total number of observations is
T 5 3600 s (over 1 h) in the workload data set. The user request arrivals
are governed by an underlying hidden semi-Markov process. The hidden
state represents the arrival rate, which is corresponding to the number of
users that are browsing the website. For instance, state 1 corresponds to
arrival rate of 13 requests/s, state10, 30.1 requests/s, and state 20, 60
24 Hidden Semi-Markov Models
70
25
60
20
50
Requests/s
State
40 15
30
10
20
5
10
0
0 10 20 30 40 50 60
Time (minute)
Figure 1.7 Data (requests/s) and the hidden states of the workload.
The grey line is the trace of the workload data (number of arrivals per second) that we observed. The black line is
the hidden state sequence estimated. The hidden state represents the arrival rate, and the state duration represents
the dwell time of the arrival rate. For given arrival rate, the number of arrivals per second is a random variable.
requests/s. For given state or arrival rate, the number of user requests per
second, Ot, is a random variable. Therefore, from an observed value one
cannot tell the actual arrival rate. In other words, the state is hidden.
Figure 1.7 plots the observed number of requests per second together with
the estimated hidden states (i.e., arrival rates). There are 41 state transi-
tions occurred during the period of 3600 s. Some states last for a long
period of time with the maximum duration of D 5 405 s (Yu et al., 2002).
From 2000 to 2009, the HSMM has been obtained more and more
attentions from vast application areas. In this decade, the main
applications are human activity recognition (see, e.g., Hongeng and
Nevatia, 2003) and speech synthesis (see, e.g., Moore and Savic, 2004).
Other application areas include change-point/end-point detection
for semiconductor manufacturing (Ge and Smyth, 2000a), protein
structure prediction (Schmidler et al., 2000), analysis of branching and
flowering patterns in plants (Guedon et al., 2001), rain events time
series model (Sansom and Thomson, 2001), brain functional MRI
sequence analysis (Faisan et al., 2002), Internet traffic modelling (Yu
et al., 2002), event recognition in videos (Hongeng and Nevatia, 2003),
image segmentation (Lanchantin and Pieczynski, 2004), semantic
learning for a mobile robot (Squire, 2004), anomaly detection for
network security (Yu, 2005), symbolic plan recognition (Duong et al.,
2005a), terrain modeling (Wellington et al., 2005), adaptive cumulative
sum test for change detection in noninvasive mean blood pressure
trend (Yang et al., 2006), equipment prognosis (Bechhoefer et al.,
2006), financial time series modeling (Bulla and Bulla, 2006), remote
sensing (Pieczynski, 2007), classification of music (Liu et al., 2008),
and prediction of particulate matter in the air (Dong et al., 2009).
Πj;d P½S½1:d 5 j
or
Πj;d P½S1:d 5 j
the forward variables in the HSMM algorithms simpler. Then the set
of model parameters for the HSMM is defined by
λ faði;hÞðj;dÞ ; bj;d ðvk1 :kd Þ; πj;d : i; jAS; h; dAD; vkd AVg;
or
λ faði;hÞð j;dÞ ; bj;d ðvk1 :kd Þ; Πj;d : i; jAS; h; dAD; vkd AVg;
where vk1 :kd represents an observable substring of length d for
vk1 . . .vkd AVd 5 V 3 . . . 3 V. This general HSMM is shown in Figure 1.6.
P
N P
n
Suppose S1:T 5ði1 ; d1 Þ?ðiN ;dN Þ, satisfying dn 5T. Let tn 5 dm .
n51 m51
N
Then P½S1:T ; o1:T jλ 5 L aðin21 ;dn21 Þðin ;dn Þ bin ;dn ðotn2111:tn Þ, where aði0 ;d0 Þði1 ;d1 Þ 5
n51
Πi1 ;d1 for simplicity. To sum over all possible S1:T , for all N $ 1,
i1 ; . . .; iN AS and d1 ; . . .; dN AD, the computational amount involved will
be huge. Therefore, a sum-product form algorithm, that is, a
forwardbackward algorithm is usually used in the literature. Now we
define the forward and backward variables.
and
P½ot11:T jS½t2d11:t 5 j; S½t11:t1h 5 i; λ 5 P½ot11:T jS½t11:t1h 5 i; λ;
General Hidden Semi-Markov Model 33
(2.6)
for t . 0, dAD, jAS, and
X
β t ð j; dÞ 5 P½S½t11:t1h 5 i; ot11:T jS½t2d11:t 5 j; λ
i6¼j;h
X
5 aðj;dÞði;hÞ UP½ot11:T jS½t11:t1h 5 i; λ
i6¼j;h
X (2.7)
5 aðj;dÞði;hÞ Ubi;h ðot11:t1h ÞUP½ot1h11:T jS½t11:t1h 5 i; λ
i6¼j;h
X
5 aðj;dÞði;hÞ Ubi;h ðot11:t1h ÞUβ t1h ði; hÞ;
i6¼j;h
or
X
βvt ð j; dÞ 5 bj;d ðot:t1d21 Þ að j;dÞði;hÞ βvt1d ði; hÞ:
i6¼j;h
We can see that the backward recursion is exactly the same as the
forward formula (2.6) when it is expressed in the reverse time order.
This can potentially reduce the requirement for the silicon area on a
chip if the backward logic module uses the forward one. A symmetric
forwardbackward algorithm for the residual time model was intro-
duced by Yu and Kobayashi (2003a).
then from the backward recursion formula (2.7) we can see that
β t1h ði; hÞ 5 1, for t 1 h $ T. Therefore, the initial conditions for the
backward recursion formula (2.7) are as follows:
If the model assumes that the first state begins at t 5 1 and the last
state ends at or after observation oT , it is a right-censored HSMM
introduced by Guedon (2003). Because this is desirable for many
applications, it is taken as a basis for an R package for analyzing
HSMMs (Bulla et al., 2010).
2. The simplified assumption of boundary conditions assumes that the
first state begins at time 1 and the last state ends at time T. This is
the most popular assumption one can find in the literature. In this
case, the initial conditions for the forward recursion formula (2.6) are
α0 ð j; dÞ 5 πj;d ; dAD;
ατ ð j; dÞ 5 0; τ , 0; dAD;
and the initial conditions for the backward recursion formula (2.7)
are
β T ði; dÞ 5 1; dAD;
β τ ði; dÞ 5 0; τ . T; dAD:
}
}
The Backward Algorithm
1. For j 5 1; . . .; M and d 5 1; . . .; D, let β T ðj; dÞ 5 1;
2. If the simplified assumption that the last state must end at t 5 T is
assumed, let β τ ðj; dÞ 5 0 for τ . T; otherwise, let β τ ðj; dÞ 5 1 for τ . T.
3. For t 5 T 2 1; . . .; 1 {
for j 5 1; :::; M and d 5 1; . . .; D {
X
β t ðj; dÞ 5 aðj;dÞði;hÞ bi;h ðot11:t1d Þβ t1d ði; hÞ;
i;h
}
}
2.2.3 Probabilities
After the forward variables fαt ðj; dÞg and the backward variables
fβ t ðj; dÞg are determined, all other probabilities of interest can be com-
puted. For instances, the filtered probability that state j starts at
t 2 d 1 1 and ends at t, with duration d, given partial observed
sequence o1:t , can be determined by
αt ðj; dÞ
P S½t2d11:t 5 jjo1:t ; λ 5 ;
P½o1:t jλ
General Hidden Semi-Markov Model 37
and the predicted probability that state j will start at t 1 1 and end at
t 1 d, with duration d, given partial observed sequence o1:t by
X
αt ði; hÞaði;hÞðj;dÞ
i6¼j;h
P S½t11:t1d 5 jjo1:t ; λ 5 ;
P½o1:t jλ
where
X XX X
P½o1:t jλ 5 P½St 5 j; o1:t jλ 5 P½S½t2d11:t1k 5 j; o1:t jλ
j j d 0 # k # D2d
XX X X
5 P½S½t2d11:t1k 5 j; o1:t1k jλ
j d 0 # k # D 2 d ot11:t1k
XX X X X
5 αt2d ði; hÞ aði;hÞðj;d1kÞ bj;d1k ðot2d11:t1k Þ:
j d i6¼j;h 0 # k # D2d ot11:t1k
τ$t$τ2d 11
(2.13)
X X
D
5 ητ ðj; dÞ;
τ$t d5τ2t11
“You will remember our call on Cardinal Martinelli and the dinner
we had with good old Cardinal Satolli who took such a pride in the
wine produced by his own vineyards, a wine, by the way, which was
not unreservedly approved by the owners of other vineyards. One of
the most delightful experiences of all was our dinner with the good
Episcopal Rector, Dr. Nevin, when ox-tongue done in the Russian
style was served as the piece de resistánce. You cannot forget how
shocked were some of the circles in Rome to find Bishop O’Gorman
and myself at such a festal board under such circumstances, and how
Pope Leo showed his thorough understanding of American
institutions by saying that American Catholics might very properly
do things which would be very much misunderstood if done by
Romans. The Episcopal Rector was a mighty hunter, a great traveller,
and gifted with a fund of anecdote which made him a most delightful
host.”
I found this highly social and sociable party rather impatiently
awaiting a reply to their formal, written proposals to the Vatican
which had been turned over to a Commission of Cardinals. They
were giving a fine imitation of outward leisurely poise, but among
themselves they were expressing very definite opinions of the
seemingly deliberate delays to which they were being subjected. Mr.
Taft was anxious to sail for Manila on the 10th of July, and already
had his passage booked on the Koenig Albert, but the immediate
prospect seemed to be that he would be held in Rome for the rest of
the summer.
He did not have the greatest confidence that he would succeed in
the mission which meant so much to his future course in the Islands,
and, indeed, it was quite evident that he would not succeed without
prolonged effort to be continued after he left Rome. The various
Cardinals lost no opportunity to assure him that the Vatican was in
full sympathy with the proposals made and that he might expect a
very early and satisfactory termination of the business, but he
decided not to believe anything until he should see the signatures to
the contract. The factions and the politics of the Vatican were most
perplexing. The monastic orders were the conservative element in
the negotiations, being willing enough to sell the Friars’ lands at a
valuation to be decided upon by a board of five members, two
representing the church, two representing the United States
government and the fifth to be selected from some other country, but
they were not willing to consent to the withdrawal of the Friars from
the Philippine Islands. Then there were wheels within wheels; Papal
candidates and candidates for Cardinals who thrust into the
negotiations considerations for agreeing or not agreeing which
greatly puzzled the purely business-like representatives of the
American government.
But I was not particularly annoyed by the delay. I found much to
interest me in Rome, and I saw my husband improving in general
health and gaining the strength he needed for a re-encounter with
the difficulties in tropic Manila. Prominent Republican leaders had
aroused his impatience at different times by publicly announcing
that, in all probability, he was “going out to the Philippines to die.”
He wrote to his brother from Rome:
“I dislike being put in such an absurd position before the country
as that of playing the martyr. I’m not asking any favours on account
of health or any other cause, nor am I taking the position that I am
making any sacrifice. I think that a great and unusual opportunity
has been offered me and if I can improve it, all well and good, but I
don’t want any sympathy or emotional support.”
He was easily aroused to resentment on the subject, but, just the
same, it was gratifying to observe him quite rapidly regaining his
normal vigour and buoyancy.
My mother-in-law was having a most wonderful time. She was
comfortably established at the Quirinal in rooms next to ours, and
was enjoying the devoted attention of every man in the party whether
he wore ecclesiastical frock, military uniform or plain citizens’
clothes. She went everywhere and saw everything and was as
indefatigable in her enjoyment as any of us. She met old-time friends
whom she had known when she and Judge Taft were in the
diplomatic corps abroad, and with them she indulged in pleasant
reminiscence. After I arrived she became more energetic than ever
and led me a lively pace at sightseeing and shopping, because, as she
wrote to another daughter-in-law, Mrs. Horace Taft, “Nellie is not at
all timid and as she speaks French we can go anywhere.”
I soon found that in spite of official and personal protest to the
contrary we were considered quite important personages, and the
elaborate hospitality we were offered kept us busy at nearly all hours
when hospitality is at all in order. There were teas and luncheons,
dinners and receptions, and functions of every description, and we
met a great many renowned and interesting people, both Roman and
foreign. Mr. W. T. Stead, the correspondent for the London Times
who was lost on the Titanic, was one of them. Then there was Mr.
Laffan, proprietor of the New York Sun, and Mrs. Laffan, and Dr.
Hillis of Brooklyn who was in Rome with his son. An attractive
personality, who interested us very much and whose hospitality we
enjoyed, was Princess Rospigliosi, the wife of an Italian nobleman,
who lived in an enchanting house. She had a very beautiful daughter
who was at that time keenly interested in the controversy as to
whether or not Catholics should vote in Rome. She was strongly in
favour of their doing so and, with extraordinary directness, carried
her advocacy straight to the Pope and insisted that it was a great
mistake for Catholics not to take advantage of the ballot and by that
means secure the political rights to which they were entitled. Pope
Leo, although very much impressed by what she said, insisted that it
was not yet time to urge the reform suggested, and wound up by
saying, “My good daughter, you go altogether too fast for me!” I don’t
doubt that by this time the young Princess is a warm supporter of
woman’s suffrage.
Also, we were entertained by a Mr. McNutt who had been in our
diplomatic corps at one time in Madrid and Constantinople, at
another time had been tutor to the sons of the Khedive of Egypt, and
was then one of the Papal Chamberlains. He had married a woman of
wealth, a Miss Ogden of New York.
Mr. McNutt had one of the most elaborate and beautiful palaces I
ever saw. He had studied the customs of Roman society in the
picturesque days of the Medicis and the Borgias, had rented the
Pamphili Palace and restored it to its pristine glory, and it was here
that he entertained us at a dinner, with cards afterward.
I felt like an actor in a mediæval pageant whose costume had not
been delivered in time for the performance. Cardinals in their
gorgeous robes, with gold snuff-boxes, gave to the scene a high
colour among the soberer tones of Bishops and Archbishops and
uniformed Ambassadors. Then there were Princes and Princesses
and other nobilities of Roman society, the men displaying gay
ribands and decorations, the women in elaborate costumes, and all
in a “stage setting” as far removed from modernity as a magnificent
old-world palace could be. To make this reproduction of old customs
complete our host made a point of having liveried attendants with
flaming torches to light the Cardinals to and from their carriages.
Before I reached Rome, Mr. Taft and his associates had been
present at a Papal consistory at which the Pope presided over the
College of Cardinals. They were the guests of the Pope and occupied
the Diplomatic Box. I was sorry to miss this exceptional privilege, but
we were given ample opportunities for seeing and hearing several
noteworthy religious festivals both at St. Peter’s and the church of St.
John of Lateran. I was educated in the strictest Presbyterianism,
while my husband’s mother was a Unitarian, and Puritan in her
training and in all her instincts. We could not help feeling that we
had been led into a prominent position in a strange environment.
But, unshaken though we were in our religious affiliations, we
appreciated the real beauty of the ceremonies and knew that we
should rejoice in the unusual privilege accorded us which would
never be ours again.
It was near the end of our stay in Rome that we had our audience
with the Pope,—Mrs. Taft, Robert, Helen and I. I wore a black
afternoon gown with a black veil on my head, while Mrs. Taft wore
her widow’s veil as usual. Helen, I dressed in white and, to her very
great excitement, she wore a white lace veil. Bishop O’Gorman
accompanied us and when we reached the door of the Vatican under
the colonnade at the right of St. Peter’s, we were met by some
members of the Swiss Guard in their curious uniforms, conducted
through endless corridors and rich apartments until we came to a
small waiting-room where we were left for a few moments by
ourselves. We had only time to adjust our veils and compose
ourselves when the door on one side opened and we were
ceremoniously ushered into the presence of Leo XIII who sat on a
low chair under a simple canopy at the far end of the room. He rose
to greet us as we entered, and as we were presented one by one he
extended his hand over which we each bowed as we received his
blessing.
He began speaking to me in French and finding that I could
answer him in that language he talked with me for perhaps half an
hour with a most charmingly graceful manner of comment and
compliment. He spoke of Mr. Roosevelt’s present and wished that he
knew English so that he might read the books. He referred to Mr.
Roosevelt as “President Roomvine” which was as near as he seemed
to be able to get to that very un-Latin name; said that he himself, in
his youth, had been devoted to the chase and would like very much to
read “The Strenuous Life.”
Later he called Robert to his side and gave him a special blessing,
saying that he hoped the little boy would follow in the footsteps of
McKinley and Roosevelt. He asked Bob what he expected to be when
he grew up and my self-confident son replied that he intended to be
Chief Justice of the Supreme Court. I suppose he had heard the Chief
Justiceship talked about by his father until he thought it the only
worthy ambition for a self-respecting citizen to entertain.
When we arose to go, His Holiness escorted us to the door and
bowed us out with a kindly smile in his fine young eyes that I shall
never forget.
Shortly after this I left Rome. It was getting hot and my husband
persuaded me to take the children away, promising to join us for a
short breath of mountain air before he sailed for Manila. It had been
decided that I should remain in Europe for a month or so and I was
to choose the place best suited for recuperation. I went first to
Florence for a week, then to the Grande Albergo Castello de
Aquabella at Vallombrosa. The sonorous name of this hotel should
have been a sufficient warning to me of the expense of living there,
but I was not in a mood to anticipate any kind of unpleasant
experience.
It is a beautiful place reached by a funicular railway from a station
about fifteen miles from Florence, and is where Milton wrote parts of
Paradise Lost. The hotel was an old castle remodelled, and as we
were almost the only guests and were attended by relays of most
obsequious servants we managed to feel quite baronial. We spent our
time being as lazy as we liked, or driving in the dense black forests of
pine which cover the mountains and through vistas of which we
could catch fascinating glimpses of the beautiful, town-dotted valley
of the Arno some thousands of feet below.
On the 20th of July my husband came up and joined us in this
delightful retreat. He had just received his final answer from the
Vatican and, while he was disappointed at not being able to settle the
matter then, he was hopeful that a way had been found which,
though it would entail much future labour, would lead to a
satisfactory solution of the problems. An Apostolic Delegate,
representing the Vatican, was to be sent to Manila to continue the
negotiations on the ground, and Pope Leo assured Mr. Taft that he
would receive instructions to bring about such an adjustment as the
United States desired. This assurance was carried out, but only after
Leo’s long pontificate had come to an end.
The final note was written by Cardinal Rampolla who rendered
“homage to the great courtesy and high capacity” with which Mr. Taft
had filled “the delicate mission,” and closed by declaring his
willingness to concede that “the favourable result” must in a large
measure be attributed to my husband’s “high personal qualities.”
I had hoped to have Mr. Taft with us at Vallombrosa for a week or
so before he sailed, but the time allotted in our plans for this was
taken up by delays in Rome, so that when he did arrive he had only
twenty-four hours to stay. His final audience with the Pope was
arranged for the following Monday, there were a number of minor
details to be attended to, and he was to sail Thursday morning from
Naples on the Princess Irene, to which he had been obliged to
transfer from the Koenig Albert.
The last audience with His Holiness consisted chiefly in an
exchange of compliments and expressions of thanks for courtesies
extended, but it had additional interest in that the Pope chose to
make it the occasion for personally presenting to the members of the
party certain small gifts, or souvenirs, which he had selected for
them. He had previously sent an inquiry through Bishop O’Gorman
as to whether or not the Commissioners would accept decorations,
but Mr. Taft replied that the American constitution forbids the
acceptance of such honours without the consent of Congress, so
nothing more was said about it.
The presents he did receive were a handsome Jubilee medal
displaying a portrait of His Holiness in bas relief, and a gold pen in
the form of a large feather with the papal arms on it. To me the Pope
sent a small piece of old German enamel showing a copy of an
ancient picture of St. Ursula and her virgins, framed in silver and
gold beautifully wrought. Smaller gold medals were given to each of
the other Commissioners, while President Roosevelt received a copy
in mosaic of a picture of a view of Rome from a corner in the Vatican
gardens in which the Pope is seen seated with three or four Cardinals
in attendance. This, together with letters from His Holiness and
Cardinal Rampolla to the President and Mr. Hay, the Secretary of
State, was given to Bishop O’Gorman to be delivered when he arrived
in the United States.
My husband sailed from Naples on the 24th of July, and I, with the
three children and their French governess, started north by Venice
and Vienna to spend a few weeks in the mountains of Switzerland
before returning to Manila.
There were rather terrifying reports of a cholera epidemic raging
in the Philippines and I dreaded the prospects of going into it with
my children, but I knew that heroic efforts were being made to check
it and I felt confident that, in Manila at least, it would have run its
course before I should arrive, so I booked passage on the German
steamer Hamburg and on the 3rd of September sailed for the East
and the tropics once more.
When Mr. Taft reached Manila he found the city en fete and in a
state of intense excitement which had prevailed for two days during
which the people had expected every hour to hear the great siren on
the cold storage plant announce that the little Alava, the government
coastguard boat which had been sent to Singapore to get him, had
been sighted off Corregidor.
When the announcement finally came, everything in the harbour
that could manage to do so steamed down the Bay to meet him, and
when the launch to which he had transferred from the Alava came
up to the mouth of the Pásig River and under the walls of old Fort
Santiago, seventeen guns boomed out a Governor’s salute, while
whistles and bells and sirens all over the bay and river and city filled
the air with a deafening din.
Wherever his eyes rested he saw people,—crowding windows,
roofs, river banks and city walls, all of them cheering wildly and
waving hats or handkerchiefs. And the thing which moved him most
was the fact that the welcoming throng was not just representative of
the wealthy and educated class, but included thousands of the
people, barefooted and in calicoes, who had come in from the
neighbouring and even the far provinces to greet him.
Mrs. Moses asked Mr. Benito Legarda, one of the Filipino
members of the Commission, whether or not there had ever been a
like demonstration in honour of the arrival of a Spanish Governor,
and his answer was:
“Yes, there were demonstrations always, but the government paid
the expenses.”
In this case the very opposite was true. The government had no
money to waste on celebrations and all government buildings, such
as the City Hall, the Post Office and the Ayuntamiento, were
conspicuously bare. Their nakedness was positively eloquent of
economy in the midst of the riot of gay bunting, the flags, the
pennants and the palm leaves in which the rest of the city was
smothered. Then there were extraordinary and elaborate arches
spanning the streets through which the Governor was to be
conducted. One of these, erected by the Partido Federal, displayed a
huge allegorical picture which had a peculiar significance. Filipina, a
lovely lady draped in flowing gauze, was seen, in an attitude which
combined appeal with condescension, presenting to Columbia a
single star, implying that she desired to be accepted as one of the
States of the Union.
I am indebted to the descriptive art of Mrs. Moses, to photographs
and to my own knowledge of the Filipino way of doing things for the
mental picture I have of this celebration.
At the landing near the Custom House my husband found a great
procession in line, ready to escort him to the Ayuntamiento where
the speeches of welcome were to be made. There were regiments of
cavalry, infantry and artillery, as well as platoon after platoon of
native and American police with as many bands as there were
divisions of the procession. Picked men from the volunteer regiments
acted as a special guard for the Governor’s carriage and they must
have added much to the impressive array, because I know of my own
observation that the volunteers were always as fine a looking body of
men as it would be possible to find anywhere.
When Mr. Taft reached the Ayuntamiento he listened to glowing
speeches of tribute and welcome in the Marble Hall, then he stood
for hours shaking hands with the people who, in orderly file, passed
in and out of the building which was large enough to hold only a very
small fraction of them. When this was over and his audience had
settled down he proceeded to tell them in a clear and simple way all
about his experiences in Rome and how far the negotiations with the
Vatican had proceeded. This was a matter of paramount importance
to the Filipinos and they listened with an intensity of interest which
Mr. Taft said seemed to promise serious consequences if the business
could not be carried to a successful conclusion.
However, despite the joy and festivity with which he was greeted
upon his return, the Governor did not find general conditions in the
islands either prosperous or happy.
Everything that could possibly happen to a country had happened
or was happening. The cholera epidemic was still raging, and while it
had abated to a considerable extent in Manila it was at its worst in
Iloilo and other provinces. There had been from seventy to eighty
cases a day in Manila for a long time, and the quarantine regulations
had incensed the ignorant people to a point where force had to be
used to secure obedience. They did not understand sanitary
measures and wanted none of them; they clung to their superstitious
beliefs, and were easily made to accept as truth wild statements to
the effect that the Americans were poisoning the wells and rivers and
had stopped transportation and business with the sole purpose of
starving or otherwise destroying the entire population. Even the
educated ones were not without their time-honoured prejudices in
this regard, for while Mr. Taft was in Rome he receive a cabled
protest from Filipino members of the Commission with a request
that he order the quarantine raised.
When he arrived in Manila the cholera cases had fallen to between
ten and twenty a day and business had been resumed to a certain
extent, but the situation was still critical and a fresh outbreak on
account of polluted water was to be expected at any time. All the
sources of water supply were patrolled by American soldiers day and
night and every precaution was taken; whole sections of the city were
burned in an attempt to stamp out the pestilence, but the disease had
to run its course and it was months before it was completely
eradicated.
While the people were dying of cholera the carabaos, the only
draught and farm animals in the Islands, were dying by thousands of
an epidemic of rinderpest. This scourge, too, was fought with all the
force of both the civil and military arms of the government, but
before it could be checked it had carried off a large majority of the
carabaos in the Archipelago with the result that agriculture and all
other industries dependent upon this mode of transportation were
paralysed. A general drought in China made a rice famine a practical
certainty, even if the people should have money to buy rice, so the
future looked black indeed.
The cholera and rinderpest had greatly reduced government
revenues and many plans for much needed public works had to be
modified or abandoned, while the condition of the currency added to
the general chaos. There was no gold standard and the fluctuations
in the value of silver made it necessary for the Governor to issue a
proclamation about once a week fixing a new rate of exchange. In
this way it was calculated that the government, with insufficient
income at the best, lost a round million dollars gold during a period
of ten months.
To cap all and add exasperation to uneasiness the ladrones had
become increasingly active with hard times and were harrying the
districts around Manila to such an extent that the people were in
constant terror. The ravages of the rinderpest had made the carabao
a very valuable animal and the chief object of the ladrones was to
steal such as were left and drive them off to be sold in distant
provinces. Nor were they at all particular about their highwaymen’s
methods or chary of sacrificing human life. There was a veritable
hotbed of ladronism at Caloocan, a suburb of Manila, which was
augmented by the roughs and toughs from the crowded and
miserable districts in the lower city, while across the Bay in Cavite
province, known as the “mother of insurrection,” there were several
hundred rifles in the hands of marauders who hid away in the hills
and jungles and made conditions such that Mr. Taft was asked by the
Director of Constabulary to suspend the writ of habeas corpus, thus
declaring the province in a practical state of siege. Mr. Taft would not
do this, saying that he thought the only course was to “hammer away
with the constabulary until the abuse was stamped out by the regular
methods of supposedly peaceful times,” but the worst feature of the
situation was that wherever ladronism showed its head there would
be cohorts of “irreconcilables”—posing in everyday life as loyal
citizens—ready, within the limits of personal safety, to encourage and
assist it. Anything to hamper and harass the government.
Shortly after Mr. Taft’s arrival in Manila, the vice-Governor,
General Wright, and Mrs. Wright left the Islands for a well-earned
vacation and my husband wrote that the amount of work which
confronted him was staggering. He took on General Wright’s
department in addition to his own duties, and if it hadn’t been that
he had at least half way learned not to try to “hurry the East” he
probably would not have lasted long.
Among the first steps to be taken was to provide against the
inevitable famine, and to do this it was necessary for the Government
to send to China and Saigon for large quantities of rice to be stored in
public godowns. They bought and brought to Manila something like
forty million pounds of this first of all necessities to an oriental
people, and the intention was to sell it at cost when the market
supply began to run low and prices began to soar beyond the poor
man’s reach. A certain degree of paternalism has always been, is
now, and probably always will be necessary in the government of the
Filipino people.
Mr. Taft besought the United States Congress to appropriate a sum
to be used for the importation of work animals, for the purchase of
rice and the furnishing of work on public improvements. The
animals were not to be given away, but were eventually to be sold at
reasonable prices. Three millions were appropriated and spent.
Congress was also petitioned to establish a gold standard of
currency, and this too was done, to the inexpressible relief of
everybody interested in the Philippine welfare, in the following
January. The currency now is as sound as our own, every silver peso,
which corresponds to the old “dollar Mex,” being worth fifty cents
gold.
When I arrived in Manila in early October I found the situation
more interesting than it had ever been, even though it was
distracting to the men who had to deal with it. My first necessity was,
of course, to settle myself once more at Malacañan. During my
absence the old Palace had been all done over, painted and patched
and cleaned and redecorated until it was quite unlike its quaint, old
dilapidated self. Some of the colours were a shade too pronounced
and some of the decorations ran a little more to “graceful patterns”
than suited my taste, but I was glad of the added comfort and
cleanliness.
It was difficult in the beginning to accustom myself to cholera
conditions. The disease was communicated to very few Americans or
other white foreigners, but safety was secured at the price of eternal
vigilance. Water could not be drunk unless it was boiled under one’s
personal supervision; nothing uncooked could be eaten, not even a
piece of imported fruit, unless it had first been washed in a carbolic
solution, a process, I may say, which added nothing desirable to its
flavour; a good many other precautions were necessary which made
us feel as if we were living always in the lowering shadow of some
dreadful catastrophe, but, even so, we were surprisingly calm about
it—everybody was—and managed to come through the experience
without any visible ill-effects.
There was one new thing for me, and that was a live cow. For two
long years we had manfully striven to make ourselves believe that we
liked canned milk and condensed cream just as much as we liked the
fresh milk we had been used to all our lives. In fact, we were fond of
declaring that we couldn’t tell the difference. But we could. And in
our secret hearts we all welcomed as the most delectable treat an
occasional gift of skimmed milk from a friend who had been a
pioneer in the momentous venture of importing an Australian cow.
The importation of our cow was a real event, and she straightway
took up a position of great dignity and importance in our
establishment. She roamed at will about the grounds of the Palace
and her general conduct was the subject of daily comment in the
family circle. A number of people brought in cows about this time,
but very few of them lived long enough to prove their dairy worth.
Our cow flourished and gave forth large quantities of milk, and this
fact became the subject of what was supposed to be a huge joke.
Mr. Worcester, who was the high chief health authority in the
Islands, decreed that all animals as they were brought in should be
inoculated for rinderpest, tuberculosis, and a number of other things,
—“including prickly heat,” said General Wright,—but it just so
happened that a great majority of these scientifically treated beasts
died almost immediately, and General Wright could always arouse
the wrath of Mr. Worcester—a thing he loved to do—by suggesting
that the only reason our cow lived was because “she had not been
inoculated.”
The presence of the cow having given me a true farmer spirit—at
least, I suppose it was the cow—I decided to have a garden. There
were very few vegetables that the Filipinos knew how to raise at that
time, and our longing for fresh things was constant and intense. I
selected a promising looking spot behind the Palace, had it prepared
for planting, then I bought a supply of fresh American seeds and
carefully buried them in places where I thought they might develop
into something. The result was positively astonishing. The soil was
rich and the sun was hot, and in an incredibly short time we were
having quantities of beans and cauliflower and big red tomatoes and
all kinds of things.
My ambition grew with success and I branched out into poultry.
The first thing anybody knew I had a big screened yard full of
chickens and turkeys little and big, which were a source of great
enjoyment to us all both in their noisy feathered state in the chicken
yard and done up in a variety of Ah Sing styles on our very well
supplied table. I wonder how my cook made up the “squeeze” out of
which he was cheated by my industry and thrift.
But, dwelling on these minor details I am getting far ahead of my
story. There were many things in the meanwhile engaging my
attention, the most important of which, I suppose, was the great
church schism.
Gregorio Aglipay, an Ilocano priest of the Roman Catholic Church,
joined the original insurrection against Spain, or the Friars rather, at
its inception and was excommunicated. He became an insurgent
leader with a reputation for great cruelty, and continued in the field
against Spain, and subsequently against the United States, until
resistance was no longer possible. He was among the last insurrecto
chiefs to surrender in northern Luzon. When peace was restored he
began immediately to solicit the interest and aid of other Filipino
priests, of politicians and influential men in a plan for organising an
Independent Filipino Catholic Church, and his temporary success
must have surprised even him.
ARCH ERECTED BY THE PARTIDO
FEDERAL REPRESENTING FILIPINA
OFFERING ANOTHER STAR TO THE
AMERICAN FLAG